├── README.md
├── feat_drift
    ├── README.md
    ├── data
    │   ├── feature_imp1.csv
    │   └── feature_imp2.csv
    ├── feat_drift_test.ipynb
    ├── feature_drift_output
    │   └── feature_drift_201802_201803.html
    ├── images
    │   ├── feat_drift.PNG
    │   └── feat_drift.gif
    └── src
    │   ├── feature_drift_draw.py
    │   └── feature_drift_template.html
├── sankey
    ├── README.md
    ├── data
    │   └── titanic_train.csv
    ├── images
    │   ├── sankey_flow_col.PNG
    │   ├── sankey_flow_col_val.PNG
    │   ├── sankey_flow_same.PNG
    │   ├── sankey_flow_tab20.PNG
    │   └── sankey_flow_val.PNG
    ├── sankey_flow_output
    │   └── sankey_flow_Titanic.html
    ├── sankey_flow_test.ipynb
    └── src
    │   ├── generate_sankey_flow.py
    │   └── sankey_flow_template.html
└── tree
    ├── README.md
    ├── data
        └── titanic_train.csv
    ├── generate_tree_test.ipynb
    ├── image
        ├── sankey_tree.gif
        └── simple_tree.gif
    ├── sankey_tree_output
        ├── sankey_tree_Iris_Tree.html
        └── sankey_tree_Titanic_Tree.html
    ├── simple_tree_output
        ├── simple_tree_Iris_Tree.html
        └── simple_tree_Titanic_Tree.html
    └── src
        ├── generate_tree.py
        ├── sankey_tree_template.html
        └── simple_tree_template.html


/README.md:
--------------------------------------------------------------------------------
 1 | # Nuance
 2 | I use Nuance to curate varied visualization thoughts during my data scientist career.   
 3 | It is not yet a package but a list of small ideas. Welcome to test them out!
 4 | 
 5 | ## Why Nuance?
 6 | **nuance n.**  
 7 | a subtle difference in meaning or opinion or attitude 
 8 | 
 9 | ## How to use?
10 | Please check instructions in the corresponding folder
11 |     
12 | ## List of ideas
13 | 1. **simple tree**: [visualize a sklearn Decision Tree](https://github.com/SauceCat/Nuance/blob/master/tree)
14 |     <img src="https://github.com/SauceCat/Nuance/blob/master/tree/image/simple_tree.gif" />      
15 |     <br>
16 | 2. **sankey tree**: [visualize a sklearn Decision Tree](https://github.com/SauceCat/Nuance/blob/master/tree)
17 |     <img src="https://github.com/SauceCat/Nuance/blob/master/tree/image/sankey_tree.gif" />  
18 |     <br>
19 | 3. **sankey flow**: [visualize a sankey flow](https://github.com/SauceCat/Nuance/tree/master/sankey)
20 |     <table>
21 |     <tr>
22 |         <td><img src="https://github.com/SauceCat/Nuance/blob/master/sankey/images/sankey_flow_tab20.PNG" /></td>
23 |         <td><img src="https://github.com/SauceCat/Nuance/blob/master/sankey/images/sankey_flow_same.PNG" /></td>
24 |     </tr>
25 |     <tr>
26 |         <td><img src="https://github.com/SauceCat/Nuance/blob/master/sankey/images/sankey_flow_val.PNG" /></td>
27 |         <td><img src="https://github.com/SauceCat/Nuance/blob/master/sankey/images/sankey_flow_col_val.PNG" /></td>
28 |     </tr>
29 |     </table>
30 |     <br>
31 | 4. **feature drift**: [visualize feature drift](https://github.com/SauceCat/Nuance/tree/master/feat_drift)  
32 |     <br>
33 |     <img src="https://github.com/SauceCat/Nuance/blob/master/feat_drift/images/feat_drift.gif"/>
34 |     
35 |     
36 | 


--------------------------------------------------------------------------------
/feat_drift/README.md:
--------------------------------------------------------------------------------
 1 | ## Visualize feature drift
 2 | <img src="https://github.com/SauceCat/Nuance/blob/master/feat_drift/images/feat_drift.gif"/>       
 3 | 
 4 | ## What's feature drift
 5 | **"Feature drifts occur whenever the relevance of a feature grows or shrinks for incoming instances."**  
 6 | Check this paper: [A survey on feature drift adaptation: Definition, benchmark, challenges and future directions](https://www.sciencedirect.com/science/article/pii/S0164121216301030)  
 7 |   
 8 | **Make it simple:** If your training dataset is relevant to time, the subset of important features selected by the same model might be quite different through time.  
 9 | The idea is to try to visualize the "feature drift" between two different training sets. Usually these two datasets are from different snapshots. So this visualization could help detect "feature drift" through time. The expected inputs are two dataframes, containing feature importance information. You can check [this notebook](https://github.com/SauceCat/Nuance/blob/master/feat_drift/feat_drift_test.ipynb) for more details.  
10 | 
11 | ## How to use?
12 | 1. Download the folder [**feat_drift**](https://github.com/SauceCat/Nuance/tree/master/feat_drift)
13 | 2. The folder structure:
14 |     ```
15 |     feat_drift
16 |       - src: folder for all codes regarding the visualization
17 |       - data: folder for the test data
18 |       - feature_drift_output: folder for outputs (will be re-generated if it was deleted)
19 |       - feat_drift_test.ipynb: instructions and examples in jupyter notebook
20 |       - ...
21 |     ```
22 | 3. Install `jinja2`
23 |     ```
24 |     pip install jinja2
25 |     ```
26 | 4. Use feature drift visualization: (visualization depends on D3.js, so you need to connect to the network)
27 |     ```python
28 |     import sys
29 |     sys.path.insert(0, 'src/')
30 |     import feature_drift_draw
31 |     
32 |     feature_drift_draw.feature_drift_graph(feat_imp1=feat_imp1, feat_imp2=feat_imp2, feature_name='feat_name', imp_name='imp',
33 |                                            ds_name1='training set', ds_name2='test set', graph_name='train_test',
34 |                                            top_n=20, max_bar_width=300, bar_height=30, middle_gap=300, fontsize=12, color_dict=None)
35 |     ```
36 | 5. A html file would be generated in [feature_drift_output](https://github.com/SauceCat/Nuance/tree/master/feat_drift/feature_drift_output). Open it using any browser you like (I like Chrome anyway). 
37 | 
38 | ## Parameters
39 | ```python
40 | def feature_drift_graph(feat_imp1, feat_imp2, feature_name, imp_name, ds_name1, ds_name2, graph_name=None,
41 |                         top_n=None, max_bar_width=300, bar_height=30, middle_gap=300, fontsize=12, color_dict=None):
42 |     """
43 |     Draw feature drift graph
44 | 
45 |     :param feat_imp1: feature importance dataframe #1
46 |     :param feat_imp2: feature importance dataframe #2
47 |     :param feature_name: column name of features
48 |     :param imp_name: column name of importance value
49 |     :param ds_name1: name of dataset #1
50 |     :param ds_name2: name of dataset #2
51 |     :param top_n: show top_n features
52 |     :param max_bar_width: maximum bar width
53 |     :param bar_height: bar height
54 |     :param middle_gap: gap between bars
55 |     :param fontsize: font size
56 |     :param color_dict: color dictionary
57 |     """
58 | ```
59 | 


--------------------------------------------------------------------------------
/feat_drift/data/feature_imp1.csv:
--------------------------------------------------------------------------------
 1 | feat_name,imp
 2 | feat_0,13.3436942857
 3 | feat_1,15.800973571400002
 4 | feat_2,21.0843816667
 5 | feat_3,48.5161874074
 6 | feat_4,113.18283999999998
 7 | feat_5,19.0311508852
 8 | feat_6,31.0144289552
 9 | feat_7,44.0083055556
10 | feat_8,21.79879273
11 | feat_9,18.5955840707
12 | feat_10,40.70143425
13 | feat_11,44.3106395238
14 | feat_12,58.8463938636
15 | feat_13,26.846384736799997
16 | feat_14,17.89738
17 | feat_15,20.8532046377
18 | feat_16,109.927572857
19 | feat_17,45.3217483929
20 | feat_18,30.324107931
21 | feat_19,19.0750985714
22 | feat_20,15.9053894444
23 | feat_21,43.1775085
24 | feat_22,16.1404971724
25 | feat_23,30.933264415700002
26 | feat_24,25.120187
27 | feat_25,19.1116620577
28 | feat_26,23.112241081100002
29 | feat_27,175.000243176
30 | feat_28,39.8828694737
31 | feat_29,54.21781872729999
32 | feat_30,23.4141987356
33 | feat_31,30.3297
34 | feat_32,39.556396
35 | feat_33,18.1330640678
36 | feat_34,44.2011352174
37 | feat_35,20.492
38 | feat_36,14.839914088099999
39 | feat_37,24.425098869
40 | feat_38,23.087398800000006
41 | feat_39,15.864219090899999
42 | feat_40,36.5840980508
43 | feat_41,52.3627042029
44 | feat_42,17.157752380999998
45 | feat_43,159.114082
46 | feat_44,30.9611028058
47 | feat_45,46.2027222222
48 | feat_46,37.8220857851
49 | feat_47,45.8025334091
50 | feat_48,50.6727130435
51 | feat_49,55.133470512799995
52 | feat_50,33.0315573684
53 | feat_51,23.7468116667
54 | feat_52,11.451292186
55 | feat_53,31.55620125
56 | feat_54,16.6248694444
57 | feat_55,29.06870825
58 | feat_56,24.9980913953
59 | feat_57,15.538474
60 | feat_58,16.471437055
61 | feat_59,20.3773357143
62 | feat_60,23.528037551
63 | feat_61,65.3004364286
64 | feat_62,45.0990146667
65 | feat_63,17.89653
66 | feat_64,16.2706778333
67 | feat_65,18.4892417005
68 | feat_66,24.78189
69 | feat_67,162.19046175399998
70 | feat_68,34.09889375
71 | feat_69,24.2538968056
72 | feat_70,430.755799091
73 | feat_71,36.1628342105
74 | feat_72,96.30485301739999
75 | feat_73,26.820805714299997
76 | feat_74,13.4593211864
77 | feat_75,65.3470578947
78 | feat_76,26.874275
79 | feat_77,60.66391
80 | feat_78,18.2826344444
81 | feat_79,139.530246267
82 | 


--------------------------------------------------------------------------------
/feat_drift/data/feature_imp2.csv:
--------------------------------------------------------------------------------
 1 | feat_name,imp
 2 | feat_0,10.8216213571
 3 | feat_1,23.4163501481
 4 | feat_2,11.565690294100001
 5 | feat_3,48.9171478125
 6 | feat_4,113.584958947
 7 | feat_5,11.0962521419
 8 | feat_6,17.8663032911
 9 | feat_7,26.624295454499997
10 | feat_8,8.02320777095
11 | feat_9,11.2232443967
12 | feat_10,24.5941423598
13 | feat_11,28.699749
14 | feat_12,25.775422679000002
15 | feat_13,16.1959695
16 | feat_14,11.397097619
17 | feat_15,10.6688073224
18 | feat_16,146.73557250000005
19 | feat_17,23.4428398347
20 | feat_18,11.718083461500001
21 | feat_19,13.567881818699998
22 | feat_20,8.789489411760002
23 | feat_21,30.6499793103
24 | feat_22,10.708858439
25 | feat_23,16.2798533926
26 | feat_24,8.307419928060003
27 | feat_25,11.1156423881
28 | feat_26,10.9661604717
29 | feat_27,28.9956640463
30 | feat_28,11.607423439200002
31 | feat_29,35.8433884211
32 | feat_30,11.9170976667
33 | feat_31,7.815233333330001
34 | feat_32,22.894614
35 | feat_33,12.3991074603
36 | feat_34,16.485753913
37 | feat_35,11.297780625
38 | feat_36,8.3932906449
39 | feat_37,15.230949728299999
40 | feat_38,8.202025200000001
41 | feat_39,10.4006885714
42 | feat_40,12.918274803800001
43 | feat_41,41.7556285075
44 | feat_42,8.27308114035
45 | feat_43,90.61743875
46 | feat_44,16.160808051900002
47 | feat_45,39.1513645833
48 | feat_46,28.195249158200003
49 | feat_47,11.925391071400002
50 | feat_48,39.5888431379
51 | feat_49,25.952234090900003
52 | feat_50,11.6053776503
53 | feat_51,10.2122409615
54 | feat_52,8.86319782
55 | feat_53,26.429423928600002
56 | feat_54,8.92956916667
57 | feat_55,13.319904821400002
58 | feat_56,22.599339901
59 | feat_57,19.431493224
60 | feat_58,9.70069634855
61 | feat_59,19.0049746667
62 | feat_60,35.4501718966
63 | feat_61,72.6334658974
64 | feat_62,38.310038
65 | feat_63,7.72638458065
66 | feat_64,9.84875153025
67 | feat_65,15.948526475
68 | feat_66,18.7450463636
69 | feat_67,151.57445307700002
70 | feat_68,10.638199758999999
71 | feat_69,21.6299969863
72 | feat_70,180.07161175
73 | feat_71,39.1086955
74 | feat_72,211.432106861
75 | feat_73,26.0100392308
76 | feat_74,9.14663557073
77 | feat_75,19.9266190476
78 | feat_76,10.540188666699999
79 | feat_77,15.1881593548
80 | feat_78,12.2875825
81 | feat_79,128.742952879
82 | 


--------------------------------------------------------------------------------
/feat_drift/feat_drift_test.ipynb:
--------------------------------------------------------------------------------
  1 | {
  2 |  "cells": [
  3 |   {
  4 |    "cell_type": "code",
  5 |    "execution_count": 1,
  6 |    "metadata": {},
  7 |    "outputs": [],
  8 |    "source": [
  9 |     "import pandas as pd"
 10 |    ]
 11 |   },
 12 |   {
 13 |    "cell_type": "markdown",
 14 |    "metadata": {},
 15 |    "source": [
 16 |     "## read fake feature importance \n",
 17 |     "The expected inputs are two feature importance dataframes to compare.   \n",
 18 |     "It is assumed that the feature set between these two dataframes is exactly same."
 19 |    ]
 20 |   },
 21 |   {
 22 |    "cell_type": "code",
 23 |    "execution_count": 2,
 24 |    "metadata": {},
 25 |    "outputs": [],
 26 |    "source": [
 27 |     "feat_imp1 = pd.read_csv('data/feature_imp1.csv')\n",
 28 |     "feat_imp2 = pd.read_csv('data/feature_imp2.csv')"
 29 |    ]
 30 |   },
 31 |   {
 32 |    "cell_type": "code",
 33 |    "execution_count": 3,
 34 |    "metadata": {},
 35 |    "outputs": [
 36 |     {
 37 |      "data": {
 38 |       "text/html": [
 39 |        "<div>\n",
 40 |        "<style scoped>\n",
 41 |        "    .dataframe tbody tr th:only-of-type {\n",
 42 |        "        vertical-align: middle;\n",
 43 |        "    }\n",
 44 |        "\n",
 45 |        "    .dataframe tbody tr th {\n",
 46 |        "        vertical-align: top;\n",
 47 |        "    }\n",
 48 |        "\n",
 49 |        "    .dataframe thead th {\n",
 50 |        "        text-align: right;\n",
 51 |        "    }\n",
 52 |        "</style>\n",
 53 |        "<table border=\"1\" class=\"dataframe\">\n",
 54 |        "  <thead>\n",
 55 |        "    <tr style=\"text-align: right;\">\n",
 56 |        "      <th></th>\n",
 57 |        "      <th>feat_name</th>\n",
 58 |        "      <th>imp</th>\n",
 59 |        "    </tr>\n",
 60 |        "  </thead>\n",
 61 |        "  <tbody>\n",
 62 |        "    <tr>\n",
 63 |        "      <th>0</th>\n",
 64 |        "      <td>feat_0</td>\n",
 65 |        "      <td>13.343694</td>\n",
 66 |        "    </tr>\n",
 67 |        "    <tr>\n",
 68 |        "      <th>1</th>\n",
 69 |        "      <td>feat_1</td>\n",
 70 |        "      <td>15.800974</td>\n",
 71 |        "    </tr>\n",
 72 |        "    <tr>\n",
 73 |        "      <th>2</th>\n",
 74 |        "      <td>feat_2</td>\n",
 75 |        "      <td>21.084382</td>\n",
 76 |        "    </tr>\n",
 77 |        "    <tr>\n",
 78 |        "      <th>3</th>\n",
 79 |        "      <td>feat_3</td>\n",
 80 |        "      <td>48.516187</td>\n",
 81 |        "    </tr>\n",
 82 |        "    <tr>\n",
 83 |        "      <th>4</th>\n",
 84 |        "      <td>feat_4</td>\n",
 85 |        "      <td>113.182840</td>\n",
 86 |        "    </tr>\n",
 87 |        "  </tbody>\n",
 88 |        "</table>\n",
 89 |        "</div>"
 90 |       ],
 91 |       "text/plain": [
 92 |        "  feat_name         imp\n",
 93 |        "0    feat_0   13.343694\n",
 94 |        "1    feat_1   15.800974\n",
 95 |        "2    feat_2   21.084382\n",
 96 |        "3    feat_3   48.516187\n",
 97 |        "4    feat_4  113.182840"
 98 |       ]
 99 |      },
100 |      "execution_count": 3,
101 |      "metadata": {},
102 |      "output_type": "execute_result"
103 |     }
104 |    ],
105 |    "source": [
106 |     "feat_imp1.head()"
107 |    ]
108 |   },
109 |   {
110 |    "cell_type": "code",
111 |    "execution_count": 4,
112 |    "metadata": {},
113 |    "outputs": [
114 |     {
115 |      "data": {
116 |       "text/html": [
117 |        "<div>\n",
118 |        "<style scoped>\n",
119 |        "    .dataframe tbody tr th:only-of-type {\n",
120 |        "        vertical-align: middle;\n",
121 |        "    }\n",
122 |        "\n",
123 |        "    .dataframe tbody tr th {\n",
124 |        "        vertical-align: top;\n",
125 |        "    }\n",
126 |        "\n",
127 |        "    .dataframe thead th {\n",
128 |        "        text-align: right;\n",
129 |        "    }\n",
130 |        "</style>\n",
131 |        "<table border=\"1\" class=\"dataframe\">\n",
132 |        "  <thead>\n",
133 |        "    <tr style=\"text-align: right;\">\n",
134 |        "      <th></th>\n",
135 |        "      <th>feat_name</th>\n",
136 |        "      <th>imp</th>\n",
137 |        "    </tr>\n",
138 |        "  </thead>\n",
139 |        "  <tbody>\n",
140 |        "    <tr>\n",
141 |        "      <th>0</th>\n",
142 |        "      <td>feat_0</td>\n",
143 |        "      <td>10.821621</td>\n",
144 |        "    </tr>\n",
145 |        "    <tr>\n",
146 |        "      <th>1</th>\n",
147 |        "      <td>feat_1</td>\n",
148 |        "      <td>23.416350</td>\n",
149 |        "    </tr>\n",
150 |        "    <tr>\n",
151 |        "      <th>2</th>\n",
152 |        "      <td>feat_2</td>\n",
153 |        "      <td>11.565690</td>\n",
154 |        "    </tr>\n",
155 |        "    <tr>\n",
156 |        "      <th>3</th>\n",
157 |        "      <td>feat_3</td>\n",
158 |        "      <td>48.917148</td>\n",
159 |        "    </tr>\n",
160 |        "    <tr>\n",
161 |        "      <th>4</th>\n",
162 |        "      <td>feat_4</td>\n",
163 |        "      <td>113.584959</td>\n",
164 |        "    </tr>\n",
165 |        "  </tbody>\n",
166 |        "</table>\n",
167 |        "</div>"
168 |       ],
169 |       "text/plain": [
170 |        "  feat_name         imp\n",
171 |        "0    feat_0   10.821621\n",
172 |        "1    feat_1   23.416350\n",
173 |        "2    feat_2   11.565690\n",
174 |        "3    feat_3   48.917148\n",
175 |        "4    feat_4  113.584959"
176 |       ]
177 |      },
178 |      "execution_count": 4,
179 |      "metadata": {},
180 |      "output_type": "execute_result"
181 |     }
182 |    ],
183 |    "source": [
184 |     "feat_imp2.head()"
185 |    ]
186 |   },
187 |   {
188 |    "cell_type": "markdown",
189 |    "metadata": {},
190 |    "source": [
191 |     "## test feature drift graph\n",
192 |     "Here we assume that these two different feature importance results are from model trained on datasets from different snapshot months.  \n",
193 |     "- feat_imp1: feature importance from dataset 201802\n",
194 |     "- feat_imp2: feature importance from dataset 201803 "
195 |    ]
196 |   },
197 |   {
198 |    "cell_type": "code",
199 |    "execution_count": 5,
200 |    "metadata": {},
201 |    "outputs": [],
202 |    "source": [
203 |     "import sys\n",
204 |     "sys.path.insert(0, 'src/')\n",
205 |     "import feature_drift_draw"
206 |    ]
207 |   },
208 |   {
209 |    "cell_type": "code",
210 |    "execution_count": 6,
211 |    "metadata": {},
212 |    "outputs": [],
213 |    "source": [
214 |     "feature_drift_draw.feature_drift_graph(feat_imp1=feat_imp1, feat_imp2=feat_imp2, feature_name='feat_name', imp_name='imp',\n",
215 |     "                                       ds_name1='201802', ds_name2='201803', graph_name='201802_201803',\n",
216 |     "                                       top_n=20, max_bar_width=300, bar_height=30, middle_gap=300, fontsize=12, color_dict=None)"
217 |    ]
218 |   }
219 |  ],
220 |  "metadata": {
221 |   "kernelspec": {
222 |    "display_name": "Python 2",
223 |    "language": "python",
224 |    "name": "python2"
225 |   },
226 |   "language_info": {
227 |    "codemirror_mode": {
228 |     "name": "ipython",
229 |     "version": 2
230 |    },
231 |    "file_extension": ".py",
232 |    "mimetype": "text/x-python",
233 |    "name": "python",
234 |    "nbconvert_exporter": "python",
235 |    "pygments_lexer": "ipython2",
236 |    "version": "2.7.14"
237 |   }
238 |  },
239 |  "nbformat": 4,
240 |  "nbformat_minor": 2
241 | }
242 | 


--------------------------------------------------------------------------------
/feat_drift/feature_drift_output/feature_drift_201802_201803.html:
--------------------------------------------------------------------------------
  1 | <!DOCTYPE html>
  2 | <html>
  3 | 
  4 | <head>
  5 |     <meta charset='utf-8' />
  6 |     <script src="https://d3js.org/d3.v3.min.js" charset="utf-8"></script>
  7 |     <style>
  8 |         body {
  9 |             font-family: "Arial", sans-serif;
 10 |             margin: 60px 50px 50px 100px;
 11 |         }
 12 |     </style>
 13 | </head>
 14 | 
 15 | <body>
 16 |     <div>
 17 |         <h1>Feature Drift</h1>
 18 |         <p>Visualize how important features drift through two datasets.</p>
 19 |     </div>
 20 |     <div id="legend"></div> 
 21 |     <div id="graphic"></div>
 22 | 
 23 |     <script>
 24 |         // variables
 25 |         var bar_left_data = [{'bar_mark': '1', 'bar_color': '#f17182', 'feat_name': 'feat_70', 'pos': 'left', 'relative_imp': 1.0}, {'bar_mark': '1', 'bar_color': '#f17182', 'feat_name': 'feat_27', 'pos': 'left', 'relative_imp': 0.406}, {'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_67', 'pos': 'left', 'relative_imp': 0.377}, {'bar_mark': '1', 'bar_color': '#f17182', 'feat_name': 'feat_43', 'pos': 'left', 'relative_imp': 0.369}, {'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_79', 'pos': 'left', 'relative_imp': 0.324}, {'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_4', 'pos': 'left', 'relative_imp': 0.263}, {'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_16', 'pos': 'left', 'relative_imp': 0.255}, {'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_72', 'pos': 'left', 'relative_imp': 0.224}, {'bar_mark': '0', 'bar_color': '#bababa', 'feat_name': 'feat_75', 'pos': 'left', 'relative_imp': 0.152}, {'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_61', 'pos': 'left', 'relative_imp': 0.152}, {'bar_mark': '0', 'bar_color': '#bababa', 'feat_name': 'feat_77', 'pos': 'left', 'relative_imp': 0.141}, {'bar_mark': '0', 'bar_color': '#bababa', 'feat_name': 'feat_12', 'pos': 'left', 'relative_imp': 0.137}, {'bar_mark': '0', 'bar_color': '#bababa', 'feat_name': 'feat_49', 'pos': 'left', 'relative_imp': 0.128}, {'bar_mark': '1', 'bar_color': '#f17182', 'feat_name': 'feat_29', 'pos': 'left', 'relative_imp': 0.126}, {'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_41', 'pos': 'left', 'relative_imp': 0.122}, {'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_48', 'pos': 'left', 'relative_imp': 0.118}, {'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_3', 'pos': 'left', 'relative_imp': 0.113}, {'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_45', 'pos': 'left', 'relative_imp': 0.107}, {'bar_mark': '0', 'bar_color': '#bababa', 'feat_name': 'feat_47', 'pos': 'left', 'relative_imp': 0.106}, {'bar_mark': '0', 'bar_color': '#bababa', 'feat_name': 'feat_17', 'pos': 'left', 'relative_imp': 0.105}],
 26 |             bar_right_data = [{'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_72', 'pos': 'right', 'relative_imp': 1.0}, {'bar_mark': '1', 'bar_color': '#f17182', 'feat_name': 'feat_70', 'pos': 'right', 'relative_imp': 0.852}, {'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_67', 'pos': 'right', 'relative_imp': 0.717}, {'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_16', 'pos': 'right', 'relative_imp': 0.694}, {'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_79', 'pos': 'right', 'relative_imp': 0.609}, {'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_4', 'pos': 'right', 'relative_imp': 0.537}, {'bar_mark': '1', 'bar_color': '#f17182', 'feat_name': 'feat_43', 'pos': 'right', 'relative_imp': 0.429}, {'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_61', 'pos': 'right', 'relative_imp': 0.344}, {'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_3', 'pos': 'right', 'relative_imp': 0.231}, {'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_41', 'pos': 'right', 'relative_imp': 0.197}, {'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_48', 'pos': 'right', 'relative_imp': 0.187}, {'bar_mark': '1', 'bar_color': '#abdda4', 'feat_name': 'feat_45', 'pos': 'right', 'relative_imp': 0.185}, {'bar_mark': '0', 'bar_color': '#9ac6df', 'feat_name': 'feat_71', 'pos': 'right', 'relative_imp': 0.185}, {'bar_mark': '0', 'bar_color': '#9ac6df', 'feat_name': 'feat_62', 'pos': 'right', 'relative_imp': 0.181}, {'bar_mark': '1', 'bar_color': '#f17182', 'feat_name': 'feat_29', 'pos': 'right', 'relative_imp': 0.17}, {'bar_mark': '0', 'bar_color': '#9ac6df', 'feat_name': 'feat_60', 'pos': 'right', 'relative_imp': 0.168}, {'bar_mark': '0', 'bar_color': '#9ac6df', 'feat_name': 'feat_21', 'pos': 'right', 'relative_imp': 0.145}, {'bar_mark': '1', 'bar_color': '#f17182', 'feat_name': 'feat_27', 'pos': 'right', 'relative_imp': 0.137}, {'bar_mark': '0', 'bar_color': '#9ac6df', 'feat_name': 'feat_11', 'pos': 'right', 'relative_imp': 0.136}, {'bar_mark': '0', 'bar_color': '#9ac6df', 'feat_name': 'feat_46', 'pos': 'right', 'relative_imp': 0.133}],
 27 |             line_data = [{'bar_color': '#f17182', 'feat_name': 'feat_70'}, {'bar_color': '#f17182', 'feat_name': 'feat_27'}, {'bar_color': '#abdda4', 'feat_name': 'feat_67'}, {'bar_color': '#f17182', 'feat_name': 'feat_43'}, {'bar_color': '#abdda4', 'feat_name': 'feat_79'}, {'bar_color': '#abdda4', 'feat_name': 'feat_4'}, {'bar_color': '#abdda4', 'feat_name': 'feat_16'}, {'bar_color': '#abdda4', 'feat_name': 'feat_72'}, {'bar_color': '#abdda4', 'feat_name': 'feat_61'}, {'bar_color': '#f17182', 'feat_name': 'feat_29'}, {'bar_color': '#abdda4', 'feat_name': 'feat_41'}, {'bar_color': '#abdda4', 'feat_name': 'feat_48'}, {'bar_color': '#abdda4', 'feat_name': 'feat_3'}, {'bar_color': '#abdda4', 'feat_name': 'feat_45'}], 
 28 |             legend_data = [{'color': '#f17182', 'name': 'Drop'}, {'color': '#abdda4', 'name': 'Up & Stable'}, {'color': '#bababa', 'name': 'Disappear'}, {'color': '#9ac6df', 'name': 'Appear'}],
 29 |             max_bar_width = 300,
 30 |             bar_height = 30,
 31 |             middle_gap = 300,
 32 |             fontsize = 12, 
 33 |             ds_name1 = '201802', 
 34 |             ds_name2 = '201803';
 35 | 
 36 |         var width = max_bar_width * 2 + middle_gap,
 37 |             height = bar_left_data.length * bar_height;
 38 | 
 39 |         // bars
 40 |         var svg = d3.select("#graphic").append("svg")
 41 |             .attr("width", width + 10)
 42 |             .attr("height", height + 10)
 43 |             .append("g");
 44 | 
 45 |         // scaler for bar width
 46 |         var x = d3.scale.linear()
 47 |             .range([0, max_bar_width])
 48 |             .domain([0, 1.0]);
 49 | 
 50 |         // scaler for bar height
 51 |         var y_left = d3.scale.ordinal()
 52 |                 .rangeRoundBands([0, height], .1)
 53 |                 .domain(bar_left_data.map(function (d) {
 54 |                     return d.feat_name;
 55 |                 }));
 56 | 
 57 |         var y_right = d3.scale.ordinal()
 58 |                 .rangeRoundBands([0, height], .1)
 59 |                 .domain(bar_right_data.map(function (d) {
 60 |                     return d.feat_name;
 61 |                 }));
 62 | 
 63 |         // adjust for titles
 64 |         var title_adjust = bar_height - y_left(bar_left_data[0].feat_name);
 65 | 
 66 |         // draw legends
 67 |         var legend_height = bar_height / 1.5,
 68 |             legend_gap = 3,
 69 |             legend_text_offset_x = 5,
 70 |             legend_text_offset_y = 4,
 71 |             legend_margin_bottom = bar_height;
 72 | 
 73 |         var legend = d3.select('#legend').append('svg')
 74 |             .attr("width", width + 10)
 75 |             .attr("height", legend_height * 2 + legend_gap + legend_margin_bottom)
 76 |             .append("g");
 77 | 
 78 |         var legend_bars = legend.selectAll('.legend_bar')
 79 |             .data(legend_data)
 80 |             .enter()
 81 |             .append("g")
 82 | 
 83 |         legend_bars.append("rect")
 84 |             .attr("class", 'legend_bar')
 85 |             .attr("y", function (d, i) {
 86 |                 if (i >= 2) {
 87 |                     return legend_height + legend_gap;
 88 |                 }else{
 89 |                     return 0;
 90 |                 }
 91 |             })
 92 |             .attr("height", legend_height)
 93 |             .attr("x", function(d, i) {
 94 |                 return (max_bar_width / 2.) * (i % 2);
 95 |             })
 96 |             .attr("width", legend_height)
 97 |             .style("fill", function(d) {
 98 |                 return d.color;
 99 |             });
100 | 
101 |         legend_bars.append("text")
102 |             .attr("class", "legend_label")
103 |             .attr("y", function (d, i) {
104 |                 if (i >= 2) {
105 |                     return 1.5 * legend_height + legend_gap + legend_text_offset_y;
106 |                 }else{
107 |                     return 0.5 * legend_height + 4;
108 |                 }
109 |             })
110 |             .attr("x", function(d, i) {
111 |                 return (max_bar_width / 2.) * (i % 2) + legend_height + legend_text_offset_x;
112 |             })
113 |             .text(function (d) {
114 |                 return d.name;
115 |             })
116 |             .style('font-size', fontsize + 'px')
117 |             .style('font-weight', 'bold');
118 | 
119 |         // draw bars and lines
120 |         updateBar('left', bar_left_data);
121 |         updateBar('right', bar_right_data);
122 |         updateLines();
123 | 
124 |         function updateBar(pos, bar_data) {
125 | 
126 |             var bar_x = 0,
127 |                 bg_bar_class = 'bg_bar_left',
128 |                 bar_class = 'bar_left',
129 |                 y = y_left, 
130 |                 other_pos = 'right', 
131 |                 title = ds_name1;
132 | 
133 |             if (pos == 'right') {
134 |                 bar_x = max_bar_width + middle_gap;
135 |                 bg_bar_class = 'bg_bar_right';
136 |                 bar_class = 'bar_right';
137 |                 y = y_right;
138 |                 other_pos = 'left', 
139 |                 title = ds_name2;
140 |             }
141 | 
142 |             // append title
143 |             svg.append("g").append('text')
144 |                 .text(title)
145 |                 .attr('y', bar_height / 1.5)
146 |                 .attr('x', bar_x + 5)
147 |                 .style('fill', '#000')
148 |                 .style('font-weight', 'bold')
149 |                 .style('font-size', fontsize + 2 + 'px');
150 |             
151 |             // draw gray bars as background
152 |             var bg_bars = svg.selectAll("." + bg_bar_class)
153 |                 .data(bar_data)
154 |                 .enter()
155 |                 .append("g")
156 | 
157 |             bg_bars.append("rect")
158 |                 .attr("class", bg_bar_class)
159 |                 .attr("y", function (d) {
160 |                     return y(d.feat_name) + title_adjust;
161 |                 })
162 |                 .attr("height", y.rangeBand())
163 |                 .attr("x", bar_x)
164 |                 .attr("width", max_bar_width)
165 |                 .style("fill", '#f0f0f0')
166 |                 .on("mouseover", barHover)
167 |                 .on("mouseout", barOut);
168 | 
169 |             // draw bars
170 |             var bars = svg.selectAll("." + bar_class)
171 |                 .data(bar_data)
172 |                 .enter()
173 |                 .append("g");
174 | 
175 |             bars.append("rect")
176 |                 .attr("class", bar_class)
177 |                 .attr("y", function (d) {
178 |                     return y(d.feat_name) + title_adjust;
179 |                 })
180 |                 .attr("height", y.rangeBand())
181 |                 .attr("x", bar_x)
182 |                 .attr("width", function (d) {
183 |                     return x(d.relative_imp);
184 |                 })
185 |                 .style('fill', function (d){
186 |                     return d.bar_color;
187 |                 })                
188 |                 .on("mouseover", barHover)
189 |                 .on("mouseout", barOut);
190 | 
191 |             // append feature names
192 |             bars.append("text")
193 |                 .attr("class", "label")
194 |                 .attr("y", function (d) {
195 |                     return y(d.feat_name) + y.rangeBand() / 1.5 + title_adjust;
196 |                 })
197 |                 .attr("x", bar_x+5)
198 |                 .text(function (d) {
199 |                     return d.feat_name;
200 |                 })
201 |                 .style('fill', function(d) { return '#000';})
202 |                 .style('font-size', fontsize + 'px');
203 |         }
204 | 
205 | 
206 |         function updateLines() {
207 | 
208 |             var lines = svg.selectAll(".lines")
209 |                 .data(line_data)
210 |                 .enter()
211 |                 .append("g");
212 | 
213 |             lines.append("line")
214 |                 .attr("class", "lines")
215 |                 .attr('x1', max_bar_width)
216 |                 .attr('y1', function (d) {
217 |                     return y_left(d.feat_name) + y_left.rangeBand() / 2 + title_adjust;
218 |                 })
219 |                 .attr('x2', max_bar_width + middle_gap)
220 |                 .attr('y2', function (d) {
221 |                     return y_right(d.feat_name) + y_right.rangeBand() / 2 + title_adjust;
222 |                 })
223 |                 .style('stroke', function (d) {
224 |                     return d.bar_color;
225 |                 });
226 | 
227 |         }
228 | 
229 |         // mouseover handler
230 |         function barHover(d) {
231 |             d3.selectAll('.bg_bar_' + d.pos).filter(function(o) {
232 |                 return o.feat_name == d.feat_name;
233 |             })
234 |             .style("stroke", d.bar_color)
235 |             .style("stroke-width", 2);
236 | 
237 |             var other_pos = 'right';
238 |             if (d.pos == 'right') other_pos = 'left';
239 | 
240 |             if (d.bar_mark == '1') {
241 |                 d3.selectAll('.bg_bar_' + other_pos).filter(function(o) {
242 |                     return o.feat_name == d.feat_name;
243 |                 })
244 |                 .style("stroke", d.bar_color)
245 |                 .style("stroke-width", 2);
246 | 
247 |                 d3.selectAll('.lines').filter(function(l) {
248 |                     return l.feat_name == d.feat_name;
249 |                 })
250 |                 .style("stroke-width", 4);
251 |             }
252 |         }
253 | 
254 |         // mouseout handler
255 |         function barOut(d) {
256 | 
257 |             d3.selectAll('.bg_bar_' + d.pos).filter(function(o) {
258 |                 return o.feat_name == d.feat_name;
259 |             })
260 |             .style("stroke", null);
261 | 
262 |             var other_pos = 'right';
263 |             if (d.pos == 'right') other_pos = 'left';
264 | 
265 |             if (d.bar_mark == '1') {
266 |                 d3.selectAll('.bg_bar_' + other_pos).filter(function(o) {
267 |                     return o.feat_name == d.feat_name;
268 |                 })
269 |                 .style("stroke", null);
270 | 
271 |                 d3.selectAll('.lines').filter(function(l) {
272 |                     return l.feat_name == d.feat_name;
273 |                 })
274 |                 .style("stroke-width", 1);
275 |             }
276 |         }
277 | 
278 |     </script>
279 | </body>
280 | </html>


--------------------------------------------------------------------------------
/feat_drift/images/feat_drift.PNG:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/sosuneko/Nuance/e8b486ae7459850a00f1e8bbd756e7a57aed4417/feat_drift/images/feat_drift.PNG


--------------------------------------------------------------------------------
/feat_drift/images/feat_drift.gif:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/sosuneko/Nuance/e8b486ae7459850a00f1e8bbd756e7a57aed4417/feat_drift/images/feat_drift.gif


--------------------------------------------------------------------------------
/feat_drift/src/feature_drift_draw.py:
--------------------------------------------------------------------------------
  1 | 
  2 | import pandas as pd
  3 | import jinja2
  4 | import os
  5 | 
  6 | 
  7 | def _process_imp(imp_df, imp_name):
  8 |     """
  9 |     Preprocessing on the input feature importance dataframe
 10 | 
 11 |     :param imp_df: feature importance pandas dataframe
 12 |     :param imp_name: column name of the importance value
 13 |     :return:
 14 |         dataframe with relative_imp and feat_rank
 15 |     """
 16 | 
 17 |     imp_df = imp_df.sort_values(by=imp_name, ascending=False).reset_index(drop=True)
 18 |     imp_df['relative_imp'] = imp_df[imp_name] * 1.0 / imp_df[imp_name].max()
 19 |     imp_df['relative_imp'] = imp_df['relative_imp'].apply(lambda x : round(x, 3))
 20 |     imp_df['feat_rank'] = imp_df.index.values + 1
 21 |     return imp_df
 22 | 
 23 | 
 24 | def _rank2color(x, color_dict):
 25 |     """
 26 |     Map change of rank to color
 27 | 
 28 |     :param x: row of dataframe
 29 |     :param color_dict: color dictionary
 30 |     """
 31 | 
 32 |     if x['feat_rank_x'] < x['feat_rank_y']:
 33 |         return color_dict['drop']
 34 |     if x['feat_rank_x'] >= x['feat_rank_y']:
 35 |         return color_dict['up_or_stable']
 36 |     if pd.isnull(x['feat_rank_y']):
 37 |         return color_dict['disappear']
 38 |     if pd.isnull(x['feat_rank_x']):
 39 |         return color_dict['appear']
 40 |     
 41 | 
 42 | def _get_mark(x):
 43 |     """
 44 |     '1' for feature appears on both feature importance dataframes
 45 |     '0' for feature disappears on either dataframe
 46 |     """
 47 |     if pd.isnull(x['feat_rank_y']) or pd.isnull(x['feat_rank_x']):
 48 |         return "0"
 49 |     else:
 50 |         return "1"
 51 |     
 52 | 
 53 | def _merge_feat_imp(imp_df1, imp_df2, feature_name, top_n, color_dict):
 54 |     """
 55 |     Merge and compare two feature importance dataframes
 56 | 
 57 |     :param imp_df1: feature importance dataframe #1
 58 |     :param imp_df2: feature importance dataframe #2
 59 |     :param feature_name: column name of features
 60 |     :param top_n: show top_n features
 61 |     :param color_dict: color dictionary
 62 |     :return:
 63 |         The merged dataframe
 64 |     """
 65 | 
 66 |     imp_df1['pos'] = 'left'
 67 |     imp_df2['pos'] = 'right'
 68 |     if top_n:
 69 |         both_imp = imp_df1.head(top_n).merge(imp_df2.head(top_n), on=feature_name, how='outer')
 70 |     else:
 71 |         both_imp = imp_df1.merge(imp_df2, on=feature_name, how='outer')
 72 |         
 73 |     both_imp['bar_color'] = both_imp.apply(lambda x : _rank2color(x, color_dict), axis=1)
 74 |     both_imp['bar_mark'] = both_imp.apply(lambda x : _get_mark(x), axis=1)
 75 |     
 76 |     return both_imp
 77 | 
 78 | 
 79 | def feature_drift_graph(feat_imp1, feat_imp2, feature_name, imp_name, ds_name1, ds_name2, graph_name=None,
 80 |                         top_n=None, max_bar_width=300, bar_height=30, middle_gap=300, fontsize=12, color_dict=None):
 81 |     """
 82 |     Draw feature drift graph
 83 | 
 84 |     :param feat_imp1: feature importance dataframe #1
 85 |     :param feat_imp2: feature importance dataframe #2
 86 |     :param feature_name: column name of features
 87 |     :param imp_name: column name of importance value
 88 |     :param ds_name1: name of dataset #1
 89 |     :param ds_name2: name of dataset #2
 90 |     :param top_n: show top_n features
 91 |     :param max_bar_width: maximum bar width
 92 |     :param bar_height: bar height
 93 |     :param middle_gap: gap between bars
 94 |     :param fontsize: font size
 95 |     :param color_dict: color dictionary
 96 |     """
 97 | 
 98 |     feat_imp1 = _process_imp(feat_imp1, imp_name)
 99 |     feat_imp2 = _process_imp(feat_imp2, imp_name)
100 |     
101 |     if color_dict is None:
102 |         color_dict = {
103 |             'drop': '#f17182',
104 |             'up_or_stable': '#abdda4',
105 |             'disappear': '#bababa',
106 |             'appear': '#9ac6df'
107 |         }
108 |     
109 |     both_imp = _merge_feat_imp(feat_imp1, feat_imp2, feature_name, top_n, color_dict)
110 |     
111 |     bar_left_data = both_imp[['feat_name', 'relative_imp_x', 'pos_x', 'bar_color', 'bar_mark']
112 |                             ].dropna().sort_values('relative_imp_x', ascending=False)
113 |     bar_left_data.columns = [col.replace('_x', '') for col in bar_left_data.columns.values]
114 | 
115 |     bar_right_data = both_imp[['feat_name', 'relative_imp_y', 'pos_y', 'bar_color', 'bar_mark']
116 |                              ].dropna().sort_values('relative_imp_y', ascending=False)
117 |     bar_right_data.columns = [col.replace('_y', '') for col in bar_right_data.columns.values]
118 | 
119 |     line_data = both_imp[['feat_name', 'bar_color', 'feat_rank_x', 'feat_rank_y']].dropna()[['feat_name', 'bar_color']]
120 |     
121 |     legend_data = [
122 |         {'name': 'Drop', 'color': color_dict['drop']},
123 |         {'name': 'Up & Stable', 'color': color_dict['up_or_stable']},
124 |         {'name': 'Disappear', 'color': color_dict['disappear']},
125 |         {'name': 'Appear', 'color': color_dict['appear']}
126 |     ]
127 |     
128 |     # render the output
129 |     temp = open('src/feature_drift_template.html').read()
130 |     template = jinja2.Template(temp)
131 | 
132 |     # create the output root if it is not exits
133 |     if not os.path.exists('feature_drift_output'):
134 |         os.mkdir('feature_drift_output')
135 | 
136 |     # generate output html
137 |     if graph_name is None:
138 |         output_path = 'feature_drift_output/feature_drift_output.html'
139 |     else:
140 |         output_path = 'feature_drift_output/feature_drift_%s.html' %graph_name
141 | 
142 |     with open(output_path, 'wb') as fh:
143 |         fh.write(template.render({'bar_left_data': bar_left_data.to_dict('records'), 
144 |                                   'bar_right_data': bar_right_data.to_dict('records'), 
145 |                                   'line_data': line_data.to_dict('records'), 
146 |                                   'legend_data': legend_data, 
147 |                                   'max_bar_width': max_bar_width, 'bar_height': bar_height, 
148 |                                   'middle_gap': middle_gap, 'fontsize': fontsize,
149 |                                   'ds_name1': ds_name1, 'ds_name2': ds_name2}))
150 | 


--------------------------------------------------------------------------------
/feat_drift/src/feature_drift_template.html:
--------------------------------------------------------------------------------
  1 | <!DOCTYPE html>
  2 | <html>
  3 | 
  4 | <head>
  5 |     <meta charset='utf-8' />
  6 |     <script src="https://d3js.org/d3.v3.min.js" charset="utf-8"></script>
  7 |     <style>
  8 |         body {
  9 |             font-family: "Arial", sans-serif;
 10 |             margin: 60px 50px 50px 100px;
 11 |         }
 12 |     </style>
 13 | </head>
 14 | 
 15 | <body>
 16 |     <div>
 17 |         <h1>Feature Drift</h1>
 18 |         <p>Visualize how important features drift through two datasets.</p>
 19 |     </div>
 20 |     <div id="legend"></div> 
 21 |     <div id="graphic"></div>
 22 | 
 23 |     <script>
 24 |         // variables
 25 |         var bar_left_data = {{ bar_left_data|safe }},
 26 |             bar_right_data = {{ bar_right_data|safe }},
 27 |             line_data = {{ line_data|safe }}, 
 28 |             legend_data = {{ legend_data|safe }},
 29 |             max_bar_width = {{ max_bar_width }},
 30 |             bar_height = {{ bar_height }},
 31 |             middle_gap = {{ middle_gap }},
 32 |             fontsize = {{ fontsize }}, 
 33 |             ds_name1 = '{{ ds_name1 }}', 
 34 |             ds_name2 = '{{ ds_name2 }}';
 35 | 
 36 |         var width = max_bar_width * 2 + middle_gap,
 37 |             height = bar_left_data.length * bar_height;
 38 | 
 39 |         // bars
 40 |         var svg = d3.select("#graphic").append("svg")
 41 |             .attr("width", width + 10)
 42 |             .attr("height", height + 10)
 43 |             .append("g");
 44 | 
 45 |         // scaler for bar width
 46 |         var x = d3.scale.linear()
 47 |             .range([0, max_bar_width])
 48 |             .domain([0, 1.0]);
 49 | 
 50 |         // scaler for bar height
 51 |         var y_left = d3.scale.ordinal()
 52 |                 .rangeRoundBands([0, height], .1)
 53 |                 .domain(bar_left_data.map(function (d) {
 54 |                     return d.feat_name;
 55 |                 }));
 56 | 
 57 |         var y_right = d3.scale.ordinal()
 58 |                 .rangeRoundBands([0, height], .1)
 59 |                 .domain(bar_right_data.map(function (d) {
 60 |                     return d.feat_name;
 61 |                 }));
 62 | 
 63 |         // adjust for titles
 64 |         var title_adjust = bar_height - y_left(bar_left_data[0].feat_name);
 65 | 
 66 |         // draw legends
 67 |         var legend_height = bar_height / 1.5,
 68 |             legend_gap = 3,
 69 |             legend_text_offset_x = 5,
 70 |             legend_text_offset_y = 4,
 71 |             legend_margin_bottom = bar_height;
 72 | 
 73 |         var legend = d3.select('#legend').append('svg')
 74 |             .attr("width", width + 10)
 75 |             .attr("height", legend_height * 2 + legend_gap + legend_margin_bottom)
 76 |             .append("g");
 77 | 
 78 |         var legend_bars = legend.selectAll('.legend_bar')
 79 |             .data(legend_data)
 80 |             .enter()
 81 |             .append("g")
 82 | 
 83 |         legend_bars.append("rect")
 84 |             .attr("class", 'legend_bar')
 85 |             .attr("y", function (d, i) {
 86 |                 if (i >= 2) {
 87 |                     return legend_height + legend_gap;
 88 |                 }else{
 89 |                     return 0;
 90 |                 }
 91 |             })
 92 |             .attr("height", legend_height)
 93 |             .attr("x", function(d, i) {
 94 |                 return (max_bar_width / 2.) * (i % 2);
 95 |             })
 96 |             .attr("width", legend_height)
 97 |             .style("fill", function(d) {
 98 |                 return d.color;
 99 |             });
100 | 
101 |         legend_bars.append("text")
102 |             .attr("class", "legend_label")
103 |             .attr("y", function (d, i) {
104 |                 if (i >= 2) {
105 |                     return 1.5 * legend_height + legend_gap + legend_text_offset_y;
106 |                 }else{
107 |                     return 0.5 * legend_height + 4;
108 |                 }
109 |             })
110 |             .attr("x", function(d, i) {
111 |                 return (max_bar_width / 2.) * (i % 2) + legend_height + legend_text_offset_x;
112 |             })
113 |             .text(function (d) {
114 |                 return d.name;
115 |             })
116 |             .style('font-size', fontsize + 'px')
117 |             .style('font-weight', 'bold');
118 | 
119 |         // draw bars and lines
120 |         updateBar('left', bar_left_data);
121 |         updateBar('right', bar_right_data);
122 |         updateLines();
123 | 
124 |         function updateBar(pos, bar_data) {
125 | 
126 |             var bar_x = 0,
127 |                 bg_bar_class = 'bg_bar_left',
128 |                 bar_class = 'bar_left',
129 |                 y = y_left, 
130 |                 other_pos = 'right', 
131 |                 title = ds_name1;
132 | 
133 |             if (pos == 'right') {
134 |                 bar_x = max_bar_width + middle_gap;
135 |                 bg_bar_class = 'bg_bar_right';
136 |                 bar_class = 'bar_right';
137 |                 y = y_right;
138 |                 other_pos = 'left', 
139 |                 title = ds_name2;
140 |             }
141 | 
142 |             // append title
143 |             svg.append("g").append('text')
144 |                 .text(title)
145 |                 .attr('y', bar_height / 1.5)
146 |                 .attr('x', bar_x + 5)
147 |                 .style('fill', '#000')
148 |                 .style('font-weight', 'bold')
149 |                 .style('font-size', fontsize + 2 + 'px');
150 |             
151 |             // draw gray bars as background
152 |             var bg_bars = svg.selectAll("." + bg_bar_class)
153 |                 .data(bar_data)
154 |                 .enter()
155 |                 .append("g")
156 | 
157 |             bg_bars.append("rect")
158 |                 .attr("class", bg_bar_class)
159 |                 .attr("y", function (d) {
160 |                     return y(d.feat_name) + title_adjust;
161 |                 })
162 |                 .attr("height", y.rangeBand())
163 |                 .attr("x", bar_x)
164 |                 .attr("width", max_bar_width)
165 |                 .style("fill", '#f0f0f0')
166 |                 .on("mouseover", barHover)
167 |                 .on("mouseout", barOut);
168 | 
169 |             // draw bars
170 |             var bars = svg.selectAll("." + bar_class)
171 |                 .data(bar_data)
172 |                 .enter()
173 |                 .append("g");
174 | 
175 |             bars.append("rect")
176 |                 .attr("class", bar_class)
177 |                 .attr("y", function (d) {
178 |                     return y(d.feat_name) + title_adjust;
179 |                 })
180 |                 .attr("height", y.rangeBand())
181 |                 .attr("x", bar_x)
182 |                 .attr("width", function (d) {
183 |                     return x(d.relative_imp);
184 |                 })
185 |                 .style('fill', function (d){
186 |                     return d.bar_color;
187 |                 })                
188 |                 .on("mouseover", barHover)
189 |                 .on("mouseout", barOut);
190 | 
191 |             // append feature names
192 |             bars.append("text")
193 |                 .attr("class", "label")
194 |                 .attr("y", function (d) {
195 |                     return y(d.feat_name) + y.rangeBand() / 1.5 + title_adjust;
196 |                 })
197 |                 .attr("x", bar_x+5)
198 |                 .text(function (d) {
199 |                     return d.feat_name;
200 |                 })
201 |                 .style('fill', function(d) { return '#000';})
202 |                 .style('font-size', fontsize + 'px');
203 |         }
204 | 
205 | 
206 |         function updateLines() {
207 | 
208 |             var lines = svg.selectAll(".lines")
209 |                 .data(line_data)
210 |                 .enter()
211 |                 .append("g");
212 | 
213 |             lines.append("line")
214 |                 .attr("class", "lines")
215 |                 .attr('x1', max_bar_width)
216 |                 .attr('y1', function (d) {
217 |                     return y_left(d.feat_name) + y_left.rangeBand() / 2 + title_adjust;
218 |                 })
219 |                 .attr('x2', max_bar_width + middle_gap)
220 |                 .attr('y2', function (d) {
221 |                     return y_right(d.feat_name) + y_right.rangeBand() / 2 + title_adjust;
222 |                 })
223 |                 .style('stroke', function (d) {
224 |                     return d.bar_color;
225 |                 });
226 | 
227 |         }
228 | 
229 |         // mouseover handler
230 |         function barHover(d) {
231 |             d3.selectAll('.bg_bar_' + d.pos).filter(function(o) {
232 |                 return o.feat_name == d.feat_name;
233 |             })
234 |             .style("stroke", d.bar_color)
235 |             .style("stroke-width", 2);
236 | 
237 |             var other_pos = 'right';
238 |             if (d.pos == 'right') other_pos = 'left';
239 | 
240 |             if (d.bar_mark == '1') {
241 |                 d3.selectAll('.bg_bar_' + other_pos).filter(function(o) {
242 |                     return o.feat_name == d.feat_name;
243 |                 })
244 |                 .style("stroke", d.bar_color)
245 |                 .style("stroke-width", 2);
246 | 
247 |                 d3.selectAll('.lines').filter(function(l) {
248 |                     return l.feat_name == d.feat_name;
249 |                 })
250 |                 .style("stroke-width", 4);
251 |             }
252 |         }
253 | 
254 |         // mouseout handler
255 |         function barOut(d) {
256 | 
257 |             d3.selectAll('.bg_bar_' + d.pos).filter(function(o) {
258 |                 return o.feat_name == d.feat_name;
259 |             })
260 |             .style("stroke", null);
261 | 
262 |             var other_pos = 'right';
263 |             if (d.pos == 'right') other_pos = 'left';
264 | 
265 |             if (d.bar_mark == '1') {
266 |                 d3.selectAll('.bg_bar_' + other_pos).filter(function(o) {
267 |                     return o.feat_name == d.feat_name;
268 |                 })
269 |                 .style("stroke", null);
270 | 
271 |                 d3.selectAll('.lines').filter(function(l) {
272 |                     return l.feat_name == d.feat_name;
273 |                 })
274 |                 .style("stroke-width", 1);
275 |             }
276 |         }
277 | 
278 |     </script>
279 | </body>
280 | </html>


--------------------------------------------------------------------------------
/sankey/README.md:
--------------------------------------------------------------------------------
 1 | ## Visualize a sankey flow
 2 | <table>
 3 | <tr>
 4 |     <td><img src="https://github.com/SauceCat/Nuance/blob/master/sankey/images/sankey_flow_tab20.PNG" /></td>
 5 |     <td><img src="https://github.com/SauceCat/Nuance/blob/master/sankey/images/sankey_flow_same.PNG" /></td>
 6 | </tr>
 7 | <tr>
 8 |     <td><img src="https://github.com/SauceCat/Nuance/blob/master/sankey/images/sankey_flow_val.PNG" /></td>
 9 |     <td><img src="https://github.com/SauceCat/Nuance/blob/master/sankey/images/sankey_flow_col_val.PNG" /></td>
10 | </tr>
11 | </table>
12 | 
13 | ## How to use?
14 | 1. Download the folder [**sankey**](https://github.com/SauceCat/Nuance/tree/master/sankey)
15 | 2. The folder structure:
16 |     ```
17 |     sankey
18 |       - src: folder for all codes regarding the visualization
19 |       - data: folder for the test data
20 |       - sankey_flow_output: folder for outputs (will be re-generated if it was deleted)
21 |       - sankey_flow_test.ipynb: instructions and examples in jupyter notebook
22 |       - ...
23 |     ```
24 | 3. Install `jinja2`
25 |     ```
26 |     pip install jinja2
27 |     ```
28 | 4. Use sankey_flow visualization: (visualization depends on D3.js, so you need to connect to the network)
29 |     ```python
30 |     import sys
31 |     sys.path.insert(0, 'src/')
32 |     import generate_sankey_flow
33 |     
34 |     generate_sankey_flow.draw_sankey_flow(df=raw[use_cols], node_color_type='col', link_color_type='source', 
35 |                                           width=1600, height=900, graph_name='Titanic', 
36 |                                           node_color_mapping=None, color_map=None, link_color=None)
37 |     ```
38 | 5. A html file would be generated in [sankey_flow_output](https://github.com/SauceCat/Nuance/tree/master/sankey/sankey_flow_output). Open it using any browser you like (I like Chrome anyway). 
39 | 
40 | ## Parameters
41 | ```python
42 | def draw_sankey_flow(df, node_color_type, link_color_type, width, height, 
43 |                      graph_name=None, node_color_mapping=None, color_map=None, link_color=None):
44 | 	'''
45 | 	:param df:
46 | 		pandas DataFrame, each column represents a state
47 | 	:param node_color_type:
48 | 		node coloring strategy, can be one of ['col', 'val', 'col_val', 'cus']
49 | 			- 'col': each column has different color
50 | 			- 'val': each unique value has different color (unique values through all columns)
51 | 			- 'col_val': each unique value in each column has different color
52 | 			- 'cus': customer provide node color mapping
53 | 	:param link_color_type:
54 | 		link coloring strategy, default='source'
55 | 		Can be one of ['source', 'target', 'both', 'same']
56 | 			- 'source': same color as the source node
57 | 			- 'target': same color as the target node
58 | 			- 'both': color from both target and source
59 | 			- 'same': all links have same color
60 | 	:param width: wdith
61 | 	:param height: height
62 | 	:param graph_name: name of the graph
63 | 	:param node_color_mapping:
64 | 		if node_color_type == 'cus', color_mapping should be provided
65 | 		example:
66 | 		node_color_mapping = {
67 | 			'type': 'col',
68 | 			'mapping': {
69 | 				column1: color1, column2: color2, ...
70 | 			}
71 | 		}
72 | 	:param color_map: matplotlib color map
73 | 	:param link_color:
74 | 		if link_color_type == 'same', link color should be provided
75 | 	'''
76 | ```
77 | 


--------------------------------------------------------------------------------
/sankey/images/sankey_flow_col.PNG:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/sosuneko/Nuance/e8b486ae7459850a00f1e8bbd756e7a57aed4417/sankey/images/sankey_flow_col.PNG


--------------------------------------------------------------------------------
/sankey/images/sankey_flow_col_val.PNG:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/sosuneko/Nuance/e8b486ae7459850a00f1e8bbd756e7a57aed4417/sankey/images/sankey_flow_col_val.PNG


--------------------------------------------------------------------------------
/sankey/images/sankey_flow_same.PNG:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/sosuneko/Nuance/e8b486ae7459850a00f1e8bbd756e7a57aed4417/sankey/images/sankey_flow_same.PNG


--------------------------------------------------------------------------------
/sankey/images/sankey_flow_tab20.PNG:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/sosuneko/Nuance/e8b486ae7459850a00f1e8bbd756e7a57aed4417/sankey/images/sankey_flow_tab20.PNG


--------------------------------------------------------------------------------
/sankey/images/sankey_flow_val.PNG:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/sosuneko/Nuance/e8b486ae7459850a00f1e8bbd756e7a57aed4417/sankey/images/sankey_flow_val.PNG


--------------------------------------------------------------------------------
/sankey/sankey_flow_output/sankey_flow_Titanic.html:
--------------------------------------------------------------------------------
  1 | <!DOCTYPE html>
  2 | <head>
  3 | <meta charset="utf-8">
  4 | 
  5 | <style>
  6 |  body {
  7 |   font-family: helvetica;
  8 |   font-size: 10px;
  9 |  }
 10 | 
 11 | .node rect {
 12 |   shape-rendering: crispEdges;
 13 | }
 14 |  
 15 | .node text {
 16 |   pointer-events: none;
 17 |   fill: #555;
 18 |   font-weight: bold;
 19 |   font-size: 12px;
 20 | }
 21 | 
 22 | #mycontainer {
 23 |   margin: 3vh 0 0 5vw;
 24 | }
 25 |  
 26 | </style>
 27 | </head>
 28 | 
 29 | <body>
 30 | <div id='mycontainer'>
 31 |   <div id="chart"></div>
 32 | </div>
 33 | 
 34 | 
 35 | <script src="https://d3js.org/d3.v3.min.js"></script>
 36 | <script type="text/javascript">
 37 |   // include sankey from here
 38 | d3.sankey = function() {
 39 |   var sankey = {},
 40 |       nodeWidth = 24,
 41 |       nodePadding = 8,
 42 |       size = [1, 1],
 43 |       nodes = [],
 44 |       links = [];
 45 | 
 46 |   sankey.nodeWidth = function(_) {
 47 |     if (!arguments.length) return nodeWidth;
 48 |     nodeWidth = +_;
 49 |     return sankey;
 50 |   };
 51 | 
 52 |   sankey.nodePadding = function(_) {
 53 |     if (!arguments.length) return nodePadding;
 54 |     nodePadding = +_;
 55 |     return sankey;
 56 |   };
 57 | 
 58 |   sankey.nodes = function(_) {
 59 |     if (!arguments.length) return nodes;
 60 |     nodes = _;
 61 |     return sankey;
 62 |   };
 63 | 
 64 |   sankey.links = function(_) {
 65 |     if (!arguments.length) return links;
 66 |     links = _;
 67 |     return sankey;
 68 |   };
 69 | 
 70 |   sankey.size = function(_) {
 71 |     if (!arguments.length) return size;
 72 |     size = _;
 73 |     return sankey;
 74 |   };
 75 | 
 76 |   sankey.layout = function(iterations) {
 77 |     computeNodeLinks();
 78 |     computeNodeValues();
 79 |     computeNodeBreadths();
 80 |     computeNodeDepths(iterations);
 81 |     computeLinkDepths();
 82 |     return sankey;
 83 |   };
 84 | 
 85 |   sankey.relayout = function() {
 86 |     computeLinkDepths();
 87 |     return sankey;
 88 |   };
 89 | 
 90 |   sankey.link = function() {
 91 |     var curvature = .5;
 92 | 
 93 |     function link(d) {
 94 |       var x0 = d.source.x + d.source.dx,
 95 |           x1 = d.target.x,
 96 |           xi = d3.interpolateNumber(x0, x1),
 97 |           x2 = xi(curvature),
 98 |           x3 = xi(1 - curvature),
 99 |           y0 = d.source.y + d.sy + d.dy / 2,
100 |           y1 = d.target.y + d.ty + d.dy / 2;
101 |       return "M" + x0 + "," + y0
102 |            + "C" + x2 + "," + y0
103 |            + " " + x3 + "," + y1
104 |            + " " + x1 + "," + y1;
105 |     }
106 | 
107 |     link.curvature = function(_) {
108 |       if (!arguments.length) return curvature;
109 |       curvature = +_;
110 |       return link;
111 |     };
112 | 
113 |     return link;
114 |   };
115 | 
116 |   // Populate the sourceLinks and targetLinks for each node.
117 |   // Also, if the source and target are not objects, assume they are indices.
118 |   function computeNodeLinks() {
119 |     nodes.forEach(function(node) {
120 |       node.sourceLinks = [];
121 |       node.targetLinks = [];
122 |     });
123 |     links.forEach(function(link) {
124 |       var source = link.source,
125 |           target = link.target;
126 |       if (typeof source === "number") source = link.source = nodes[link.source];
127 |       if (typeof target === "number") target = link.target = nodes[link.target];
128 |       source.sourceLinks.push(link);
129 |       target.targetLinks.push(link);
130 |     });
131 |   }
132 | 
133 |   // Compute the value (size) of each node by summing the associated links.
134 |   function computeNodeValues() {
135 |     nodes.forEach(function(node) {
136 |       node.value = Math.max(
137 |         d3.sum(node.sourceLinks, value),
138 |         d3.sum(node.targetLinks, value)
139 |       );
140 |     });
141 |   }
142 | 
143 |   // Iteratively assign the breadth (x-position) for each node.
144 |   // Nodes are assigned the maximum breadth of incoming neighbors plus one;
145 |   // nodes with no incoming links are assigned breadth zero, while
146 |   // nodes with no outgoing links are assigned the maximum breadth.
147 |   function computeNodeBreadths() {
148 |     var remainingNodes = nodes,
149 |         nextNodes,
150 |         x = 0;
151 | 
152 |     while (remainingNodes.length) {
153 |       nextNodes = [];
154 |       remainingNodes.forEach(function(node) {
155 |         node.x = x;
156 |         node.dx = nodeWidth;
157 |         node.sourceLinks.forEach(function(link) {
158 |           nextNodes.push(link.target);
159 |         });
160 |       });
161 |       remainingNodes = nextNodes;
162 |       ++x;
163 |     }
164 | 
165 |     //
166 |     moveSinksRight(x);
167 |     scaleNodeBreadths((width - nodeWidth) / (x - 1));
168 |   }
169 | 
170 |   function moveSourcesRight() {
171 |     nodes.forEach(function(node) {
172 |       if (!node.targetLinks.length) {
173 |         node.x = d3.min(node.sourceLinks, function(d) { return d.target.x; }) - 1;
174 |       }
175 |     });
176 |   }
177 | 
178 |   function moveSinksRight(x) {
179 |     nodes.forEach(function(node) {
180 |       if (!node.sourceLinks.length) {
181 |         node.x = x - 1;
182 |       }
183 |     });
184 |   }
185 | 
186 |   function scaleNodeBreadths(kx) {
187 |     nodes.forEach(function(node) {
188 |       node.x *= kx;
189 |     });
190 |   }
191 | 
192 |   function computeNodeDepths(iterations) {
193 |     var nodesByBreadth = d3.nest()
194 |         .key(function(d) { return d.x; })
195 |         .sortKeys(d3.ascending)
196 |         .entries(nodes)
197 |         .map(function(d) { return d.values; });
198 | 
199 |     //
200 |     initializeNodeDepth();
201 |     resolveCollisions();
202 |     for (var alpha = 1; iterations > 0; --iterations) {
203 |       relaxRightToLeft(alpha *= .99);
204 |       resolveCollisions();
205 |       relaxLeftToRight(alpha);
206 |       resolveCollisions();
207 |     }
208 | 
209 |     function initializeNodeDepth() {
210 |       var ky = d3.min(nodesByBreadth, function(nodes) {
211 |         return (size[1] - (nodes.length - 1) * nodePadding) / d3.sum(nodes, value);
212 |       });
213 | 
214 |       nodesByBreadth.forEach(function(nodes) {
215 |         nodes.forEach(function(node, i) {
216 |           node.y = i;
217 |           node.dy = node.value * ky;
218 |         });
219 |       });
220 | 
221 |       links.forEach(function(link) {
222 |         link.dy = link.value * ky;
223 |       });
224 |     }
225 | 
226 |     function relaxLeftToRight(alpha) {
227 |       nodesByBreadth.forEach(function(nodes, breadth) {
228 |         nodes.forEach(function(node) {
229 |           if (node.targetLinks.length) {
230 |             var y = d3.sum(node.targetLinks, weightedSource) / d3.sum(node.targetLinks, value);
231 |             node.y += (y - center(node)) * alpha;
232 |           }
233 |         });
234 |       });
235 | 
236 |       function weightedSource(link) {
237 |         return center(link.source) * link.value;
238 |       }
239 |     }
240 | 
241 |     function relaxRightToLeft(alpha) {
242 |       nodesByBreadth.slice().reverse().forEach(function(nodes) {
243 |         nodes.forEach(function(node) {
244 |           if (node.sourceLinks.length) {
245 |             var y = d3.sum(node.sourceLinks, weightedTarget) / d3.sum(node.sourceLinks, value);
246 |             node.y += (y - center(node)) * alpha;
247 |           }
248 |         });
249 |       });
250 | 
251 |       function weightedTarget(link) {
252 |         return center(link.target) * link.value;
253 |       }
254 |     }
255 | 
256 |     function resolveCollisions() {
257 |       nodesByBreadth.forEach(function(nodes) {
258 |         var node,
259 |             dy,
260 |             y0 = 0,
261 |             n = nodes.length,
262 |             i;
263 | 
264 |         // Push any overlapping nodes down.
265 |         nodes.sort(ascendingDepth);
266 |         for (i = 0; i < n; ++i) {
267 |           node = nodes[i];
268 |           dy = y0 - node.y;
269 |           if (dy > 0) node.y += dy;
270 |           y0 = node.y + node.dy + nodePadding;
271 |         }
272 | 
273 |         // If the bottommost node goes outside the bounds, push it back up.
274 |         dy = y0 - nodePadding - size[1];
275 |         if (dy > 0) {
276 |           y0 = node.y -= dy;
277 | 
278 |           // Push any overlapping nodes back up.
279 |           for (i = n - 2; i >= 0; --i) {
280 |             node = nodes[i];
281 |             dy = node.y + node.dy + nodePadding - y0;
282 |             if (dy > 0) node.y -= dy;
283 |             y0 = node.y;
284 |           }
285 |         }
286 |       });
287 |     }
288 | 
289 |     function ascendingDepth(a, b) {
290 |       return a.y - b.y;
291 |     }
292 |   }
293 | 
294 |   function computeLinkDepths() {
295 |     nodes.forEach(function(node) {
296 |       node.sourceLinks.sort(ascendingTargetDepth);
297 |       node.targetLinks.sort(ascendingSourceDepth);
298 |     });
299 |     nodes.forEach(function(node) {
300 |       var sy = 0, ty = 0;
301 |       node.sourceLinks.forEach(function(link) {
302 |         link.sy = sy;
303 |         sy += link.dy;
304 |       });
305 |       node.targetLinks.forEach(function(link) {
306 |         link.ty = ty;
307 |         ty += link.dy;
308 |       });
309 |     });
310 | 
311 |     function ascendingSourceDepth(a, b) {
312 |       return a.source.y - b.source.y;
313 |     }
314 | 
315 |     function ascendingTargetDepth(a, b) {
316 |       return a.target.y - b.target.y;
317 |     }
318 |   }
319 | 
320 |   function center(node) {
321 |     return node.y + node.dy / 2;
322 |   }
323 | 
324 |   function value(link) {
325 |     return link.value;
326 |   }
327 | 
328 |   return sankey;
329 | };
330 | </script>
331 | <script>
332 | 
333 | var data = {"nodes": [{"color": "#f46d43", "name": "Pclass: 3"}, {"color": "#3288bd", "name": "Pclass: 2"}, {"color": "#d53e4f", "name": "Pclass: 1"}, {"color": "#f46d43", "name": "Embarked: unknown"}, {"color": "#fdae61", "name": "Sex: male"}, {"color": "#5e4fa2", "name": "SibSp: 5"}, {"color": "#9e0142", "name": "SibSp: 4"}, {"color": "#abdda4", "name": "SibSp: 3"}, {"color": "#e6f598", "name": "SibSp: 2"}, {"color": "#ffffbf", "name": "SibSp: 1"}, {"color": "#fee08b", "name": "SibSp: 0"}, {"color": "#d53e4f", "name": "SibSp: 8"}, {"color": "#5e4fa2", "name": "Survived: 1"}, {"color": "#9e0142", "name": "Survived: 0"}, {"color": "#9e0142", "name": "Embarked: C"}, {"color": "#fee08b", "name": "Parch: 2"}, {"color": "#ffffbf", "name": "Parch: 3"}, {"color": "#f46d43", "name": "Parch: 0"}, {"color": "#fdae61", "name": "Parch: 1"}, {"color": "#66c2a5", "name": "Parch: 6"}, {"color": "#66c2a5", "name": "Sex: female"}, {"color": "#e6f598", "name": "Parch: 4"}, {"color": "#abdda4", "name": "Parch: 5"}, {"color": "#d53e4f", "name": "Embarked: Q"}, {"color": "#3288bd", "name": "Embarked: S"}], "links": [{"source": "Survived: 0", "target": "Pclass: 1", "value": "80"}, {"source": "Survived: 0", "target": "Pclass: 2", "value": "97"}, {"source": "Survived: 0", "target": "Pclass: 3", "value": "372"}, {"source": "Survived: 1", "target": "Pclass: 1", "value": "136"}, {"source": "Survived: 1", "target": "Pclass: 2", "value": "87"}, {"source": "Survived: 1", "target": "Pclass: 3", "value": "119"}, {"source": "Pclass: 1", "target": "Sex: female", "value": "94"}, {"source": "Pclass: 1", "target": "Sex: male", "value": "122"}, {"source": "Pclass: 2", "target": "Sex: female", "value": "76"}, {"source": "Pclass: 2", "target": "Sex: male", "value": "108"}, {"source": "Pclass: 3", "target": "Sex: female", "value": "144"}, {"source": "Pclass: 3", "target": "Sex: male", "value": "347"}, {"source": "Sex: female", "target": "SibSp: 0", "value": "174"}, {"source": "Sex: female", "target": "SibSp: 1", "value": "106"}, {"source": "Sex: female", "target": "SibSp: 2", "value": "13"}, {"source": "Sex: female", "target": "SibSp: 3", "value": "11"}, {"source": "Sex: female", "target": "SibSp: 4", "value": "6"}, {"source": "Sex: female", "target": "SibSp: 5", "value": "1"}, {"source": "Sex: female", "target": "SibSp: 8", "value": "3"}, {"source": "Sex: male", "target": "SibSp: 0", "value": "434"}, {"source": "Sex: male", "target": "SibSp: 1", "value": "103"}, {"source": "Sex: male", "target": "SibSp: 2", "value": "15"}, {"source": "Sex: male", "target": "SibSp: 3", "value": "5"}, {"source": "Sex: male", "target": "SibSp: 4", "value": "12"}, {"source": "Sex: male", "target": "SibSp: 5", "value": "4"}, {"source": "Sex: male", "target": "SibSp: 8", "value": "4"}, {"source": "SibSp: 0", "target": "Parch: 0", "value": "537"}, {"source": "SibSp: 0", "target": "Parch: 1", "value": "38"}, {"source": "SibSp: 0", "target": "Parch: 2", "value": "29"}, {"source": "SibSp: 0", "target": "Parch: 3", "value": "1"}, {"source": "SibSp: 0", "target": "Parch: 4", "value": "1"}, {"source": "SibSp: 0", "target": "Parch: 5", "value": "2"}, {"source": "SibSp: 1", "target": "Parch: 0", "value": "123"}, {"source": "SibSp: 1", "target": "Parch: 1", "value": "57"}, {"source": "SibSp: 1", "target": "Parch: 2", "value": "19"}, {"source": "SibSp: 1", "target": "Parch: 3", "value": "3"}, {"source": "SibSp: 1", "target": "Parch: 4", "value": "3"}, {"source": "SibSp: 1", "target": "Parch: 5", "value": "3"}, {"source": "SibSp: 1", "target": "Parch: 6", "value": "1"}, {"source": "SibSp: 2", "target": "Parch: 0", "value": "16"}, {"source": "SibSp: 2", "target": "Parch: 1", "value": "7"}, {"source": "SibSp: 2", "target": "Parch: 2", "value": "4"}, {"source": "SibSp: 2", "target": "Parch: 3", "value": "1"}, {"source": "SibSp: 3", "target": "Parch: 0", "value": "2"}, {"source": "SibSp: 3", "target": "Parch: 1", "value": "7"}, {"source": "SibSp: 3", "target": "Parch: 2", "value": "7"}, {"source": "SibSp: 4", "target": "Parch: 1", "value": "9"}, {"source": "SibSp: 4", "target": "Parch: 2", "value": "9"}, {"source": "SibSp: 5", "target": "Parch: 2", "value": "5"}, {"source": "SibSp: 8", "target": "Parch: 2", "value": "7"}, {"source": "Parch: 0", "target": "Embarked: C", "value": "123"}, {"source": "Parch: 0", "target": "Embarked: Q", "value": "69"}, {"source": "Parch: 0", "target": "Embarked: S", "value": "484"}, {"source": "Parch: 0", "target": "Embarked: unknown", "value": "2"}, {"source": "Parch: 1", "target": "Embarked: C", "value": "30"}, {"source": "Parch: 1", "target": "Embarked: Q", "value": "6"}, {"source": "Parch: 1", "target": "Embarked: S", "value": "82"}, {"source": "Parch: 2", "target": "Embarked: C", "value": "14"}, {"source": "Parch: 2", "target": "Embarked: Q", "value": "1"}, {"source": "Parch: 2", "target": "Embarked: S", "value": "65"}, {"source": "Parch: 3", "target": "Embarked: C", "value": "1"}, {"source": "Parch: 3", "target": "Embarked: S", "value": "4"}, {"source": "Parch: 4", "target": "Embarked: S", "value": "4"}, {"source": "Parch: 5", "target": "Embarked: Q", "value": "1"}, {"source": "Parch: 5", "target": "Embarked: S", "value": "4"}, {"source": "Parch: 6", "target": "Embarked: S", "value": "1"}]};
334 | var link_color_type = 'source';
335 | var link_color = 'None';
336 | var width = 1600, height = 900;
337 | 
338 | var margin = {top: 50, right: 200, bottom: 100, left: 0},
339 |     width = width - margin.left - margin.right,
340 |     height = height - margin.top - margin.bottom;
341 | 
342 | // Set the sankey diagram properties
343 | var sankey = d3.sankey()
344 |     .nodeWidth(25)
345 |     .nodePadding(20)
346 |     .size([width, height]);
347 |  
348 | var path = sankey.link();
349 | update_sankey(data);
350 | 
351 | function update_sankey(data){
352 |   // remember deep copy here
353 |   var graph = data;
354 | 
355 |   // append the svg canvas to the page
356 |   d3.select('#chart').select('svg').remove();
357 |   var svg = d3.select("#chart").append("svg")
358 |       .attr("width", width + margin.left + margin.right)
359 |       .attr("height", height + margin.top + margin.bottom)
360 |     .append("g")
361 |       .attr("transform", 
362 |             "translate(" + margin.left + "," + margin.top + ")");
363 | 
364 |   var nodeMap = {};
365 |   graph.nodes.forEach(function(x) { nodeMap[x.name] = x; });
366 |   graph.links = graph.links.map(function(x) {
367 |     return {
368 |       source: nodeMap[x.source],
369 |       target: nodeMap[x.target],
370 |       value: x.value
371 |     };
372 |   });
373 | 
374 |   sankey
375 |       .nodes(graph.nodes)
376 |       .links(graph.links)
377 |       .layout(32);
378 |    
379 |   var defs = svg.append("defs");
380 | 
381 |   // create gradients for the links
382 |   var grads = defs.selectAll("linearGradient")
383 |     .data(graph.links, getGradID);
384 | 
385 |   grads.enter().append("linearGradient")
386 |     .attr("id", getGradID)
387 |     .attr("gradientUnits", "userSpaceOnUse");
388 | 
389 |   positionGrads(grads);
390 | 
391 |   grads.html("") //erase any existing <stop> elements on update
392 |     .append("stop")
393 |     .attr("offset", "0%")
394 |     .attr("stop-color", function(d) {
395 |       return nodeColor((+d.source.x <= +d.target.x) ?
396 |         d.source : d.target);
397 |     });
398 | 
399 |   grads.append("stop")
400 |     .attr("offset", "100%")
401 |     .attr("stop-color", function(d) {
402 |       return nodeColor((+d.source.x > +d.target.x) ?
403 |         d.source : d.target)
404 |     });
405 | 
406 |   // add in the links
407 |   var link = svg.append("g").selectAll(".link")
408 |     .data(graph.links)
409 |     .enter().append("path")
410 |     .attr("class", "link")
411 |     .attr("d", path)
412 |     .style("fill", "none")
413 |     .style("stroke", function(d) {
414 |       if(link_color_type=="source") return d.source.color;
415 |       else if(link_color_type=="target") return d.target.color;
416 |       else if(link_color_type=="both") return "url(#" + getGradID(d) + ")";
417 |       else return link_color;
418 |     })
419 |     .style("stroke-opacity", "0.3")
420 |     .on("mouseover", function() {
421 |       d3.select(this).style("stroke-opacity", "0.6")
422 |     })
423 |     .on("mouseout", function() {
424 |       d3.select(this).style("stroke-opacity", "0.3")
425 |     })
426 |     .style("stroke-width", function(d) {
427 |       return Math.max(1, d.dy);
428 |     })
429 |     .sort(function(a, b) {
430 |       return b.dy - a.dy;
431 |     });
432 | 
433 |   // add the link titles
434 |   link.append("title")
435 |         .text(function(d) {
436 |         return d.source.name + " → " + 
437 |                 d.target.name + "\n" + d.value; });
438 |  
439 | // add in the nodes
440 |   var node = svg.append("g").selectAll(".node")
441 |       .data(graph.nodes)
442 |     .enter().append("g")
443 |       .attr("class", "node")
444 |       .attr("transform", function(d) { 
445 |       return "translate(" + d.x + "," + d.y + ")"; });
446 |  
447 | // add the rectangles for the nodes
448 |   node.append("rect")
449 |       .attr("height", function(d) { return d.dy; })
450 |       .attr("width", sankey.nodeWidth())
451 |       .style("fill", function(d) { 
452 |       return d.color = d.color; })
453 |       .style('stroke-width', 0)
454 |     .append("title")
455 |       .text(function(d) { 
456 |       return d.name + "\n" + d.value; });
457 |  
458 | // add in the title for the nodes
459 |   node.append("text")
460 |       .attr("x", 6 + sankey.nodeWidth())
461 |       .attr("y", function(d) { return d.dy / 2; })
462 |       .attr("dy", ".35em")
463 |       .attr("text-anchor", "start")
464 |       .attr("transform", null)
465 |       .text(function(d) { return d.name; });
466 | }
467 | 
468 | // define utility functions
469 | function getGradID(d) {
470 |   return "linkGrad-" + d.source.name.replace(/ /g, "") + "-" + d.target.name.replace(/ /g, "");
471 | }
472 | 
473 | function nodeColor(d) {
474 |   return d.color;
475 | }
476 | 
477 | function positionGrads(grads) {
478 |   grads.attr("x1", function(d) {
479 |       return d.source.x;
480 |     })
481 |     .attr("y1", function(d) {
482 |       return d.source.y;
483 |     })
484 |     .attr("x2", function(d) {
485 |       return d.target.x;
486 |     })
487 |     .attr("y2", function(d) {
488 |       return d.target.y;
489 |     });
490 | }
491 | 
492 | </script>
493 |  
494 | </body>
495 | </html>


--------------------------------------------------------------------------------
/sankey/sankey_flow_test.ipynb:
--------------------------------------------------------------------------------
  1 | {
  2 |  "cells": [
  3 |   {
  4 |    "cell_type": "markdown",
  5 |    "metadata": {},
  6 |    "source": [
  7 |     "## test sankey_flow visualization with Titanic dataset"
  8 |    ]
  9 |   },
 10 |   {
 11 |    "cell_type": "code",
 12 |    "execution_count": 1,
 13 |    "metadata": {
 14 |     "collapsed": true
 15 |    },
 16 |    "outputs": [],
 17 |    "source": [
 18 |     "import pandas as pd\n",
 19 |     "import numpy as np"
 20 |    ]
 21 |   },
 22 |   {
 23 |    "cell_type": "markdown",
 24 |    "metadata": {},
 25 |    "source": [
 26 |     "## read data"
 27 |    ]
 28 |   },
 29 |   {
 30 |    "cell_type": "code",
 31 |    "execution_count": 2,
 32 |    "metadata": {
 33 |     "collapsed": true
 34 |    },
 35 |    "outputs": [],
 36 |    "source": [
 37 |     "raw = pd.read_csv('data/titanic_train.csv')"
 38 |    ]
 39 |   },
 40 |   {
 41 |    "cell_type": "code",
 42 |    "execution_count": 3,
 43 |    "metadata": {
 44 |     "collapsed": false
 45 |    },
 46 |    "outputs": [
 47 |     {
 48 |      "data": {
 49 |       "text/html": [
 50 |        "<div>\n",
 51 |        "<table border=\"1\" class=\"dataframe\">\n",
 52 |        "  <thead>\n",
 53 |        "    <tr style=\"text-align: right;\">\n",
 54 |        "      <th></th>\n",
 55 |        "      <th>PassengerId</th>\n",
 56 |        "      <th>Survived</th>\n",
 57 |        "      <th>Pclass</th>\n",
 58 |        "      <th>Name</th>\n",
 59 |        "      <th>Sex</th>\n",
 60 |        "      <th>Age</th>\n",
 61 |        "      <th>SibSp</th>\n",
 62 |        "      <th>Parch</th>\n",
 63 |        "      <th>Ticket</th>\n",
 64 |        "      <th>Fare</th>\n",
 65 |        "      <th>Cabin</th>\n",
 66 |        "      <th>Embarked</th>\n",
 67 |        "    </tr>\n",
 68 |        "  </thead>\n",
 69 |        "  <tbody>\n",
 70 |        "    <tr>\n",
 71 |        "      <th>0</th>\n",
 72 |        "      <td>1</td>\n",
 73 |        "      <td>0</td>\n",
 74 |        "      <td>3</td>\n",
 75 |        "      <td>Braund, Mr. Owen Harris</td>\n",
 76 |        "      <td>male</td>\n",
 77 |        "      <td>22.0</td>\n",
 78 |        "      <td>1</td>\n",
 79 |        "      <td>0</td>\n",
 80 |        "      <td>A/5 21171</td>\n",
 81 |        "      <td>7.2500</td>\n",
 82 |        "      <td>NaN</td>\n",
 83 |        "      <td>S</td>\n",
 84 |        "    </tr>\n",
 85 |        "    <tr>\n",
 86 |        "      <th>1</th>\n",
 87 |        "      <td>2</td>\n",
 88 |        "      <td>1</td>\n",
 89 |        "      <td>1</td>\n",
 90 |        "      <td>Cumings, Mrs. John Bradley (Florence Briggs Th...</td>\n",
 91 |        "      <td>female</td>\n",
 92 |        "      <td>38.0</td>\n",
 93 |        "      <td>1</td>\n",
 94 |        "      <td>0</td>\n",
 95 |        "      <td>PC 17599</td>\n",
 96 |        "      <td>71.2833</td>\n",
 97 |        "      <td>C85</td>\n",
 98 |        "      <td>C</td>\n",
 99 |        "    </tr>\n",
100 |        "    <tr>\n",
101 |        "      <th>2</th>\n",
102 |        "      <td>3</td>\n",
103 |        "      <td>1</td>\n",
104 |        "      <td>3</td>\n",
105 |        "      <td>Heikkinen, Miss. Laina</td>\n",
106 |        "      <td>female</td>\n",
107 |        "      <td>26.0</td>\n",
108 |        "      <td>0</td>\n",
109 |        "      <td>0</td>\n",
110 |        "      <td>STON/O2. 3101282</td>\n",
111 |        "      <td>7.9250</td>\n",
112 |        "      <td>NaN</td>\n",
113 |        "      <td>S</td>\n",
114 |        "    </tr>\n",
115 |        "    <tr>\n",
116 |        "      <th>3</th>\n",
117 |        "      <td>4</td>\n",
118 |        "      <td>1</td>\n",
119 |        "      <td>1</td>\n",
120 |        "      <td>Futrelle, Mrs. Jacques Heath (Lily May Peel)</td>\n",
121 |        "      <td>female</td>\n",
122 |        "      <td>35.0</td>\n",
123 |        "      <td>1</td>\n",
124 |        "      <td>0</td>\n",
125 |        "      <td>113803</td>\n",
126 |        "      <td>53.1000</td>\n",
127 |        "      <td>C123</td>\n",
128 |        "      <td>S</td>\n",
129 |        "    </tr>\n",
130 |        "    <tr>\n",
131 |        "      <th>4</th>\n",
132 |        "      <td>5</td>\n",
133 |        "      <td>0</td>\n",
134 |        "      <td>3</td>\n",
135 |        "      <td>Allen, Mr. William Henry</td>\n",
136 |        "      <td>male</td>\n",
137 |        "      <td>35.0</td>\n",
138 |        "      <td>0</td>\n",
139 |        "      <td>0</td>\n",
140 |        "      <td>373450</td>\n",
141 |        "      <td>8.0500</td>\n",
142 |        "      <td>NaN</td>\n",
143 |        "      <td>S</td>\n",
144 |        "    </tr>\n",
145 |        "  </tbody>\n",
146 |        "</table>\n",
147 |        "</div>"
148 |       ],
149 |       "text/plain": [
150 |        "   PassengerId  Survived  Pclass  \\\n",
151 |        "0            1         0       3   \n",
152 |        "1            2         1       1   \n",
153 |        "2            3         1       3   \n",
154 |        "3            4         1       1   \n",
155 |        "4            5         0       3   \n",
156 |        "\n",
157 |        "                                                Name     Sex   Age  SibSp  \\\n",
158 |        "0                            Braund, Mr. Owen Harris    male  22.0      1   \n",
159 |        "1  Cumings, Mrs. John Bradley (Florence Briggs Th...  female  38.0      1   \n",
160 |        "2                             Heikkinen, Miss. Laina  female  26.0      0   \n",
161 |        "3       Futrelle, Mrs. Jacques Heath (Lily May Peel)  female  35.0      1   \n",
162 |        "4                           Allen, Mr. William Henry    male  35.0      0   \n",
163 |        "\n",
164 |        "   Parch            Ticket     Fare Cabin Embarked  \n",
165 |        "0      0         A/5 21171   7.2500   NaN        S  \n",
166 |        "1      0          PC 17599  71.2833   C85        C  \n",
167 |        "2      0  STON/O2. 3101282   7.9250   NaN        S  \n",
168 |        "3      0            113803  53.1000  C123        S  \n",
169 |        "4      0            373450   8.0500   NaN        S  "
170 |       ]
171 |      },
172 |      "execution_count": 3,
173 |      "metadata": {},
174 |      "output_type": "execute_result"
175 |     }
176 |    ],
177 |    "source": [
178 |     "raw.head()"
179 |    ]
180 |   },
181 |   {
182 |    "cell_type": "code",
183 |    "execution_count": 4,
184 |    "metadata": {
185 |     "collapsed": false
186 |    },
187 |    "outputs": [
188 |     {
189 |      "data": {
190 |       "text/plain": [
191 |        "array(['PassengerId', 'Survived', 'Pclass', 'Name', 'Sex', 'Age', 'SibSp',\n",
192 |        "       'Parch', 'Ticket', 'Fare', 'Cabin', 'Embarked'], dtype=object)"
193 |       ]
194 |      },
195 |      "execution_count": 4,
196 |      "metadata": {},
197 |      "output_type": "execute_result"
198 |     }
199 |    ],
200 |    "source": [
201 |     "raw.columns.values"
202 |    ]
203 |   },
204 |   {
205 |    "cell_type": "code",
206 |    "execution_count": 5,
207 |    "metadata": {
208 |     "collapsed": false
209 |    },
210 |    "outputs": [
211 |     {
212 |      "name": "stdout",
213 |      "output_type": "stream",
214 |      "text": [
215 |       "Survived: [0 1]\n",
216 |       "Pclass: [3 1 2]\n",
217 |       "Sex: ['male' 'female']\n",
218 |       "SibSp: [1 0 3 4 2 5 8]\n",
219 |       "Parch: [0 1 2 5 3 4 6]\n",
220 |       "Embarked: ['S' 'C' 'Q' nan]\n"
221 |      ]
222 |     }
223 |    ],
224 |    "source": [
225 |     "# use these columns as layers of nodes\n",
226 |     "use_cols = ['Survived', 'Pclass', 'Sex', 'SibSp', 'Parch', 'Embarked']\n",
227 |     "for col in use_cols:\n",
228 |     "    print '%s: %s' %(col, str(raw[col].unique()))"
229 |    ]
230 |   },
231 |   {
232 |    "cell_type": "code",
233 |    "execution_count": 6,
234 |    "metadata": {
235 |     "collapsed": true
236 |    },
237 |    "outputs": [],
238 |    "source": [
239 |     "raw['Embarked'] = raw['Embarked'].fillna('unknown')"
240 |    ]
241 |   },
242 |   {
243 |    "cell_type": "markdown",
244 |    "metadata": {},
245 |    "source": [
246 |     "## input for generate_sankey_flow\n",
247 |     "The function is expecting a pandas DataFrame. In this DataFrame, each colum represents a state.  \n",
248 |     "Take Titanic for example, columns **['Survived', 'Pclass', 'Sex', 'SibSp', 'Parch', 'Embarked']** are 6 different states. Unique values in each column are all possible state value in that state. And each row represents a unit and its flowing path.  \n",
249 |     "Well, to draw a proper sankey path, you need to calculate how many units is flowing from (status1, value1) to (status2, value1), how many units is flowing from (status2, value1) to (status3, value3), ..., which was tedious. Fortunately, it is handled automatically in **generate_sankey_flow** function."
250 |    ]
251 |   },
252 |   {
253 |    "cell_type": "code",
254 |    "execution_count": 7,
255 |    "metadata": {
256 |     "collapsed": false
257 |    },
258 |    "outputs": [
259 |     {
260 |      "data": {
261 |       "text/html": [
262 |        "<div>\n",
263 |        "<table border=\"1\" class=\"dataframe\">\n",
264 |        "  <thead>\n",
265 |        "    <tr style=\"text-align: right;\">\n",
266 |        "      <th></th>\n",
267 |        "      <th>Survived</th>\n",
268 |        "      <th>Pclass</th>\n",
269 |        "      <th>Sex</th>\n",
270 |        "      <th>SibSp</th>\n",
271 |        "      <th>Parch</th>\n",
272 |        "      <th>Embarked</th>\n",
273 |        "    </tr>\n",
274 |        "  </thead>\n",
275 |        "  <tbody>\n",
276 |        "    <tr>\n",
277 |        "      <th>0</th>\n",
278 |        "      <td>0</td>\n",
279 |        "      <td>3</td>\n",
280 |        "      <td>male</td>\n",
281 |        "      <td>1</td>\n",
282 |        "      <td>0</td>\n",
283 |        "      <td>S</td>\n",
284 |        "    </tr>\n",
285 |        "    <tr>\n",
286 |        "      <th>1</th>\n",
287 |        "      <td>1</td>\n",
288 |        "      <td>1</td>\n",
289 |        "      <td>female</td>\n",
290 |        "      <td>1</td>\n",
291 |        "      <td>0</td>\n",
292 |        "      <td>C</td>\n",
293 |        "    </tr>\n",
294 |        "    <tr>\n",
295 |        "      <th>2</th>\n",
296 |        "      <td>1</td>\n",
297 |        "      <td>3</td>\n",
298 |        "      <td>female</td>\n",
299 |        "      <td>0</td>\n",
300 |        "      <td>0</td>\n",
301 |        "      <td>S</td>\n",
302 |        "    </tr>\n",
303 |        "    <tr>\n",
304 |        "      <th>3</th>\n",
305 |        "      <td>1</td>\n",
306 |        "      <td>1</td>\n",
307 |        "      <td>female</td>\n",
308 |        "      <td>1</td>\n",
309 |        "      <td>0</td>\n",
310 |        "      <td>S</td>\n",
311 |        "    </tr>\n",
312 |        "    <tr>\n",
313 |        "      <th>4</th>\n",
314 |        "      <td>0</td>\n",
315 |        "      <td>3</td>\n",
316 |        "      <td>male</td>\n",
317 |        "      <td>0</td>\n",
318 |        "      <td>0</td>\n",
319 |        "      <td>S</td>\n",
320 |        "    </tr>\n",
321 |        "  </tbody>\n",
322 |        "</table>\n",
323 |        "</div>"
324 |       ],
325 |       "text/plain": [
326 |        "   Survived  Pclass     Sex  SibSp  Parch Embarked\n",
327 |        "0         0       3    male      1      0        S\n",
328 |        "1         1       1  female      1      0        C\n",
329 |        "2         1       3  female      0      0        S\n",
330 |        "3         1       1  female      1      0        S\n",
331 |        "4         0       3    male      0      0        S"
332 |       ]
333 |      },
334 |      "execution_count": 7,
335 |      "metadata": {},
336 |      "output_type": "execute_result"
337 |     }
338 |    ],
339 |    "source": [
340 |     "raw[use_cols].head()"
341 |    ]
342 |   },
343 |   {
344 |    "cell_type": "markdown",
345 |    "metadata": {},
346 |    "source": [
347 |     "## color strategy\n",
348 |     "**node_color_type**, default=\"col_val\"  \n",
349 |     "Can be ['col', 'val', 'col_val', 'cus']\n",
350 |     "- 'col': each column has different color\n",
351 |     "- 'val': each unique value has different color (unique values through all columns)\n",
352 |     "- 'col_val': each unique value in each column has different color\n",
353 |     "- 'cus': customer provide node color mapping\n",
354 |     "\n",
355 |     "**node_color_mapping**, default=None  \n",
356 |     "Customized color mapping.\n",
357 |     "- node_color_mapping = '#fff'  \n",
358 |     "All nodes have the same color\n",
359 |     "- node_color_mapping = dict()  \n",
360 |     "\n",
361 |     "```python\n",
362 |     "node_color_mapping = {\n",
363 |     "    'type': 'col',\n",
364 |     "    'mapping': {\n",
365 |     "        column1: color1, column2: color2, ...\n",
366 |     "    }\n",
367 |     "}\n",
368 |     "\n",
369 |     "node_color_mapping = {\n",
370 |     "    'type': 'val',\n",
371 |     "    'mapping': {\n",
372 |     "        value1: color1, value2: color2, ...\n",
373 |     "    }\n",
374 |     "}\n",
375 |     "\n",
376 |     "node_color_mapping = {\n",
377 |     "    'type': 'col_val',\n",
378 |     "    'mapping': {\n",
379 |     "        column1: {value1: color1, value2: color2, ...},\n",
380 |     "        column2: {value1: color3, value2: color4, ...}\n",
381 |     "    }\n",
382 |     "}\n",
383 |     "```\n",
384 |     "\n",
385 |     "**link_color_type**, default='source'  \n",
386 |     "Can be ['source', 'target', 'both', 'same']\n",
387 |     "- 'source': same color as the source node\n",
388 |     "- 'target': same color as the target node\n",
389 |     "- 'both': color from both target and source\n",
390 |     "- 'same': all links have same color\n",
391 |     "\n",
392 |     "**link_color**, default=None  \n",
393 |     "Only required when `link_color_type=\"same\"`"
394 |    ]
395 |   },
396 |   {
397 |    "cell_type": "markdown",
398 |    "metadata": {},
399 |    "source": [
400 |     "## start generating sankey flow!"
401 |    ]
402 |   },
403 |   {
404 |    "cell_type": "code",
405 |    "execution_count": 8,
406 |    "metadata": {
407 |     "collapsed": true
408 |    },
409 |    "outputs": [],
410 |    "source": [
411 |     "import sys\n",
412 |     "sys.path.insert(0, 'src/')\n",
413 |     "import generate_sankey_flow"
414 |    ]
415 |   },
416 |   {
417 |    "cell_type": "markdown",
418 |    "metadata": {},
419 |    "source": [
420 |     "### use node_color_type"
421 |    ]
422 |   },
423 |   {
424 |    "cell_type": "code",
425 |    "execution_count": 9,
426 |    "metadata": {
427 |     "collapsed": true
428 |    },
429 |    "outputs": [],
430 |    "source": [
431 |     "generate_sankey_flow.draw_sankey_flow(df=raw[use_cols], node_color_type='col', link_color_type='source', \n",
432 |     "                                      width=1600, height=900, graph_name='Titanic', \n",
433 |     "                                      node_color_mapping=None, color_map=None, link_color=None)"
434 |    ]
435 |   },
436 |   {
437 |    "cell_type": "code",
438 |    "execution_count": 18,
439 |    "metadata": {
440 |     "collapsed": true
441 |    },
442 |    "outputs": [],
443 |    "source": [
444 |     "# change color_map\n",
445 |     "generate_sankey_flow.draw_sankey_flow(df=raw[use_cols], node_color_type='col', link_color_type='same', \n",
446 |     "                                      width=1600, height=900, graph_name='Titanic_same', \n",
447 |     "                                      node_color_mapping=None, color_map='tab20', link_color='#ccc')"
448 |    ]
449 |   },
450 |   {
451 |    "cell_type": "markdown",
452 |    "metadata": {},
453 |    "source": [
454 |     "### use customized node_color_type and provide node_color_mapping"
455 |    ]
456 |   },
457 |   {
458 |    "cell_type": "code",
459 |    "execution_count": 15,
460 |    "metadata": {
461 |     "collapsed": false
462 |    },
463 |    "outputs": [],
464 |    "source": [
465 |     "node_color_mapping = {\n",
466 |     "    'type': 'col',\n",
467 |     "    'mapping': {\n",
468 |     "        'Survived': '#9e0142', \n",
469 |     "        'Pclass': '#d53e4f', \n",
470 |     "        'Sex': '#f46d43', \n",
471 |     "        'SibSp': '#fdae61', \n",
472 |     "        'Parch': '#fee08b', \n",
473 |     "        'Embarked': '#ffffbf'\n",
474 |     "    }\n",
475 |     "}\n",
476 |     "generate_sankey_flow.draw_sankey_flow(df=raw[use_cols], node_color_type='cus', link_color_type='source', \n",
477 |     "                                      width=1600, height=900, graph_name='Titanic_col', \n",
478 |     "                                      node_color_mapping=node_color_mapping, color_map=None, link_color=None)"
479 |    ]
480 |   },
481 |   {
482 |    "cell_type": "code",
483 |    "execution_count": 16,
484 |    "metadata": {
485 |     "collapsed": true
486 |    },
487 |    "outputs": [],
488 |    "source": [
489 |     "node_color_mapping = {\n",
490 |     "    'type': 'val',\n",
491 |     "    'mapping': {\n",
492 |     "        0: '#9e0142',\n",
493 |     "        1: '#d53e4f',\n",
494 |     "        2: '#f46d43',\n",
495 |     "        3: '#fdae61',\n",
496 |     "        4: '#fee08b',\n",
497 |     "        5: '#ffffbf',\n",
498 |     "        6: '#e6f598',\n",
499 |     "        8: '#abdda4',\n",
500 |     "        'male': '#66c2a5',\n",
501 |     "        'female': '#3288bd',\n",
502 |     "        'S': '#5e4fa2',\n",
503 |     "        'C': '#9e0142',\n",
504 |     "        'Q': '#d53e4f',\n",
505 |     "        'unknown': '#f46d43'\n",
506 |     "    }\n",
507 |     "}\n",
508 |     "generate_sankey_flow.draw_sankey_flow(df=raw[use_cols], node_color_type='cus', link_color_type='source', \n",
509 |     "                                      width=1600, height=900, graph_name='Titanic_val', \n",
510 |     "                                      node_color_mapping=node_color_mapping, color_map=None, link_color=None)"
511 |    ]
512 |   },
513 |   {
514 |    "cell_type": "code",
515 |    "execution_count": 17,
516 |    "metadata": {
517 |     "collapsed": true
518 |    },
519 |    "outputs": [],
520 |    "source": [
521 |     "node_color_mapping = {\n",
522 |     "    'type': 'col_val',\n",
523 |     "    'mapping': {\n",
524 |     "        'Survived': {0: '#9e0142', 1: '#5e4fa2'},\n",
525 |     "        'Pclass': {1: '#d53e4f', 2: '#3288bd', 3: '#f46d43'},\n",
526 |     "        'Sex': {'male': '#fdae61', 'female': '#66c2a5'},\n",
527 |     "        'SibSp': {0: '#fee08b', 1: '#ffffbf', 2: '#e6f598', 3: '#abdda4', 4: '#9e0142', 5: '#5e4fa2', 8: '#d53e4f'},\n",
528 |     "        'Parch': {0: '#f46d43', 1: '#fdae61', 2: '#fee08b', 3: '#ffffbf', 4: '#e6f598', 5: '#abdda4', 6: '#66c2a5'},\n",
529 |     "        'Embarked': {'S': '#3288bd', 'C': '#9e0142', 'Q': '#d53e4f', 'unknown': '#f46d43'}\n",
530 |     "    }\n",
531 |     "}\n",
532 |     "generate_sankey_flow.draw_sankey_flow(df=raw[use_cols], node_color_type='cus', link_color_type='source', \n",
533 |     "                                      width=1600, height=900, graph_name='Titanic_col_val', \n",
534 |     "                                      node_color_mapping=node_color_mapping, color_map=None, link_color=None)"
535 |    ]
536 |   }
537 |  ],
538 |  "metadata": {
539 |   "anaconda-cloud": {},
540 |   "kernelspec": {
541 |    "display_name": "Python [default]",
542 |    "language": "python",
543 |    "name": "python2"
544 |   },
545 |   "language_info": {
546 |    "codemirror_mode": {
547 |     "name": "ipython",
548 |     "version": 2.0
549 |    },
550 |    "file_extension": ".py",
551 |    "mimetype": "text/x-python",
552 |    "name": "python",
553 |    "nbconvert_exporter": "python",
554 |    "pygments_lexer": "ipython2",
555 |    "version": "2.7.12"
556 |   }
557 |  },
558 |  "nbformat": 4,
559 |  "nbformat_minor": 0
560 | }


--------------------------------------------------------------------------------
/sankey/src/generate_sankey_flow.py:
--------------------------------------------------------------------------------
  1 | 
  2 | import matplotlib
  3 | import matplotlib.pyplot as plt
  4 | 
  5 | import json
  6 | import jinja2
  7 | import os
  8 | 
  9 | import sys
 10 | reload(sys)
 11 | sys.setdefaultencoding('utf-8')
 12 | 
 13 | 
 14 | def _get_node_names(df):
 15 | 	'''
 16 | 	:param df: pandas DataFrame, each column represents a state
 17 | 	:return:
 18 | 		dictionary of (column name, column unique value list)
 19 | 		which means (state name, unique values in this state)
 20 | 	'''
 21 | 
 22 | 	node_infos = {}
 23 | 
 24 | 	for col in df.columns.values:
 25 | 		col_values = sorted(df[col].unique())
 26 | 		col_names = []
 27 | 
 28 | 		for col_value in col_values:
 29 | 			node_name = '%s: %s' %(col, str(col_value))
 30 | 			col_names.append(node_name)
 31 | 
 32 | 		node_infos[col] = {'col_names': col_names, 'col_values': col_values}
 33 | 
 34 | 	return node_infos
 35 | 
 36 | 
 37 | def _get_node_colors(node_infos, node_color_type, node_color_mapping, cm):
 38 | 	'''
 39 | 	:param node_infos:
 40 | 		dictionary of (state name, unique values in this state)
 41 | 	:param node_color_type:
 42 | 		node coloring strategy, can be one of ['col', 'val', 'col_val', 'cus']
 43 | 		- 'col': each column has different color
 44 | 		- 'val': each unique value has different color (unique values through all columns)
 45 | 		- 'col_val': each unique value in each column has different color
 46 | 		- 'cus': customer provide node color mapping
 47 | 	:param node_color_mapping:
 48 | 		if node_color_type == 'cus', node_color_mapping should't be None
 49 | 	:param cm: matplotlib color map
 50 | 	:return:
 51 | 		dictionary of (node name, node color)
 52 | 	'''
 53 | 
 54 | 	node_colors = {}
 55 | 
 56 | 	if (node_color_type == 'col') or (node_color_type == 'cus' and node_color_mapping['type'] == 'col'):
 57 | 		for col_idx, col in enumerate(node_infos.keys()):
 58 | 			if node_color_type == 'cus':
 59 | 				col_color = node_color_mapping['mapping'][col]
 60 | 			else:
 61 | 				col_color = cm(col_idx % 20)
 62 | 			for col_name in node_infos[col]['col_names']:
 63 | 				node_colors[col_name] = matplotlib.colors.rgb2hex(col_color)
 64 | 
 65 | 	if (node_color_type == 'val') or (node_color_type == 'cus' and node_color_mapping['type'] == 'val'):
 66 | 		unique_values = []
 67 | 		for col in node_infos.keys():
 68 | 			unique_values += list(node_infos[col]['col_values'])
 69 | 		unique_values = list(set(unique_values))
 70 | 
 71 | 		for col in node_infos.keys():
 72 | 			for col_value in node_infos[col]['col_values']:
 73 | 				if node_color_type == 'cus':
 74 | 					val_color = node_color_mapping['mapping'][col_value]
 75 | 				else:
 76 | 					val_idx = unique_values.index(col_value)
 77 | 					val_color = matplotlib.colors.rgb2hex(cm(val_idx % 20))
 78 | 				node_colors['%s: %s' %(col, str(col_value))] = val_color
 79 | 
 80 | 	if (node_color_type == 'col_val') or (node_color_type == 'cus' and node_color_mapping['type'] == 'col_val'):
 81 | 		for col_idx, col in enumerate(node_infos.keys()):
 82 | 			if node_color_type == 'cus':
 83 | 				for col_value in node_color_mapping['mapping'][col]:
 84 | 					node_colors['%s: %s' % (col, str(col_value))] = node_color_mapping['mapping'][col][col_value]
 85 | 			else:
 86 | 				for col_name_idx, col_name in enumerate(node_infos[col]['col_names']):
 87 | 					col_name_idx_true = col_idx * len(node_infos[col]['col_names']) + col_name_idx
 88 | 					node_colors[col_name] = matplotlib.colors.rgb2hex(cm(col_name_idx_true % 20))
 89 | 
 90 | 	return node_colors
 91 | 
 92 | 
 93 | def _prepare_sankey_data(df, node_colors):
 94 | 	'''
 95 | 	:param df: pandas DataFrame, each column represents a state
 96 | 	:param node_colors: dictionary of (node name, node color)
 97 | 	:return:
 98 | 		dictionary of links and nodes
 99 | 	'''
100 | 
101 | 	links = []
102 | 	for i in range(len(df.columns.values)-1):
103 | 		source_col, target_col = df.columns.values[i], df.columns.values[i+1]
104 | 		temp_df = df[[source_col, target_col]]
105 | 		temp_df['count'] = 1
106 | 		temp_df = temp_df.rename(columns={source_col: 'source', target_col: 'target'})
107 | 		temp_df_gp = temp_df.groupby(['source', 'target'], as_index=False).count()
108 | 
109 | 		temp_df_gp['source'] = temp_df_gp['source'].apply(lambda x : '%s: %s' %(source_col, str(x)))
110 | 		temp_df_gp['target'] = temp_df_gp['target'].apply(lambda x : '%s: %s' %(target_col, str(x)))
111 | 
112 | 		temp_df_gp['color_source'] = temp_df_gp['source'].apply(lambda x : node_colors[x] if x in node_colors.keys() else '#000')
113 | 		temp_df_gp['color_target'] = temp_df_gp['target'].apply(lambda x : node_colors[x] if x in node_colors.keys() else '#000')
114 | 		temp_df_gp['value'] = temp_df_gp['count'].map(str)
115 | 
116 | 		links+= temp_df_gp[['source', 'target', 'value']].to_dict('records')
117 | 
118 | 	nodes = [{'name': n, 'color': c} for (n, c) in node_colors.items()]
119 | 	data = {
120 | 		'links': links,
121 | 		'nodes': nodes
122 | 	}
123 | 	return data
124 | 
125 | 
126 | def draw_sankey_flow(df, node_color_type, link_color_type, width, height, graph_name=None,
127 | 					 node_color_mapping=None, color_map=None, link_color=None):
128 | 	'''
129 | 	:param df:
130 | 		pandas DataFrame, each column represents a state
131 | 	:param node_color_type:
132 | 		node coloring strategy, can be one of ['col', 'val', 'col_val', 'cus']
133 | 			- 'col': each column has different color
134 | 			- 'val': each unique value has different color (unique values through all columns)
135 | 			- 'col_val': each unique value in each column has different color
136 | 			- 'cus': customer provide node color mapping
137 | 	:param link_color_type:
138 | 		link coloring strategy, default='source'
139 | 		Can be one of ['source', 'target', 'both', 'same']
140 | 			- 'source': same color as the source node
141 | 			- 'target': same color as the target node
142 | 			- 'both': color from both target and source
143 | 			- 'same': all links have same color
144 | 	:param width: wdith
145 | 	:param height: height
146 | 	:param graph_name: name of the graph
147 | 	:param node_color_mapping:
148 | 		if node_color_type == 'cus', color_mapping should be provided
149 | 		example:
150 | 		node_color_mapping = {
151 | 			'type': 'col',
152 | 			'mapping': {
153 | 				column1: color1, column2: color2, ...
154 | 			}
155 | 		}
156 | 	:param color_map: matplotlib color map
157 | 	:param link_color:
158 | 		if link_color_type == 'same', link color should be provided
159 | 	'''
160 | 
161 | 	# get node infos
162 | 	node_infos = _get_node_names(df)
163 | 
164 | 	# get node colors
165 | 	if color_map is None:
166 | 		cm = plt.cm.get_cmap('Vega20')
167 | 	else:
168 | 		try:
169 | 			cm = plt.cm.get_cmap(color_map)
170 | 		except:
171 | 			cm = plt.cm.get_cmap('Vega20')
172 | 
173 | 	node_colors = _get_node_colors(node_infos, node_color_type, node_color_mapping, cm)
174 | 
175 | 	# prepare data for sankey
176 | 	sankey_data = _prepare_sankey_data(df, node_colors)
177 | 
178 | 	# render the output
179 | 	temp = open('src/sankey_flow_template.html').read()
180 | 	template = jinja2.Template(temp)
181 | 
182 | 	# create the output root if it is not exits
183 | 	if not os.path.exists('sankey_flow_output'):
184 | 		os.mkdir('sankey_flow_output')
185 | 
186 | 	# generate output html
187 | 	if graph_name is None:
188 | 		output_path = 'sankey_flow_output/sankey_flow_output.html'
189 | 	else:
190 | 		output_path = 'sankey_flow_output/sankey_flow_%s.html' %(graph_name)
191 | 	with open(output_path, 'wb') as fh:
192 | 		fh.write(template.render({'data': json.dumps(sankey_data), 'link_color_type': link_color_type,
193 | 								  'link_color': link_color, 'width': width, 'height': height}))
194 | 	print('The output is in %s. Enjoy!' %(output_path))
195 | 


--------------------------------------------------------------------------------
/sankey/src/sankey_flow_template.html:
--------------------------------------------------------------------------------
  1 | <!DOCTYPE html>
  2 | <head>
  3 | <meta charset="utf-8">
  4 | 
  5 | <style>
  6 |  body {
  7 |   font-family: helvetica;
  8 |   font-size: 10px;
  9 |  }
 10 | 
 11 | .node rect {
 12 |   shape-rendering: crispEdges;
 13 | }
 14 |  
 15 | .node text {
 16 |   pointer-events: none;
 17 |   fill: #555;
 18 |   font-weight: bold;
 19 |   font-size: 12px;
 20 | }
 21 | 
 22 | #mycontainer {
 23 |   margin: 3vh 0 0 5vw;
 24 | }
 25 |  
 26 | </style>
 27 | </head>
 28 | 
 29 | <body>
 30 | <div id='mycontainer'>
 31 |   <div id="chart"></div>
 32 | </div>
 33 | 
 34 | 
 35 | <script src="https://d3js.org/d3.v3.min.js"></script>
 36 | <script type="text/javascript">
 37 |   // include sankey from here
 38 | d3.sankey = function() {
 39 |   var sankey = {},
 40 |       nodeWidth = 24,
 41 |       nodePadding = 8,
 42 |       size = [1, 1],
 43 |       nodes = [],
 44 |       links = [];
 45 | 
 46 |   sankey.nodeWidth = function(_) {
 47 |     if (!arguments.length) return nodeWidth;
 48 |     nodeWidth = +_;
 49 |     return sankey;
 50 |   };
 51 | 
 52 |   sankey.nodePadding = function(_) {
 53 |     if (!arguments.length) return nodePadding;
 54 |     nodePadding = +_;
 55 |     return sankey;
 56 |   };
 57 | 
 58 |   sankey.nodes = function(_) {
 59 |     if (!arguments.length) return nodes;
 60 |     nodes = _;
 61 |     return sankey;
 62 |   };
 63 | 
 64 |   sankey.links = function(_) {
 65 |     if (!arguments.length) return links;
 66 |     links = _;
 67 |     return sankey;
 68 |   };
 69 | 
 70 |   sankey.size = function(_) {
 71 |     if (!arguments.length) return size;
 72 |     size = _;
 73 |     return sankey;
 74 |   };
 75 | 
 76 |   sankey.layout = function(iterations) {
 77 |     computeNodeLinks();
 78 |     computeNodeValues();
 79 |     computeNodeBreadths();
 80 |     computeNodeDepths(iterations);
 81 |     computeLinkDepths();
 82 |     return sankey;
 83 |   };
 84 | 
 85 |   sankey.relayout = function() {
 86 |     computeLinkDepths();
 87 |     return sankey;
 88 |   };
 89 | 
 90 |   sankey.link = function() {
 91 |     var curvature = .5;
 92 | 
 93 |     function link(d) {
 94 |       var x0 = d.source.x + d.source.dx,
 95 |           x1 = d.target.x,
 96 |           xi = d3.interpolateNumber(x0, x1),
 97 |           x2 = xi(curvature),
 98 |           x3 = xi(1 - curvature),
 99 |           y0 = d.source.y + d.sy + d.dy / 2,
100 |           y1 = d.target.y + d.ty + d.dy / 2;
101 |       return "M" + x0 + "," + y0
102 |            + "C" + x2 + "," + y0
103 |            + " " + x3 + "," + y1
104 |            + " " + x1 + "," + y1;
105 |     }
106 | 
107 |     link.curvature = function(_) {
108 |       if (!arguments.length) return curvature;
109 |       curvature = +_;
110 |       return link;
111 |     };
112 | 
113 |     return link;
114 |   };
115 | 
116 |   // Populate the sourceLinks and targetLinks for each node.
117 |   // Also, if the source and target are not objects, assume they are indices.
118 |   function computeNodeLinks() {
119 |     nodes.forEach(function(node) {
120 |       node.sourceLinks = [];
121 |       node.targetLinks = [];
122 |     });
123 |     links.forEach(function(link) {
124 |       var source = link.source,
125 |           target = link.target;
126 |       if (typeof source === "number") source = link.source = nodes[link.source];
127 |       if (typeof target === "number") target = link.target = nodes[link.target];
128 |       source.sourceLinks.push(link);
129 |       target.targetLinks.push(link);
130 |     });
131 |   }
132 | 
133 |   // Compute the value (size) of each node by summing the associated links.
134 |   function computeNodeValues() {
135 |     nodes.forEach(function(node) {
136 |       node.value = Math.max(
137 |         d3.sum(node.sourceLinks, value),
138 |         d3.sum(node.targetLinks, value)
139 |       );
140 |     });
141 |   }
142 | 
143 |   // Iteratively assign the breadth (x-position) for each node.
144 |   // Nodes are assigned the maximum breadth of incoming neighbors plus one;
145 |   // nodes with no incoming links are assigned breadth zero, while
146 |   // nodes with no outgoing links are assigned the maximum breadth.
147 |   function computeNodeBreadths() {
148 |     var remainingNodes = nodes,
149 |         nextNodes,
150 |         x = 0;
151 | 
152 |     while (remainingNodes.length) {
153 |       nextNodes = [];
154 |       remainingNodes.forEach(function(node) {
155 |         node.x = x;
156 |         node.dx = nodeWidth;
157 |         node.sourceLinks.forEach(function(link) {
158 |           nextNodes.push(link.target);
159 |         });
160 |       });
161 |       remainingNodes = nextNodes;
162 |       ++x;
163 |     }
164 | 
165 |     //
166 |     moveSinksRight(x);
167 |     scaleNodeBreadths((width - nodeWidth) / (x - 1));
168 |   }
169 | 
170 |   function moveSourcesRight() {
171 |     nodes.forEach(function(node) {
172 |       if (!node.targetLinks.length) {
173 |         node.x = d3.min(node.sourceLinks, function(d) { return d.target.x; }) - 1;
174 |       }
175 |     });
176 |   }
177 | 
178 |   function moveSinksRight(x) {
179 |     nodes.forEach(function(node) {
180 |       if (!node.sourceLinks.length) {
181 |         node.x = x - 1;
182 |       }
183 |     });
184 |   }
185 | 
186 |   function scaleNodeBreadths(kx) {
187 |     nodes.forEach(function(node) {
188 |       node.x *= kx;
189 |     });
190 |   }
191 | 
192 |   function computeNodeDepths(iterations) {
193 |     var nodesByBreadth = d3.nest()
194 |         .key(function(d) { return d.x; })
195 |         .sortKeys(d3.ascending)
196 |         .entries(nodes)
197 |         .map(function(d) { return d.values; });
198 | 
199 |     //
200 |     initializeNodeDepth();
201 |     resolveCollisions();
202 |     for (var alpha = 1; iterations > 0; --iterations) {
203 |       relaxRightToLeft(alpha *= .99);
204 |       resolveCollisions();
205 |       relaxLeftToRight(alpha);
206 |       resolveCollisions();
207 |     }
208 | 
209 |     function initializeNodeDepth() {
210 |       var ky = d3.min(nodesByBreadth, function(nodes) {
211 |         return (size[1] - (nodes.length - 1) * nodePadding) / d3.sum(nodes, value);
212 |       });
213 | 
214 |       nodesByBreadth.forEach(function(nodes) {
215 |         nodes.forEach(function(node, i) {
216 |           node.y = i;
217 |           node.dy = node.value * ky;
218 |         });
219 |       });
220 | 
221 |       links.forEach(function(link) {
222 |         link.dy = link.value * ky;
223 |       });
224 |     }
225 | 
226 |     function relaxLeftToRight(alpha) {
227 |       nodesByBreadth.forEach(function(nodes, breadth) {
228 |         nodes.forEach(function(node) {
229 |           if (node.targetLinks.length) {
230 |             var y = d3.sum(node.targetLinks, weightedSource) / d3.sum(node.targetLinks, value);
231 |             node.y += (y - center(node)) * alpha;
232 |           }
233 |         });
234 |       });
235 | 
236 |       function weightedSource(link) {
237 |         return center(link.source) * link.value;
238 |       }
239 |     }
240 | 
241 |     function relaxRightToLeft(alpha) {
242 |       nodesByBreadth.slice().reverse().forEach(function(nodes) {
243 |         nodes.forEach(function(node) {
244 |           if (node.sourceLinks.length) {
245 |             var y = d3.sum(node.sourceLinks, weightedTarget) / d3.sum(node.sourceLinks, value);
246 |             node.y += (y - center(node)) * alpha;
247 |           }
248 |         });
249 |       });
250 | 
251 |       function weightedTarget(link) {
252 |         return center(link.target) * link.value;
253 |       }
254 |     }
255 | 
256 |     function resolveCollisions() {
257 |       nodesByBreadth.forEach(function(nodes) {
258 |         var node,
259 |             dy,
260 |             y0 = 0,
261 |             n = nodes.length,
262 |             i;
263 | 
264 |         // Push any overlapping nodes down.
265 |         nodes.sort(ascendingDepth);
266 |         for (i = 0; i < n; ++i) {
267 |           node = nodes[i];
268 |           dy = y0 - node.y;
269 |           if (dy > 0) node.y += dy;
270 |           y0 = node.y + node.dy + nodePadding;
271 |         }
272 | 
273 |         // If the bottommost node goes outside the bounds, push it back up.
274 |         dy = y0 - nodePadding - size[1];
275 |         if (dy > 0) {
276 |           y0 = node.y -= dy;
277 | 
278 |           // Push any overlapping nodes back up.
279 |           for (i = n - 2; i >= 0; --i) {
280 |             node = nodes[i];
281 |             dy = node.y + node.dy + nodePadding - y0;
282 |             if (dy > 0) node.y -= dy;
283 |             y0 = node.y;
284 |           }
285 |         }
286 |       });
287 |     }
288 | 
289 |     function ascendingDepth(a, b) {
290 |       return a.y - b.y;
291 |     }
292 |   }
293 | 
294 |   function computeLinkDepths() {
295 |     nodes.forEach(function(node) {
296 |       node.sourceLinks.sort(ascendingTargetDepth);
297 |       node.targetLinks.sort(ascendingSourceDepth);
298 |     });
299 |     nodes.forEach(function(node) {
300 |       var sy = 0, ty = 0;
301 |       node.sourceLinks.forEach(function(link) {
302 |         link.sy = sy;
303 |         sy += link.dy;
304 |       });
305 |       node.targetLinks.forEach(function(link) {
306 |         link.ty = ty;
307 |         ty += link.dy;
308 |       });
309 |     });
310 | 
311 |     function ascendingSourceDepth(a, b) {
312 |       return a.source.y - b.source.y;
313 |     }
314 | 
315 |     function ascendingTargetDepth(a, b) {
316 |       return a.target.y - b.target.y;
317 |     }
318 |   }
319 | 
320 |   function center(node) {
321 |     return node.y + node.dy / 2;
322 |   }
323 | 
324 |   function value(link) {
325 |     return link.value;
326 |   }
327 | 
328 |   return sankey;
329 | };
330 | </script>
331 | <script>
332 | 
333 | var data = {{ data|safe }};
334 | var link_color_type = '{{ link_color_type }}';
335 | var link_color = '{{ link_color }}';
336 | var width = {{ width }}, height = {{ height }};
337 | 
338 | var margin = {top: 50, right: 200, bottom: 100, left: 0},
339 |     width = width - margin.left - margin.right,
340 |     height = height - margin.top - margin.bottom;
341 | 
342 | // Set the sankey diagram properties
343 | var sankey = d3.sankey()
344 |     .nodeWidth(25)
345 |     .nodePadding(20)
346 |     .size([width, height]);
347 |  
348 | var path = sankey.link();
349 | update_sankey(data);
350 | 
351 | function update_sankey(data){
352 |   // remember deep copy here
353 |   var graph = data;
354 | 
355 |   // append the svg canvas to the page
356 |   d3.select('#chart').select('svg').remove();
357 |   var svg = d3.select("#chart").append("svg")
358 |       .attr("width", width + margin.left + margin.right)
359 |       .attr("height", height + margin.top + margin.bottom)
360 |     .append("g")
361 |       .attr("transform", 
362 |             "translate(" + margin.left + "," + margin.top + ")");
363 | 
364 |   var nodeMap = {};
365 |   graph.nodes.forEach(function(x) { nodeMap[x.name] = x; });
366 |   graph.links = graph.links.map(function(x) {
367 |     return {
368 |       source: nodeMap[x.source],
369 |       target: nodeMap[x.target],
370 |       value: x.value
371 |     };
372 |   });
373 | 
374 |   sankey
375 |       .nodes(graph.nodes)
376 |       .links(graph.links)
377 |       .layout(32);
378 |    
379 |   var defs = svg.append("defs");
380 | 
381 |   // create gradients for the links
382 |   var grads = defs.selectAll("linearGradient")
383 |     .data(graph.links, getGradID);
384 | 
385 |   grads.enter().append("linearGradient")
386 |     .attr("id", getGradID)
387 |     .attr("gradientUnits", "userSpaceOnUse");
388 | 
389 |   positionGrads(grads);
390 | 
391 |   grads.html("") //erase any existing <stop> elements on update
392 |     .append("stop")
393 |     .attr("offset", "0%")
394 |     .attr("stop-color", function(d) {
395 |       return nodeColor((+d.source.x <= +d.target.x) ?
396 |         d.source : d.target);
397 |     });
398 | 
399 |   grads.append("stop")
400 |     .attr("offset", "100%")
401 |     .attr("stop-color", function(d) {
402 |       return nodeColor((+d.source.x > +d.target.x) ?
403 |         d.source : d.target)
404 |     });
405 | 
406 |   // add in the links
407 |   var link = svg.append("g").selectAll(".link")
408 |     .data(graph.links)
409 |     .enter().append("path")
410 |     .attr("class", "link")
411 |     .attr("d", path)
412 |     .style("fill", "none")
413 |     .style("stroke", function(d) {
414 |       if(link_color_type=="source") return d.source.color;
415 |       else if(link_color_type=="target") return d.target.color;
416 |       else if(link_color_type=="both") return "url(#" + getGradID(d) + ")";
417 |       else return link_color;
418 |     })
419 |     .style("stroke-opacity", "0.3")
420 |     .on("mouseover", function() {
421 |       d3.select(this).style("stroke-opacity", "0.6")
422 |     })
423 |     .on("mouseout", function() {
424 |       d3.select(this).style("stroke-opacity", "0.3")
425 |     })
426 |     .style("stroke-width", function(d) {
427 |       return Math.max(1, d.dy);
428 |     })
429 |     .sort(function(a, b) {
430 |       return b.dy - a.dy;
431 |     });
432 | 
433 |   // add the link titles
434 |   link.append("title")
435 |         .text(function(d) {
436 |         return d.source.name + " → " + 
437 |                 d.target.name + "\n" + d.value; });
438 |  
439 | // add in the nodes
440 |   var node = svg.append("g").selectAll(".node")
441 |       .data(graph.nodes)
442 |     .enter().append("g")
443 |       .attr("class", "node")
444 |       .attr("transform", function(d) { 
445 |       return "translate(" + d.x + "," + d.y + ")"; });
446 |  
447 | // add the rectangles for the nodes
448 |   node.append("rect")
449 |       .attr("height", function(d) { return d.dy; })
450 |       .attr("width", sankey.nodeWidth())
451 |       .style("fill", function(d) { 
452 |       return d.color = d.color; })
453 |       .style('stroke-width', 0)
454 |     .append("title")
455 |       .text(function(d) { 
456 |       return d.name + "\n" + d.value; });
457 |  
458 | // add in the title for the nodes
459 |   node.append("text")
460 |       .attr("x", 6 + sankey.nodeWidth())
461 |       .attr("y", function(d) { return d.dy / 2; })
462 |       .attr("dy", ".35em")
463 |       .attr("text-anchor", "start")
464 |       .attr("transform", null)
465 |       .text(function(d) { return d.name; });
466 | }
467 | 
468 | // define utility functions
469 | function getGradID(d) {
470 |   return "linkGrad-" + d.source.name.replace(/ /g, "") + "-" + d.target.name.replace(/ /g, "");
471 | }
472 | 
473 | function nodeColor(d) {
474 |   return d.color;
475 | }
476 | 
477 | function positionGrads(grads) {
478 |   grads.attr("x1", function(d) {
479 |       return d.source.x;
480 |     })
481 |     .attr("y1", function(d) {
482 |       return d.source.y;
483 |     })
484 |     .attr("x2", function(d) {
485 |       return d.target.x;
486 |     })
487 |     .attr("y2", function(d) {
488 |       return d.target.y;
489 |     });
490 | }
491 | 
492 | </script>
493 |  
494 | </body>
495 | </html>


--------------------------------------------------------------------------------
/tree/README.md:
--------------------------------------------------------------------------------
 1 | ## Visualize a sklearn Decision Tree Classifier
 2 | - simple tree   
 3 |   <img src="https://github.com/SauceCat/Nuance/blob/master/tree/image/simple_tree.gif" />    
 4 | - sankey tree  
 5 |   <img src="https://github.com/SauceCat/Nuance/blob/master/tree/image/sankey_tree.gif" />      
 6 | 
 7 | ## How to use?
 8 | Take **simple_tree** as an example:
 9 | 1. Download the folder [**tree**](https://github.com/SauceCat/Nuance/tree/master/tree)
10 | 2. The folder structure:
11 |     ```
12 |     tree
13 |       - src: folder for all codes regarding the visualization
14 |       - data: folder for the test data
15 |       - simple_tree_output: folder for simple_tree outputs (will be re-generated if it was deleted)
16 |       - sankey_tree_output: folder for sankey_tree outputs (will be re-generated if it was deleted)
17 |       - generate_tree_test.ipynb: instructions and examples in jupyter notebook
18 |       - ...
19 |     ```
20 | 3. Install `jinja2`
21 |     ```
22 |     pip install jinja2
23 |     ```
24 | 4. Use simple_tree or sankey_tree visualization: (visualization depends on D3.js, so you need to connect to the network)
25 |     ```python
26 |     import sys
27 |     sys.path.insert(0, 'src/')
28 |     import generate_tree
29 |     
30 |     # simple tree
31 |     generate_tree.generate_simple_tree(tree_title='Titanic_Tree', tree_model=dt, X=titanic[features], 
32 |                                    target_names=['Not Survived', 'Survived'], target_colors = None,
33 |                                    color_map=None, width=1500, height=1000)
34 | 				   
35 |     # sankey tree
36 |     generate_tree.generate_sankey_tree(tree_title='Titanic_Tree', tree_model=dt, X=titanic[features], 
37 |                                    target_names=['Not Survived', 'Survived'], target_colors = None,
38 |                                    color_map=None, width=1500, height=1200)
39 |     ```
40 | 5. A html file would be generated in [simple_tree_output](https://github.com/SauceCat/Nuance/tree/master/tree/simple_tree_output) or [sankey_tree_output](https://github.com/SauceCat/Nuance/tree/master/tree/sankey_tree_output) folder. Open it using any browser you like (I like Chrome anyway). 
41 | 
42 | ## Parameters
43 | ```python
44 | def generate_simple_tree(tree_title, tree_model, X, target_names, 
45 |                          target_colors=None, color_map=None, width=None, height=None):
46 | 	'''
47 | 	visualize a sklearn Decision Tree Classifier
48 | 
49 | 	:param tree_title: string
50 | 		name of the tree
51 | 	:param tree_model: a fitted sklearn Decision Tree Classifier
52 | 	:param X: pandas DataFrame
53 | 		dataset model was fitted on
54 | 	:param target_names: list
55 | 		list of names for targets
56 | 	:param target_colors: list, default=None
57 | 		list of colors for targets
58 | 	:param color_map: string, default=None
59 | 		matplotlib color map name, like 'Vega20'
60 | 	:param width: int
61 | 		width of the html page
62 | 	:param height: int
63 | 		height of the html page
64 | 	'''
65 | 	
66 | def generate_sankey_tree(tree_title, tree_model, X, target_names,
67 | 			 target_colors=None, color_map=None, width=None, height=None):
68 | 	'''
69 | 	visualize a sklearn Decision Tree Classifier
70 | 
71 | 	:param tree_title: string
72 | 		name of the tree
73 | 	:param tree_model: a fitted sklearn Decision Tree Classifier
74 | 	:param X: pandas DataFrame
75 | 		dataset model was fitted on
76 | 	:param target_names: list
77 | 		list of names for targets
78 | 	:param target_colors: list, default=None
79 | 		list of colors for targets
80 | 	:param color_map: string, default=None
81 | 		matplotlib color map name, like 'Vega20'
82 | 	:param width: int
83 | 		width of the html page
84 | 	:param height: int
85 | 		height of the html page
86 | 	'''
87 | ```
88 | 


--------------------------------------------------------------------------------
/tree/generate_tree_test.ipynb:
--------------------------------------------------------------------------------
  1 | {
  2 |  "cells": [
  3 |   {
  4 |    "cell_type": "markdown",
  5 |    "metadata": {},
  6 |    "source": [
  7 |     "## test generate_tree visualization with \n",
  8 |     "1. Titanic dataset: Binary classification Decision Tree\n",
  9 |     "2. Iris dataset: Multiclass classification Decision Tree"
 10 |    ]
 11 |   },
 12 |   {
 13 |    "cell_type": "markdown",
 14 |    "metadata": {},
 15 |    "source": [
 16 |     "## Titanic: Binary classifier"
 17 |    ]
 18 |   },
 19 |   {
 20 |    "cell_type": "code",
 21 |    "execution_count": 1,
 22 |    "metadata": {
 23 |     "collapsed": true
 24 |    },
 25 |    "outputs": [],
 26 |    "source": [
 27 |     "import pandas as pd\n",
 28 |     "import numpy as np\n",
 29 |     "\n",
 30 |     "from sklearn.tree import DecisionTreeClassifier"
 31 |    ]
 32 |   },
 33 |   {
 34 |    "cell_type": "code",
 35 |    "execution_count": 2,
 36 |    "metadata": {
 37 |     "collapsed": true
 38 |    },
 39 |    "outputs": [],
 40 |    "source": [
 41 |     "# read dataset\n",
 42 |     "titanic = pd.read_csv('data/titanic_train.csv')\n",
 43 |     "\n",
 44 |     "# impute null values\n",
 45 |     "titanic[\"Age\"] = titanic[\"Age\"].fillna(titanic[\"Age\"].dropna().median())\n",
 46 |     "titanic[\"Embarked\"] = titanic[\"Embarked\"].fillna(\"S\")\n",
 47 |     "\n",
 48 |     "# handle categrical features\n",
 49 |     "titanic['Sex'] = titanic['Sex'].apply(lambda x : 1 if x == 'male' else 0)\n",
 50 |     "titanic = pd.get_dummies(titanic, columns=['Embarked'])\n",
 51 |     "\n",
 52 |     "# features to used\n",
 53 |     "features = ['Pclass', 'Sex', 'Age', 'SibSp', 'Parch', 'Fare', 'Embarked_C', 'Embarked_Q', 'Embarked_S']"
 54 |    ]
 55 |   },
 56 |   {
 57 |    "cell_type": "code",
 58 |    "execution_count": 3,
 59 |    "metadata": {},
 60 |    "outputs": [
 61 |     {
 62 |      "name": "stdout",
 63 |      "output_type": "stream",
 64 |      "text": [
 65 |       "CPU times: user 4.45 ms, sys: 411 µs, total: 4.86 ms\n",
 66 |       "Wall time: 3.83 ms\n"
 67 |      ]
 68 |     },
 69 |     {
 70 |      "data": {
 71 |       "text/plain": [
 72 |        "DecisionTreeClassifier(class_weight=None, criterion='gini', max_depth=None,\n",
 73 |        "            max_features=None, max_leaf_nodes=20,\n",
 74 |        "            min_impurity_decrease=0.0, min_impurity_split=None,\n",
 75 |        "            min_samples_leaf=1, min_samples_split=2,\n",
 76 |        "            min_weight_fraction_leaf=0.0, presort=False, random_state=24,\n",
 77 |        "            splitter='best')"
 78 |       ]
 79 |      },
 80 |      "execution_count": 3,
 81 |      "metadata": {},
 82 |      "output_type": "execute_result"
 83 |     }
 84 |    ],
 85 |    "source": [
 86 |     "dt = DecisionTreeClassifier(random_state=24, max_leaf_nodes=20)\n",
 87 |     "%time dt.fit(titanic[features], titanic['Survived'])"
 88 |    ]
 89 |   },
 90 |   {
 91 |    "cell_type": "markdown",
 92 |    "metadata": {},
 93 |    "source": [
 94 |     "### visualize the tree model"
 95 |    ]
 96 |   },
 97 |   {
 98 |    "cell_type": "code",
 99 |    "execution_count": 4,
100 |    "metadata": {
101 |     "collapsed": true
102 |    },
103 |    "outputs": [],
104 |    "source": [
105 |     "import sys\n",
106 |     "sys.path.insert(0, 'src/')\n",
107 |     "import generate_tree"
108 |    ]
109 |   },
110 |   {
111 |    "cell_type": "markdown",
112 |    "metadata": {},
113 |    "source": [
114 |     "#### generate simple tree"
115 |    ]
116 |   },
117 |   {
118 |    "cell_type": "code",
119 |    "execution_count": 6,
120 |    "metadata": {},
121 |    "outputs": [
122 |     {
123 |      "name": "stdout",
124 |      "output_type": "stream",
125 |      "text": [
126 |       "The output is in simple_tree_output/simple_tree_Titanic_Tree.html. Enjoy!\n",
127 |       "CPU times: user 11.8 ms, sys: 0 ns, total: 11.8 ms\n",
128 |       "Wall time: 11.4 ms\n"
129 |      ]
130 |     }
131 |    ],
132 |    "source": [
133 |     "%%time\n",
134 |     "generate_tree.generate_simple_tree(tree_title='Titanic_Tree', tree_model=dt, X=titanic[features], \n",
135 |     "                                   target_names=['Not Survived', 'Survived'], target_colors = None,\n",
136 |     "                                   color_map=None, width=1500, height=1000)"
137 |    ]
138 |   },
139 |   {
140 |    "cell_type": "markdown",
141 |    "metadata": {},
142 |    "source": [
143 |     "#### generate sankey tree"
144 |    ]
145 |   },
146 |   {
147 |    "cell_type": "code",
148 |    "execution_count": 7,
149 |    "metadata": {},
150 |    "outputs": [
151 |     {
152 |      "name": "stdout",
153 |      "output_type": "stream",
154 |      "text": [
155 |       "The output is in sankey_tree_output/sankey_tree_Titanic_Tree.html. Enjoy!\n",
156 |       "CPU times: user 28.6 ms, sys: 5.19 ms, total: 33.8 ms\n",
157 |       "Wall time: 31.1 ms\n"
158 |      ]
159 |     }
160 |    ],
161 |    "source": [
162 |     "%%time\n",
163 |     "generate_tree.generate_sankey_tree(tree_title='Titanic_Tree', tree_model=dt, X=titanic[features], \n",
164 |     "                                   target_names=['Not Survived', 'Survived'], target_colors = None,\n",
165 |     "                                   color_map=None, width=1500, height=1200)"
166 |    ]
167 |   },
168 |   {
169 |    "cell_type": "markdown",
170 |    "metadata": {},
171 |    "source": [
172 |     "## Iris: multiclass classifier"
173 |    ]
174 |   },
175 |   {
176 |    "cell_type": "code",
177 |    "execution_count": 8,
178 |    "metadata": {},
179 |    "outputs": [],
180 |    "source": [
181 |     "from sklearn.datasets import load_iris\n",
182 |     "import pandas as pd\n",
183 |     "\n",
184 |     "iris = load_iris()\n",
185 |     "clf = DecisionTreeClassifier()\n",
186 |     "clf = clf.fit(iris.data, iris.target)"
187 |    ]
188 |   },
189 |   {
190 |    "cell_type": "markdown",
191 |    "metadata": {},
192 |    "source": [
193 |     "#### generate simple tree"
194 |    ]
195 |   },
196 |   {
197 |    "cell_type": "code",
198 |    "execution_count": 10,
199 |    "metadata": {},
200 |    "outputs": [
201 |     {
202 |      "name": "stdout",
203 |      "output_type": "stream",
204 |      "text": [
205 |       "The output is in simple_tree_output/simple_tree_Iris_Tree.html. Enjoy!\n",
206 |       "CPU times: user 13.8 ms, sys: 6.28 ms, total: 20.1 ms\n",
207 |       "Wall time: 17.5 ms\n"
208 |      ]
209 |     }
210 |    ],
211 |    "source": [
212 |     "%%time\n",
213 |     "generate_tree.generate_simple_tree(tree_title='Iris_Tree', tree_model=clf, \n",
214 |     "                                   X=pd.DataFrame(iris.data, columns=iris.feature_names), \n",
215 |     "                                   target_names=list(iris.target_names), target_colors = None,\n",
216 |     "                                   color_map='Vega10', width=1200, height=1000)"
217 |    ]
218 |   },
219 |   {
220 |    "cell_type": "markdown",
221 |    "metadata": {},
222 |    "source": [
223 |     "#### generate sankey tree"
224 |    ]
225 |   },
226 |   {
227 |    "cell_type": "code",
228 |    "execution_count": 11,
229 |    "metadata": {},
230 |    "outputs": [
231 |     {
232 |      "name": "stdout",
233 |      "output_type": "stream",
234 |      "text": [
235 |       "The output is in sankey_tree_output/sankey_tree_Iris_Tree.html. Enjoy!\n",
236 |       "CPU times: user 22.1 ms, sys: 2.52 ms, total: 24.7 ms\n",
237 |       "Wall time: 21.8 ms\n"
238 |      ]
239 |     }
240 |    ],
241 |    "source": [
242 |     "%%time\n",
243 |     "generate_tree.generate_sankey_tree(tree_title='Iris_Tree', tree_model=clf, \n",
244 |     "                                   X=pd.DataFrame(iris.data, columns=iris.feature_names), \n",
245 |     "                                   target_names=list(iris.target_names), target_colors = None,\n",
246 |     "                                   color_map='Vega10', width=1200, height=1000)"
247 |    ]
248 |   }
249 |  ],
250 |  "metadata": {
251 |   "anaconda-cloud": {},
252 |   "kernelspec": {
253 |    "display_name": "Python 2",
254 |    "language": "python",
255 |    "name": "python2"
256 |   },
257 |   "language_info": {
258 |    "codemirror_mode": {
259 |     "name": "ipython",
260 |     "version": 2
261 |    },
262 |    "file_extension": ".py",
263 |    "mimetype": "text/x-python",
264 |    "name": "python",
265 |    "nbconvert_exporter": "python",
266 |    "pygments_lexer": "ipython2",
267 |    "version": "2.7.14"
268 |   }
269 |  },
270 |  "nbformat": 4,
271 |  "nbformat_minor": 1
272 | }
273 | 


--------------------------------------------------------------------------------
/tree/image/sankey_tree.gif:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/sosuneko/Nuance/e8b486ae7459850a00f1e8bbd756e7a57aed4417/tree/image/sankey_tree.gif


--------------------------------------------------------------------------------
/tree/image/simple_tree.gif:
--------------------------------------------------------------------------------
https://raw.githubusercontent.com/sosuneko/Nuance/e8b486ae7459850a00f1e8bbd756e7a57aed4417/tree/image/simple_tree.gif


--------------------------------------------------------------------------------
/tree/sankey_tree_output/sankey_tree_Iris_Tree.html:
--------------------------------------------------------------------------------
  1 | <!DOCTYPE html>
  2 | <html lang='en'>
  3 | <head>
  4 | <meta charset='utf-8'>
  5 | <style>
  6 | .node {
  7 | 	cursor: pointer;
  8 | }
  9 | 
 10 | .node text {
 11 | 	font: 12px helvetica;
 12 | }
 13 | 
 14 | .link {
 15 | 	fill: #eee;
 16 | 	opacity: 0.8;
 17 | }
 18 | 
 19 | </style>
 20 | </head>
 21 | 
 22 | <body>
 23 | <script src="https://d3js.org/d3.v3.min.js"></script>
 24 | <script>
 25 | 
 26 | var margin = {top: 50, right: 50, bottom: 150, left: 50};
 27 | 
 28 | // collect all inputs
 29 | var tree_depth = 5;
 30 | var rules = {"0": [], "1": ["petal width (cm) <= 0.8"], "2": ["petal width (cm) > 0.8"], "3": ["0.8 < petal width (cm) <= 1.75"], "4": ["petal length (cm) <= 4.95", "0.8 < petal width (cm) <= 1.75"], "5": ["petal length (cm) <= 4.95", "0.8 < petal width (cm) <= 1.65"], "6": ["petal length (cm) <= 4.95", "1.65 < petal width (cm) <= 1.75"], "7": ["petal length (cm) > 4.95", "0.8 < petal width (cm) <= 1.75"], "8": ["petal length (cm) > 4.95", "0.8 < petal width (cm) <= 1.55"], "9": ["petal length (cm) > 4.95", "1.55 < petal width (cm) <= 1.75"], "10": ["1.55 < petal width (cm) <= 1.75", "4.95 < petal length (cm) <= 5.45"], "11": ["petal length (cm) > 5.45", "1.55 < petal width (cm) <= 1.75"], "12": ["petal width (cm) > 1.75"], "13": ["petal width (cm) > 1.75", "petal length (cm) <= 4.85"], "14": ["petal width (cm) > 1.75", "petal length (cm) <= 4.85", "sepal length (cm) <= 5.95"], "15": ["petal width (cm) > 1.75", "sepal length (cm) > 5.95", "petal length (cm) <= 4.85"], "16": ["petal width (cm) > 1.75", "petal length (cm) > 4.85"]};
 31 | var num_node = 17;
 32 | var width = 1200, height = 1000;
 33 | var max_samples = 150;
 34 | var min_samples = 1;
 35 | var target_colors = ['#ff7f0e', '#2ca02c', '#d62728'];
 36 | var root = {"name": "Iris_Tree", "parent": "null", "predict": "setosa", "self": 0, "pos": "null", "value": [50, 50, 50], "sample": 150, "color": "#ff7f0e", "children": [{"name": "petal width (cm) <= 0.8", "parent": 0, "predict": "setosa", "self": 1, "pos": "left", "value": [50, 0, 0], "sample": 50, "color": "#ff7f0e", "impurity": 0.0}, {"name": "petal width (cm) > 0.8", "parent": 0, "predict": "versicolor", "self": 2, "pos": "right", "value": [0, 50, 50], "sample": 100, "color": "#2ca02c", "children": [{"name": "petal width (cm) <= 1.75", "parent": 2, "predict": "versicolor", "self": 3, "pos": "left", "value": [0, 49, 5], "sample": 54, "color": "#2ca02c", "children": [{"name": "petal length (cm) <= 4.95", "parent": 3, "predict": "versicolor", "self": 4, "pos": "left", "value": [0, 47, 1], "sample": 48, "color": "#2ca02c", "children": [{"name": "petal width (cm) <= 1.65", "parent": 4, "predict": "versicolor", "self": 5, "pos": "left", "value": [0, 47, 0], "sample": 47, "color": "#2ca02c", "impurity": 0.0}, {"name": "petal width (cm) > 1.65", "parent": 4, "predict": "virginica", "self": 6, "pos": "right", "value": [0, 0, 1], "sample": 1, "color": "#d62728", "impurity": 0.0}], "impurity": 0.041}, {"name": "petal length (cm) > 4.95", "parent": 3, "predict": "virginica", "self": 7, "pos": "right", "value": [0, 2, 4], "sample": 6, "color": "#d62728", "children": [{"name": "petal width (cm) <= 1.55", "parent": 7, "predict": "virginica", "self": 8, "pos": "left", "value": [0, 0, 3], "sample": 3, "color": "#d62728", "impurity": 0.0}, {"name": "petal width (cm) > 1.55", "parent": 7, "predict": "versicolor", "self": 9, "pos": "right", "value": [0, 2, 1], "sample": 3, "color": "#2ca02c", "children": [{"name": "petal length (cm) <= 5.45", "parent": 9, "predict": "versicolor", "self": 10, "pos": "left", "value": [0, 2, 0], "sample": 2, "color": "#2ca02c", "impurity": 0.0}, {"name": "petal length (cm) > 5.45", "parent": 9, "predict": "virginica", "self": 11, "pos": "right", "value": [0, 0, 1], "sample": 1, "color": "#d62728", "impurity": 0.0}], "impurity": 0.444}], "impurity": 0.444}], "impurity": 0.168}, {"name": "petal width (cm) > 1.75", "parent": 2, "predict": "virginica", "self": 12, "pos": "right", "value": [0, 1, 45], "sample": 46, "color": "#d62728", "children": [{"name": "petal length (cm) <= 4.85", "parent": 12, "predict": "virginica", "self": 13, "pos": "left", "value": [0, 1, 2], "sample": 3, "color": "#d62728", "children": [{"name": "sepal length (cm) <= 5.95", "parent": 13, "predict": "versicolor", "self": 14, "pos": "left", "value": [0, 1, 0], "sample": 1, "color": "#2ca02c", "impurity": 0.0}, {"name": "sepal length (cm) > 5.95", "parent": 13, "predict": "virginica", "self": 15, "pos": "right", "value": [0, 0, 2], "sample": 2, "color": "#d62728", "impurity": 0.0}], "impurity": 0.444}, {"name": "petal length (cm) > 4.85", "parent": 12, "predict": "virginica", "self": 16, "pos": "right", "value": [0, 0, 43], "sample": 43, "color": "#d62728", "impurity": 0.0}], "impurity": 0.043}], "impurity": 0.5}], "impurity": 0.667};
 37 | 
 38 | // calculate tree gap
 39 | var tree_gap = (width - margin['left'] - margin['right']) * 1.0 / tree_depth;
 40 | 
 41 | // node height scale based on number of samples
 42 | var h_node = d3.scale.linear()
 43 |     .domain([parseInt(min_samples), parseInt(max_samples)])
 44 |     .clamp(true)
 45 |     .range([10, 350]);
 46 | 
 47 | // generate tree layout (horizonal)
 48 | var tree = d3.layout.tree().size([height, width])
 49 |                 .separation(function(a, b) {
 50 |                     var height = h_node(a.sample) + h_node(b.sample),
 51 |                     // horizontal distance between nodes = 16
 52 |                     distance = height / 2 + 16; 
 53 |                     return distance;
 54 |                 });
 55 | var diagonal = d3.svg.diagonal()
 56 |         .projection(function(d){
 57 |             return [d.y, d.x];
 58 |         });
 59 | 
 60 | var svg = d3.select('body').append('svg')
 61 |         .attr('width', width + margin.right + margin.left)
 62 |         .attr('height', height + margin.top + margin.bottom)
 63 |         .append('g')
 64 |         .attr('transform', 'translate(' + margin.left + ',' + margin.top + ')');
 65 | 
 66 | var i = 0, duration = 750;
 67 | var rect_width = 15;
 68 | // height scale for each tree node
 69 | var y = d3.scale.linear();
 70 | 
 71 | root.x0 = height / 2;
 72 | root.y0 = 0;
 73 | // update tree
 74 | update(root);
 75 | 
 76 | 
 77 | function update(source) {
 78 | 
 79 |     // compute the tree layout
 80 |     var nodes = tree.nodes(root).reverse(),
 81 |     links = tree.links(nodes);
 82 | 
 83 |     // normalize for fixed-depth
 84 |     nodes.forEach(function(d) {
 85 |     	d.y = d.depth * tree_gap;
 86 |     });
 87 | 
 88 |     // draw the nodes
 89 |     var node = svg.selectAll('g.node')
 90 |     		.data(nodes, function(d) {
 91 |     			return d.id || (d.id = ++i);
 92 |     		});
 93 | 
 94 |     // enter any new nodes at the parent's previous position
 95 |     var nodeEnter = node.enter().append('g')
 96 |     		.attr('class', 'node')
 97 |     		.on('click', click)
 98 |     		.attr('transform', function(d) {
 99 |     			return 'translate(' + source.y0 + ',' + source.x0 + ')';
100 |     		})
101 |     		.on('mouseover', mouseover)
102 |     		.on('mouseout', mouseout);
103 | 
104 |     // draw tree nodes as a stacked bar
105 |     nodeEnter.append("g").selectAll("rect")
106 |             .data(function(d) { 
107 |                 var y0 = 0;
108 |                 var node_samples = d.sample;
109 |                 y.rangeRound([h_node(node_samples), 0]).domain([0, node_samples]);
110 |                 d.values_info = d3.keys(target_colors).map(function(i, target_name){
111 |                     // y0 will update 
112 |                     return {target_name: target_name, y0: y(y0), y1: y(y0 += d.value[i])}
113 |                 })
114 |                 return d.values_info; 
115 |               })
116 |             .enter().append("rect")
117 |                 .attr("width", rect_width)
118 |                 .attr("y", function(d) { return d.y1; })
119 |                 .attr("height", function(d) { return d.y0 - d.y1; })
120 |                 .style("fill", function(d) { 
121 |                     return target_colors[d.target_name];
122 |                 });
123 | 
124 |     // add text on tree nodes
125 |     var text_gap = 15;
126 |     var text = nodeEnter.append('text')
127 |     		.attr('x', 18)
128 |     		.attr('y', function(d){
129 |                 return h_node(d.sample) / 2 - text_gap + 3;             
130 |     		});
131 | 
132 |     text.append('tspan')
133 |     		.style('font-weight', 'bold')
134 |     		.text(function(d) {
135 |                 return d.name;
136 |     		});
137 |     
138 |     text.append('tspan')
139 |             .attr('x', 18)
140 |             .attr('y', function(d){
141 |                 return h_node(d.sample) / 2 + 3;
142 |             })
143 |             .style('font-size', '11px')
144 |             .style('fill', '#555')
145 |             .text(function(d) {
146 |                 return 'Values: [' + d.value + ']';
147 |             });
148 | 
149 |     text.append('tspan')
150 |             .attr('x', 18)
151 |             .attr('y', function(d){
152 |                 return h_node(d.sample) / 2 + text_gap + 3;
153 |             })
154 |             .style('font-size', '11px')
155 |             .style('fill', '#555')
156 |             .text(function(d) {
157 |                 return 'Predict: ' + d.predict;
158 |             });
159 | 
160 |     nodeEnter.select('text').style('fill-opacity', 0); 
161 | 
162 |     var nodeUpdate = node.transition()
163 |     		.duration(duration)
164 |     		.attr('transform', function(d) {
165 |                 var height = h_node(+d.sample);                
166 |     			var width = rect_width;
167 |     			return 'translate(' + (d.y - width / 2) + ',' + (d.x - height / 2) + ')';
168 |     		});
169 | 
170 |     nodeUpdate.select('text').style('fill-opacity', 1);
171 | 
172 |     // transition exiting nodes to the parent's new position
173 |     var nodeExit = node.exit().transition()
174 |     		.duration(duration)
175 |     		.attr('transform', function(d) {
176 |     			return 'translate(' + source.y + ',' + source.x + ')';
177 |     		})
178 |     		.remove();
179 | 
180 |     // hide nodes and texts
181 |     nodeExit.select('rect').attr('height', 1e-6);
182 |     nodeExit.select('text').style('fill-opacity', 0);
183 | 
184 |     // draw the links
185 |     var link = svg.selectAll('path.link')
186 |     		.data(links, function(d) {
187 |     			return d.target.id;
188 |     		});
189 | 
190 |     // enter any new links at the parent's previous position
191 |     link.enter().insert('path', 'g')
192 |     		.attr('class', 'link')
193 |     		.attr('d', function(d) {
194 |                 var o = {x: source.x0, y : source.y0};
195 |                 d_new = {source: {x: o.y, y:o.x, sample:d.source.sample}, target: {x: o.y, y: o.x, sample:d.target.sample}};
196 |                 return sankey_link(d_new);
197 |     		})
198 |             .style('stroke', '#eee');
199 | 
200 |     // transition links to their new position
201 |     link.transition()
202 |     		.duration(duration)
203 |             .attr('d', function(d){
204 |                 var source_h = h_node(d.source.sample);
205 |                 var target_h = h_node(d.target.sample);
206 |                 // center the starting points 
207 |                 if(d.target.pos=="left") {
208 |                     var o = {x: d.source.x - (source_h - target_h) / 2., y : d.source.y};
209 |                 }else{
210 |                     var o = {x: d.source.x + (source_h - target_h) / 2., y : d.source.y};
211 |                 }
212 |                 d_new = {source: {x: o.y, y:o.x, sample:d.source.sample}, 
213 |                          target: {x: d.target.y, y: d.target.x, sample:d.target.sample}};
214 |                 return sankey_link(d_new);
215 |             });
216 | 
217 |     // transition the existing links to the parent's new position
218 |     link.exit().transition()
219 |     		.duration(duration)
220 |     		.attr('d', function(d) {
221 |     			var o = {x: source.x, y: source.y};
222 |                 d_new = {source: {x: o.y, y:o.x, sample:d.source.sample}, 
223 |                          target: {x: o.y, y: o.x, sample:d.target.sample}};
224 |                 return sankey_link(d_new);
225 |     		}).remove();
226 | 
227 |     // stash the old positions for transition
228 |     nodes.forEach(function(d) {
229 |     	d.x0 = d.x;
230 |     	d.y0 = d.y;
231 |     });
232 | }
233 | 
234 | function sankey_link(d_new) {
235 |     var source_h = h_node(d_new.source.sample);
236 |     var target_h = h_node(d_new.target.sample);
237 | 
238 |     var curvature = .5;
239 |     var x0 = d_new.source.x,
240 |         x1 = d_new.target.x,
241 |         xi = d3.interpolateNumber(x0, x1),
242 |         x2 = xi(curvature),
243 |         x3 = xi(1 - curvature),
244 |         y0 = d_new.source.y - target_h / 2.,
245 |         y1 = d_new.target.y - target_h / 2.;
246 |     return "M" + x0 + "," + y0
247 |          + "C" + x2 + "," + y0
248 |          + " " + x3 + "," + y1
249 |          + " " + x1 + "," + y1
250 |          + "L" + x1 + "," + (y1+target_h)
251 |          + "C" + x3 + "," + (y1+target_h)
252 |          + " " + x2 + "," + (y0+target_h)
253 |          + " " + x0 + "," + (y0+target_h)
254 |          + "L" + x0 + "," + y0;
255 | }
256 | 
257 | // recursively color the links when hovering on a tree node
258 | function recurse(d, color) {
259 | 	d3.selectAll('.link').each(function(l) {
260 | 		if(l.target === d) {
261 | 			d3.select(this).style('fill', color);
262 | 		}
263 | 	});
264 | 	if(d.parent) {
265 | 		recurse(d.parent, color);
266 | 	}
267 | }
268 | 
269 | function mouseover(d) {
270 |     // if this node has no children nodes
271 | 	if(!d.children) {
272 |         var rule = rules[d.self];
273 |         var append = ['Predict: ' + d.predict, 'Impurity: ' + d.impurity, ''];
274 |         rule = append.concat(rule);
275 | 
276 |         var len = rule.length;
277 |         var info = d3.select(this).append('g').attr('class', 'hover');
278 |         var y0 = h_node(d.sample) + 20;
279 |         var rules_text = info.append('g').append('text').attr('y', y0);
280 | 
281 |         for(var i = 1; i <= len; i++) {
282 |             rules_text.append('tspan')
283 |                     .attr('x', 0)
284 |                     .attr('y', y0 + i * 15)
285 |                     .attr('text-anchor', 'left')
286 |                     .style('font-weight', function(){
287 |                         if(i <= 3) {
288 |                             return 'bold';
289 |                         }else{
290 |                             return null;
291 |                         }
292 |                     })
293 |                     .style('fill', function(){
294 |                         if(i <= 3) {
295 |                             return '#000';
296 |                         }else{
297 |                             return d.color;
298 |                         }
299 |                     })
300 |                     .text(function() {
301 |                         return rule[i - 1];
302 |                     });
303 |         }
304 | 
305 |         // draw the box
306 |         var bbox = info.node().getBBox();
307 |         var padding = 10;
308 |         var rect = d3.select(this).insert('rect', 'g')
309 |                 .attr('class', 'info_bg')
310 |                 .attr('x', bbox.x - padding)
311 |                 .attr('y', bbox.y - padding)
312 |                 .attr('width', bbox.width + padding * 2)
313 |                 .attr('height', bbox.height + padding * 2)
314 |                 .style('fill', 'white')
315 |                 .style('stroke', d.color)
316 |                 .style('stroke-width', 2);
317 | 
318 |         // make the rule infomation on the first layer
319 |         this.parentNode.appendChild(this);      
320 | 		recurse(d, d.color);
321 | 	
322 | 	}
323 | }
324 | 
325 | function mouseout(d) {
326 | 	d3.select('.hover').remove();
327 | 	d3.select('.info_bg').remove();
328 | 
329 | 	d3.select(this).style('stroke', function(d) {
330 | 		if(d._children){
331 | 			var stroke_color = '#000';
332 | 		}else{
333 | 			var stroke_color = null;
334 | 		}
335 | 		d3.select(this).selectAll('rect').style('stroke-width', 2).style('stroke', stroke_color);
336 | 		return null;
337 | 	});
338 | 
339 |     d3.selectAll('.link').style('fill', '#eee');
340 | }
341 | 
342 | function click(d) {
343 | 	if(d.children) {
344 | 		d._children = d.children;
345 | 		d.children = null;
346 | 	}else {
347 | 		d.children = d._children;
348 | 		d._children = null;
349 | 	}
350 | 	update(d);
351 | }
352 | 
353 | </script>
354 | </body>
355 | </html>


--------------------------------------------------------------------------------
/tree/sankey_tree_output/sankey_tree_Titanic_Tree.html:
--------------------------------------------------------------------------------
  1 | <!DOCTYPE html>
  2 | <html lang='en'>
  3 | <head>
  4 | <meta charset='utf-8'>
  5 | <style>
  6 | .node {
  7 | 	cursor: pointer;
  8 | }
  9 | 
 10 | .node text {
 11 | 	font: 12px helvetica;
 12 | }
 13 | 
 14 | .link {
 15 | 	fill: #eee;
 16 | 	opacity: 0.8;
 17 | }
 18 | 
 19 | </style>
 20 | </head>
 21 | 
 22 | <body>
 23 | <script src="https://d3js.org/d3.v3.min.js"></script>
 24 | <script>
 25 | 
 26 | var margin = {top: 50, right: 50, bottom: 150, left: 50};
 27 | 
 28 | // collect all inputs
 29 | var tree_depth = 8;
 30 | var rules = {"0": [], "1": ["Sex: 0"], "2": ["Sex: 1"], "3": ["Sex: 0", "Pclass <= 2"], "4": ["Sex: 0", "Pclass > 2"], "5": ["Sex: 1", "Age <= 6.5"], "6": ["Sex: 1", "Age > 6.5"], "7": ["Sex: 1", "Age > 6.5", "Pclass <= 1"], "8": ["Sex: 1", "Age > 6.5", "Pclass > 1"], "9": ["Sex: 0", "Pclass > 2", "Fare <= 23.35"], "10": ["Sex: 0", "Pclass > 2", "Fare > 23.35"], "11": ["Sex: 1", "Age <= 6.5", "SibSp <= 2"], "12": ["Sex: 1", "SibSp > 2", "Age <= 6.5"], "13": ["Sex: 0", "Pclass > 2", "Age <= 36.5", "Fare <= 23.35"], "14": ["Sex: 0", "Age > 36.5", "Pclass > 2", "Fare <= 23.35"], "15": ["Sex: 1", "Age > 6.5", "Pclass <= 1", "Fare <= 26.144"], "16": ["Sex: 1", "Age > 6.5", "Pclass <= 1", "Fare > 26.144"], "17": ["Sex: 1", "Pclass <= 1", "Fare > 26.144", "6.5 < Age <= 53.0"], "18": ["Sex: 1", "Age > 53.0", "Pclass <= 1", "Fare > 26.144"], "19": ["Sex: 1", "Pclass <= 1", "6.5 < Age <= 53.0", "26.144 < Fare <= 27.135"], "20": ["Sex: 1", "Pclass <= 1", "Fare > 27.135", "6.5 < Age <= 53.0"], "21": ["Sex: 0", "Pclass > 2", "Age <= 36.5", "Fare <= 23.35", "Embarked_S: 0"], "22": ["Sex: 0", "Pclass > 2", "Age <= 36.5", "Fare <= 23.35", "Embarked_S: 1"], "23": ["Sex: 0", "Pclass > 2", "Age <= 29.75", "Fare <= 23.35", "Embarked_S: 0"], "24": ["Sex: 0", "Pclass > 2", "Fare <= 23.35", "Embarked_S: 0", "29.75 < Age <= 36.5"], "25": ["Sex: 0", "Pclass > 2", "Age <= 29.75", "Embarked_S: 0", "Fare <= 15.373"], "26": ["Sex: 0", "Pclass > 2", "Age <= 29.75", "Embarked_S: 0", "15.373 < Fare <= 23.35"], "27": ["Sex: 0", "Pclass > 2", "Age <= 29.75", "Embarked_S: 0", "Fare <= 13.935"], "28": ["Sex: 0", "Pclass > 2", "Age <= 29.75", "Embarked_S: 0", "13.935 < Fare <= 15.373"], "29": ["Sex: 0", "Pclass > 2", "Fare <= 23.35", "36.5 < Age <= 55.0"], "30": ["Sex: 0", "Age > 55.0", "Pclass > 2", "Fare <= 23.35"], "31": ["Sex: 0", "Parch <= 0", "Pclass > 2", "Fare > 23.35"], "32": ["Sex: 0", "Parch > 0", "Pclass > 2", "Fare > 23.35"], "33": ["Sex: 0", "Pclass > 2", "Age <= 36.5", "Embarked_S: 1", "Fare <= 10.825"], "34": ["Sex: 0", "Pclass > 2", "Age <= 36.5", "Embarked_S: 1", "10.825 < Fare <= 23.35"], "35": ["Sex: 0", "Pclass > 2", "Age <= 36.5", "Embarked_S: 1", "10.825 < Fare <= 17.6"], "36": ["Sex: 0", "Pclass > 2", "Age <= 36.5", "Embarked_S: 1", "17.6 < Fare <= 23.35"], "37": ["Sex: 0", "Pclass > 2", "Age <= 36.5", "Fare <= 7.762", "Embarked_S: 1"], "38": ["Sex: 0", "Pclass > 2", "Age <= 36.5", "Embarked_S: 1", "7.762 < Fare <= 10.825"]};
 31 | var num_node = 39;
 32 | var width = 1500, height = 1200;
 33 | var max_samples = 891;
 34 | var min_samples = 1;
 35 | var target_colors = ['#aec7e8', '#ff7f0e'];
 36 | var root = {"name": "Titanic_Tree", "parent": "null", "predict": "Not Survived", "self": 0, "pos": "null", "value": [549, 342], "sample": 891, "color": "#aec7e8", "children": [{"name": "Sex: 0", "parent": 0, "predict": "Survived", "self": 1, "pos": "left", "value": [81, 233], "sample": 314, "color": "#ff7f0e", "children": [{"name": "Pclass <= 2", "parent": 1, "predict": "Survived", "self": 3, "pos": "left", "value": [9, 161], "sample": 170, "color": "#ff7f0e", "impurity": 0.1}, {"name": "Pclass > 2", "parent": 1, "predict": "Not Survived", "self": 4, "pos": "right", "value": [72, 72], "sample": 144, "color": "#aec7e8", "children": [{"name": "Fare <= 23.35", "parent": 4, "predict": "Survived", "self": 9, "pos": "left", "value": [48, 69], "sample": 117, "color": "#ff7f0e", "children": [{"name": "Age <= 36.5", "parent": 9, "predict": "Survived", "self": 13, "pos": "left", "value": [42, 68], "sample": 110, "color": "#ff7f0e", "children": [{"name": "Embarked_S: 0", "parent": 13, "predict": "Survived", "self": 21, "pos": "left", "value": [15, 38], "sample": 53, "color": "#ff7f0e", "children": [{"name": "Age <= 29.75", "parent": 21, "predict": "Survived", "self": 23, "pos": "left", "value": [13, 38], "sample": 51, "color": "#ff7f0e", "children": [{"name": "Fare <= 15.373", "parent": 23, "predict": "Survived", "self": 25, "pos": "left", "value": [13, 27], "sample": 40, "color": "#ff7f0e", "children": [{"name": "Fare <= 13.935", "parent": 25, "predict": "Survived", "self": 27, "pos": "left", "value": [6, 25], "sample": 31, "color": "#ff7f0e", "impurity": 0.312}, {"name": "Fare > 13.935", "parent": 25, "predict": "Not Survived", "self": 28, "pos": "right", "value": [7, 2], "sample": 9, "color": "#aec7e8", "impurity": 0.346}], "impurity": 0.439}, {"name": "Fare > 15.373", "parent": 23, "predict": "Survived", "self": 26, "pos": "right", "value": [0, 11], "sample": 11, "color": "#ff7f0e", "impurity": 0.0}], "impurity": 0.38}, {"name": "Age > 29.75", "parent": 21, "predict": "Not Survived", "self": 24, "pos": "right", "value": [2, 0], "sample": 2, "color": "#aec7e8", "impurity": 0.0}], "impurity": 0.406}, {"name": "Embarked_S: 1", "parent": 13, "predict": "Survived", "self": 22, "pos": "right", "value": [27, 30], "sample": 57, "color": "#ff7f0e", "children": [{"name": "Fare <= 10.825", "parent": 22, "predict": "Not Survived", "self": 33, "pos": "left", "value": [19, 14], "sample": 33, "color": "#aec7e8", "children": [{"name": "Fare <= 7.762", "parent": 33, "predict": "Survived", "self": 37, "pos": "left", "value": [1, 5], "sample": 6, "color": "#ff7f0e", "impurity": 0.278}, {"name": "Fare > 7.762", "parent": 33, "predict": "Not Survived", "self": 38, "pos": "right", "value": [18, 9], "sample": 27, "color": "#aec7e8", "impurity": 0.444}], "impurity": 0.489}, {"name": "Fare > 10.825", "parent": 22, "predict": "Survived", "self": 34, "pos": "right", "value": [8, 16], "sample": 24, "color": "#ff7f0e", "children": [{"name": "Fare <= 17.6", "parent": 34, "predict": "Survived", "self": 35, "pos": "left", "value": [2, 13], "sample": 15, "color": "#ff7f0e", "impurity": 0.231}, {"name": "Fare > 17.6", "parent": 34, "predict": "Not Survived", "self": 36, "pos": "right", "value": [6, 3], "sample": 9, "color": "#aec7e8", "impurity": 0.444}], "impurity": 0.444}], "impurity": 0.499}], "impurity": 0.472}, {"name": "Age > 36.5", "parent": 9, "predict": "Not Survived", "self": 14, "pos": "right", "value": [6, 1], "sample": 7, "color": "#aec7e8", "children": [{"name": "Age <= 55.0", "parent": 14, "predict": "Not Survived", "self": 29, "pos": "left", "value": [6, 0], "sample": 6, "color": "#aec7e8", "impurity": 0.0}, {"name": "Age > 55.0", "parent": 14, "predict": "Survived", "self": 30, "pos": "right", "value": [0, 1], "sample": 1, "color": "#ff7f0e", "impurity": 0.0}], "impurity": 0.245}], "impurity": 0.484}, {"name": "Fare > 23.35", "parent": 4, "predict": "Not Survived", "self": 10, "pos": "right", "value": [24, 3], "sample": 27, "color": "#aec7e8", "children": [{"name": "Parch <= 0", "parent": 10, "predict": "Survived", "self": 31, "pos": "left", "value": [0, 1], "sample": 1, "color": "#ff7f0e", "impurity": 0.0}, {"name": "Parch > 0", "parent": 10, "predict": "Not Survived", "self": 32, "pos": "right", "value": [24, 2], "sample": 26, "color": "#aec7e8", "impurity": 0.142}], "impurity": 0.198}], "impurity": 0.5}], "impurity": 0.383}, {"name": "Sex: 1", "parent": 0, "predict": "Not Survived", "self": 2, "pos": "right", "value": [468, 109], "sample": 577, "color": "#aec7e8", "children": [{"name": "Age <= 6.5", "parent": 2, "predict": "Survived", "self": 5, "pos": "left", "value": [8, 16], "sample": 24, "color": "#ff7f0e", "children": [{"name": "SibSp <= 2", "parent": 5, "predict": "Survived", "self": 11, "pos": "left", "value": [0, 15], "sample": 15, "color": "#ff7f0e", "impurity": 0.0}, {"name": "SibSp > 2", "parent": 5, "predict": "Not Survived", "self": 12, "pos": "right", "value": [8, 1], "sample": 9, "color": "#aec7e8", "impurity": 0.198}], "impurity": 0.444}, {"name": "Age > 6.5", "parent": 2, "predict": "Not Survived", "self": 6, "pos": "right", "value": [460, 93], "sample": 553, "color": "#aec7e8", "children": [{"name": "Pclass <= 1", "parent": 6, "predict": "Not Survived", "self": 7, "pos": "left", "value": [77, 43], "sample": 120, "color": "#aec7e8", "children": [{"name": "Fare <= 26.144", "parent": 7, "predict": "Not Survived", "self": 15, "pos": "left", "value": [10, 0], "sample": 10, "color": "#aec7e8", "impurity": 0.0}, {"name": "Fare > 26.144", "parent": 7, "predict": "Not Survived", "self": 16, "pos": "right", "value": [67, 43], "sample": 110, "color": "#aec7e8", "children": [{"name": "Age <= 53.0", "parent": 16, "predict": "Not Survived", "self": 17, "pos": "left", "value": [49, 40], "sample": 89, "color": "#aec7e8", "children": [{"name": "Fare <= 27.135", "parent": 17, "predict": "Survived", "self": 19, "pos": "left", "value": [2, 11], "sample": 13, "color": "#ff7f0e", "impurity": 0.26}, {"name": "Fare > 27.135", "parent": 17, "predict": "Not Survived", "self": 20, "pos": "right", "value": [47, 29], "sample": 76, "color": "#aec7e8", "impurity": 0.472}], "impurity": 0.495}, {"name": "Age > 53.0", "parent": 16, "predict": "Not Survived", "self": 18, "pos": "right", "value": [18, 3], "sample": 21, "color": "#aec7e8", "impurity": 0.245}], "impurity": 0.476}], "impurity": 0.46}, {"name": "Pclass > 1", "parent": 6, "predict": "Not Survived", "self": 8, "pos": "right", "value": [383, 50], "sample": 433, "color": "#aec7e8", "impurity": 0.204}], "impurity": 0.28}], "impurity": 0.306}], "impurity": 0.473};
 37 | 
 38 | // calculate tree gap
 39 | var tree_gap = (width - margin['left'] - margin['right']) * 1.0 / tree_depth;
 40 | 
 41 | // node height scale based on number of samples
 42 | var h_node = d3.scale.linear()
 43 |     .domain([parseInt(min_samples), parseInt(max_samples)])
 44 |     .clamp(true)
 45 |     .range([10, 350]);
 46 | 
 47 | // generate tree layout (horizonal)
 48 | var tree = d3.layout.tree().size([height, width])
 49 |                 .separation(function(a, b) {
 50 |                     var height = h_node(a.sample) + h_node(b.sample),
 51 |                     // horizontal distance between nodes = 16
 52 |                     distance = height / 2 + 16; 
 53 |                     return distance;
 54 |                 });
 55 | var diagonal = d3.svg.diagonal()
 56 |         .projection(function(d){
 57 |             return [d.y, d.x];
 58 |         });
 59 | 
 60 | var svg = d3.select('body').append('svg')
 61 |         .attr('width', width + margin.right + margin.left)
 62 |         .attr('height', height + margin.top + margin.bottom)
 63 |         .append('g')
 64 |         .attr('transform', 'translate(' + margin.left + ',' + margin.top + ')');
 65 | 
 66 | var i = 0, duration = 750;
 67 | var rect_width = 15;
 68 | // height scale for each tree node
 69 | var y = d3.scale.linear();
 70 | 
 71 | root.x0 = height / 2;
 72 | root.y0 = 0;
 73 | // update tree
 74 | update(root);
 75 | 
 76 | 
 77 | function update(source) {
 78 | 
 79 |     // compute the tree layout
 80 |     var nodes = tree.nodes(root).reverse(),
 81 |     links = tree.links(nodes);
 82 | 
 83 |     // normalize for fixed-depth
 84 |     nodes.forEach(function(d) {
 85 |     	d.y = d.depth * tree_gap;
 86 |     });
 87 | 
 88 |     // draw the nodes
 89 |     var node = svg.selectAll('g.node')
 90 |     		.data(nodes, function(d) {
 91 |     			return d.id || (d.id = ++i);
 92 |     		});
 93 | 
 94 |     // enter any new nodes at the parent's previous position
 95 |     var nodeEnter = node.enter().append('g')
 96 |     		.attr('class', 'node')
 97 |     		.on('click', click)
 98 |     		.attr('transform', function(d) {
 99 |     			return 'translate(' + source.y0 + ',' + source.x0 + ')';
100 |     		})
101 |     		.on('mouseover', mouseover)
102 |     		.on('mouseout', mouseout);
103 | 
104 |     // draw tree nodes as a stacked bar
105 |     nodeEnter.append("g").selectAll("rect")
106 |             .data(function(d) { 
107 |                 var y0 = 0;
108 |                 var node_samples = d.sample;
109 |                 y.rangeRound([h_node(node_samples), 0]).domain([0, node_samples]);
110 |                 d.values_info = d3.keys(target_colors).map(function(i, target_name){
111 |                     // y0 will update 
112 |                     return {target_name: target_name, y0: y(y0), y1: y(y0 += d.value[i])}
113 |                 })
114 |                 return d.values_info; 
115 |               })
116 |             .enter().append("rect")
117 |                 .attr("width", rect_width)
118 |                 .attr("y", function(d) { return d.y1; })
119 |                 .attr("height", function(d) { return d.y0 - d.y1; })
120 |                 .style("fill", function(d) { 
121 |                     return target_colors[d.target_name];
122 |                 });
123 | 
124 |     // add text on tree nodes
125 |     var text_gap = 15;
126 |     var text = nodeEnter.append('text')
127 |     		.attr('x', 18)
128 |     		.attr('y', function(d){
129 |                 return h_node(d.sample) / 2 - text_gap + 3;             
130 |     		});
131 | 
132 |     text.append('tspan')
133 |     		.style('font-weight', 'bold')
134 |     		.text(function(d) {
135 |                 return d.name;
136 |     		});
137 |     
138 |     text.append('tspan')
139 |             .attr('x', 18)
140 |             .attr('y', function(d){
141 |                 return h_node(d.sample) / 2 + 3;
142 |             })
143 |             .style('font-size', '11px')
144 |             .style('fill', '#555')
145 |             .text(function(d) {
146 |                 return 'Values: [' + d.value + ']';
147 |             });
148 | 
149 |     text.append('tspan')
150 |             .attr('x', 18)
151 |             .attr('y', function(d){
152 |                 return h_node(d.sample) / 2 + text_gap + 3;
153 |             })
154 |             .style('font-size', '11px')
155 |             .style('fill', '#555')
156 |             .text(function(d) {
157 |                 return 'Predict: ' + d.predict;
158 |             });
159 | 
160 |     nodeEnter.select('text').style('fill-opacity', 0); 
161 | 
162 |     var nodeUpdate = node.transition()
163 |     		.duration(duration)
164 |     		.attr('transform', function(d) {
165 |                 var height = h_node(+d.sample);                
166 |     			var width = rect_width;
167 |     			return 'translate(' + (d.y - width / 2) + ',' + (d.x - height / 2) + ')';
168 |     		});
169 | 
170 |     nodeUpdate.select('text').style('fill-opacity', 1);
171 | 
172 |     // transition exiting nodes to the parent's new position
173 |     var nodeExit = node.exit().transition()
174 |     		.duration(duration)
175 |     		.attr('transform', function(d) {
176 |     			return 'translate(' + source.y + ',' + source.x + ')';
177 |     		})
178 |     		.remove();
179 | 
180 |     // hide nodes and texts
181 |     nodeExit.select('rect').attr('height', 1e-6);
182 |     nodeExit.select('text').style('fill-opacity', 0);
183 | 
184 |     // draw the links
185 |     var link = svg.selectAll('path.link')
186 |     		.data(links, function(d) {
187 |     			return d.target.id;
188 |     		});
189 | 
190 |     // enter any new links at the parent's previous position
191 |     link.enter().insert('path', 'g')
192 |     		.attr('class', 'link')
193 |     		.attr('d', function(d) {
194 |                 var o = {x: source.x0, y : source.y0};
195 |                 d_new = {source: {x: o.y, y:o.x, sample:d.source.sample}, target: {x: o.y, y: o.x, sample:d.target.sample}};
196 |                 return sankey_link(d_new);
197 |     		})
198 |             .style('stroke', '#eee');
199 | 
200 |     // transition links to their new position
201 |     link.transition()
202 |     		.duration(duration)
203 |             .attr('d', function(d){
204 |                 var source_h = h_node(d.source.sample);
205 |                 var target_h = h_node(d.target.sample);
206 |                 // center the starting points 
207 |                 if(d.target.pos=="left") {
208 |                     var o = {x: d.source.x - (source_h - target_h) / 2., y : d.source.y};
209 |                 }else{
210 |                     var o = {x: d.source.x + (source_h - target_h) / 2., y : d.source.y};
211 |                 }
212 |                 d_new = {source: {x: o.y, y:o.x, sample:d.source.sample}, 
213 |                          target: {x: d.target.y, y: d.target.x, sample:d.target.sample}};
214 |                 return sankey_link(d_new);
215 |             });
216 | 
217 |     // transition the existing links to the parent's new position
218 |     link.exit().transition()
219 |     		.duration(duration)
220 |     		.attr('d', function(d) {
221 |     			var o = {x: source.x, y: source.y};
222 |                 d_new = {source: {x: o.y, y:o.x, sample:d.source.sample}, 
223 |                          target: {x: o.y, y: o.x, sample:d.target.sample}};
224 |                 return sankey_link(d_new);
225 |     		}).remove();
226 | 
227 |     // stash the old positions for transition
228 |     nodes.forEach(function(d) {
229 |     	d.x0 = d.x;
230 |     	d.y0 = d.y;
231 |     });
232 | }
233 | 
234 | function sankey_link(d_new) {
235 |     var source_h = h_node(d_new.source.sample);
236 |     var target_h = h_node(d_new.target.sample);
237 | 
238 |     var curvature = .5;
239 |     var x0 = d_new.source.x,
240 |         x1 = d_new.target.x,
241 |         xi = d3.interpolateNumber(x0, x1),
242 |         x2 = xi(curvature),
243 |         x3 = xi(1 - curvature),
244 |         y0 = d_new.source.y - target_h / 2.,
245 |         y1 = d_new.target.y - target_h / 2.;
246 |     return "M" + x0 + "," + y0
247 |          + "C" + x2 + "," + y0
248 |          + " " + x3 + "," + y1
249 |          + " " + x1 + "," + y1
250 |          + "L" + x1 + "," + (y1+target_h)
251 |          + "C" + x3 + "," + (y1+target_h)
252 |          + " " + x2 + "," + (y0+target_h)
253 |          + " " + x0 + "," + (y0+target_h)
254 |          + "L" + x0 + "," + y0;
255 | }
256 | 
257 | // recursively color the links when hovering on a tree node
258 | function recurse(d, color) {
259 | 	d3.selectAll('.link').each(function(l) {
260 | 		if(l.target === d) {
261 | 			d3.select(this).style('fill', color);
262 | 		}
263 | 	});
264 | 	if(d.parent) {
265 | 		recurse(d.parent, color);
266 | 	}
267 | }
268 | 
269 | function mouseover(d) {
270 |     // if this node has no children nodes
271 | 	if(!d.children) {
272 |         var rule = rules[d.self];
273 |         var append = ['Predict: ' + d.predict, 'Impurity: ' + d.impurity, ''];
274 |         rule = append.concat(rule);
275 | 
276 |         var len = rule.length;
277 |         var info = d3.select(this).append('g').attr('class', 'hover');
278 |         var y0 = h_node(d.sample) + 20;
279 |         var rules_text = info.append('g').append('text').attr('y', y0);
280 | 
281 |         for(var i = 1; i <= len; i++) {
282 |             rules_text.append('tspan')
283 |                     .attr('x', 0)
284 |                     .attr('y', y0 + i * 15)
285 |                     .attr('text-anchor', 'left')
286 |                     .style('font-weight', function(){
287 |                         if(i <= 3) {
288 |                             return 'bold';
289 |                         }else{
290 |                             return null;
291 |                         }
292 |                     })
293 |                     .style('fill', function(){
294 |                         if(i <= 3) {
295 |                             return '#000';
296 |                         }else{
297 |                             return d.color;
298 |                         }
299 |                     })
300 |                     .text(function() {
301 |                         return rule[i - 1];
302 |                     });
303 |         }
304 | 
305 |         // draw the box
306 |         var bbox = info.node().getBBox();
307 |         var padding = 10;
308 |         var rect = d3.select(this).insert('rect', 'g')
309 |                 .attr('class', 'info_bg')
310 |                 .attr('x', bbox.x - padding)
311 |                 .attr('y', bbox.y - padding)
312 |                 .attr('width', bbox.width + padding * 2)
313 |                 .attr('height', bbox.height + padding * 2)
314 |                 .style('fill', 'white')
315 |                 .style('stroke', d.color)
316 |                 .style('stroke-width', 2);
317 | 
318 |         // make the rule infomation on the first layer
319 |         this.parentNode.appendChild(this);      
320 | 		recurse(d, d.color);
321 | 	
322 | 	}
323 | }
324 | 
325 | function mouseout(d) {
326 | 	d3.select('.hover').remove();
327 | 	d3.select('.info_bg').remove();
328 | 
329 | 	d3.select(this).style('stroke', function(d) {
330 | 		if(d._children){
331 | 			var stroke_color = '#000';
332 | 		}else{
333 | 			var stroke_color = null;
334 | 		}
335 | 		d3.select(this).selectAll('rect').style('stroke-width', 2).style('stroke', stroke_color);
336 | 		return null;
337 | 	});
338 | 
339 |     d3.selectAll('.link').style('fill', '#eee');
340 | }
341 | 
342 | function click(d) {
343 | 	if(d.children) {
344 | 		d._children = d.children;
345 | 		d.children = null;
346 | 	}else {
347 | 		d.children = d._children;
348 | 		d._children = null;
349 | 	}
350 | 	update(d);
351 | }
352 | 
353 | </script>
354 | </body>
355 | </html>


--------------------------------------------------------------------------------
/tree/simple_tree_output/simple_tree_Iris_Tree.html:
--------------------------------------------------------------------------------
  1 | <!DOCTYPE html>
  2 | <html lang='en'>
  3 | <head>
  4 | <meta charset='utf-8'>
  5 | <style>
  6 | .node {
  7 | 	cursor: pointer;
  8 | }
  9 | 
 10 | .node circle {
 11 | 	stroke-width: 3px;
 12 | }
 13 | 
 14 | .node text {
 15 | 	font: 12px helvetica;
 16 | }
 17 | 
 18 | .link {
 19 | 	fill: none;
 20 | 	stroke: #ccc;
 21 | 	stroke-width: 2px;
 22 | }
 23 | 
 24 | </style>
 25 | </head>
 26 | 
 27 | <body>
 28 | <script src="https://d3js.org/d3.v3.min.js"></script>
 29 | <script>
 30 | 
 31 | var margin = {top: 50, right: 100, bottom: 150, left: 50};
 32 | 
 33 | // collect all inputs
 34 | var tree_depth = 5;
 35 | var rules = {"0": [], "1": ["petal width (cm) <= 0.8"], "2": ["petal width (cm) > 0.8"], "3": ["0.8 < petal width (cm) <= 1.75"], "4": ["petal length (cm) <= 4.95", "0.8 < petal width (cm) <= 1.75"], "5": ["petal length (cm) <= 4.95", "0.8 < petal width (cm) <= 1.65"], "6": ["petal length (cm) <= 4.95", "1.65 < petal width (cm) <= 1.75"], "7": ["petal length (cm) > 4.95", "0.8 < petal width (cm) <= 1.75"], "8": ["petal length (cm) > 4.95", "0.8 < petal width (cm) <= 1.55"], "9": ["petal length (cm) > 4.95", "1.55 < petal width (cm) <= 1.75"], "10": ["1.55 < petal width (cm) <= 1.75", "4.95 < petal length (cm) <= 5.45"], "11": ["petal length (cm) > 5.45", "1.55 < petal width (cm) <= 1.75"], "12": ["petal width (cm) > 1.75"], "13": ["petal width (cm) > 1.75", "petal length (cm) <= 4.85"], "14": ["petal width (cm) > 1.75", "petal length (cm) <= 4.85", "sepal length (cm) <= 5.95"], "15": ["petal width (cm) > 1.75", "sepal length (cm) > 5.95", "petal length (cm) <= 4.85"], "16": ["petal width (cm) > 1.75", "petal length (cm) > 4.85"]};
 36 | var num_node = 17;
 37 | var root = {"name": "Iris_Tree", "parent": "null", "predict": "setosa", "self": 0, "pos": "null", "value": [50, 50, 50], "sample": 150, "color": "#ff7f0e", "children": [{"name": "petal width (cm) <= 0.8", "parent": 0, "predict": "setosa", "self": 1, "pos": "left", "value": [50, 0, 0], "sample": 50, "color": "#ff7f0e", "impurity": 0.0}, {"name": "petal width (cm) > 0.8", "parent": 0, "predict": "versicolor", "self": 2, "pos": "right", "value": [0, 50, 50], "sample": 100, "color": "#2ca02c", "children": [{"name": "petal width (cm) <= 1.75", "parent": 2, "predict": "versicolor", "self": 3, "pos": "left", "value": [0, 49, 5], "sample": 54, "color": "#2ca02c", "children": [{"name": "petal length (cm) <= 4.95", "parent": 3, "predict": "versicolor", "self": 4, "pos": "left", "value": [0, 47, 1], "sample": 48, "color": "#2ca02c", "children": [{"name": "petal width (cm) <= 1.65", "parent": 4, "predict": "versicolor", "self": 5, "pos": "left", "value": [0, 47, 0], "sample": 47, "color": "#2ca02c", "impurity": 0.0}, {"name": "petal width (cm) > 1.65", "parent": 4, "predict": "virginica", "self": 6, "pos": "right", "value": [0, 0, 1], "sample": 1, "color": "#d62728", "impurity": 0.0}], "impurity": 0.041}, {"name": "petal length (cm) > 4.95", "parent": 3, "predict": "virginica", "self": 7, "pos": "right", "value": [0, 2, 4], "sample": 6, "color": "#d62728", "children": [{"name": "petal width (cm) <= 1.55", "parent": 7, "predict": "virginica", "self": 8, "pos": "left", "value": [0, 0, 3], "sample": 3, "color": "#d62728", "impurity": 0.0}, {"name": "petal width (cm) > 1.55", "parent": 7, "predict": "versicolor", "self": 9, "pos": "right", "value": [0, 2, 1], "sample": 3, "color": "#2ca02c", "children": [{"name": "petal length (cm) <= 5.45", "parent": 9, "predict": "versicolor", "self": 10, "pos": "left", "value": [0, 2, 0], "sample": 2, "color": "#2ca02c", "impurity": 0.0}, {"name": "petal length (cm) > 5.45", "parent": 9, "predict": "virginica", "self": 11, "pos": "right", "value": [0, 0, 1], "sample": 1, "color": "#d62728", "impurity": 0.0}], "impurity": 0.444}], "impurity": 0.444}], "impurity": 0.168}, {"name": "petal width (cm) > 1.75", "parent": 2, "predict": "virginica", "self": 12, "pos": "right", "value": [0, 1, 45], "sample": 46, "color": "#d62728", "children": [{"name": "petal length (cm) <= 4.85", "parent": 12, "predict": "virginica", "self": 13, "pos": "left", "value": [0, 1, 2], "sample": 3, "color": "#d62728", "children": [{"name": "sepal length (cm) <= 5.95", "parent": 13, "predict": "versicolor", "self": 14, "pos": "left", "value": [0, 1, 0], "sample": 1, "color": "#2ca02c", "impurity": 0.0}, {"name": "sepal length (cm) > 5.95", "parent": 13, "predict": "virginica", "self": 15, "pos": "right", "value": [0, 0, 2], "sample": 2, "color": "#d62728", "impurity": 0.0}], "impurity": 0.444}, {"name": "petal length (cm) > 4.85", "parent": 12, "predict": "virginica", "self": 16, "pos": "right", "value": [0, 0, 43], "sample": 43, "color": "#d62728", "impurity": 0.0}], "impurity": 0.043}], "impurity": 0.5}], "impurity": 0.667};
 38 | var width = 1200, height = 1000;
 39 | var n_classes = 3;
 40 | 
 41 | // calculate tree gap
 42 | var tree_gap = (width - margin['left'] - margin['right']) * 1.0 / tree_depth;
 43 | 
 44 | // generate tree layout (horizonal)
 45 | var tree = d3.layout.tree().size([height, width]);
 46 | var diagonal = d3.svg.diagonal()
 47 |         .projection(function(d){
 48 | 			return [d.y, d.x];
 49 | 		});
 50 | 
 51 | var svg = d3.select('body').append('svg')
 52 | 		.attr('width', width + margin.right + margin.left)
 53 | 		.attr('height', height + margin.top + margin.bottom)
 54 | 		.append('g')
 55 | 		.attr('transform', 'translate(' + margin.left + ',' + margin.top + ')');
 56 | 
 57 | // color scale based on node impurity
 58 | var node_color_scale = d3.scale.linear()
 59 |     .domain([1.0 / n_classes + 0.1, 0.])
 60 |     .clamp(true);
 61 | 
 62 | var i = 0, duration = 750;
 63 | root.x0 = height / 2;
 64 | root.y0 = 0;
 65 | 
 66 | // update tree
 67 | update(root);
 68 | 
 69 | 
 70 | function update(source) {
 71 | 
 72 |     // compute the tree layout
 73 |     var nodes = tree.nodes(root).reverse(),
 74 |     links = tree.links(nodes);
 75 | 
 76 |     // normalize for fixed-depth
 77 |     nodes.forEach(function(d) {
 78 |     	d.y = d.depth * tree_gap;
 79 |     });
 80 | 
 81 |     // draw the nodes
 82 |     var node = svg.selectAll('g.node')
 83 |     		.data(nodes, function(d) {
 84 |     			return d.id || (d.id = ++i);
 85 |     		});
 86 | 
 87 |     // enter any new nodes at the parent's previous position
 88 |     var nodeEnter = node.enter().append('g')
 89 |     		.attr('class', 'node')
 90 |     		.on('click', click)
 91 |     		.attr('transform', function(d) {
 92 |     			return 'translate(' + source.y0 + ',' + source.x0 + ')';
 93 |     		})
 94 |     		.on('click', click)
 95 |     		.on('mouseover', mouseover)
 96 |     		.on('mouseout', mouseout);
 97 | 
 98 |     // draw tree nodes
 99 |     nodeEnter.append('circle')
100 |     		.attr('r', 1e-6)
101 |     		.style('fill', function(d) {
102 |                 // d.color is the target color
103 |     			node_color_scale.range(['#fff', d.color]);
104 |     			return node_color_scale(d.impurity);
105 |     		})
106 |     		.style('stroke', function(d) {
107 |     			return d.color;
108 |     		})
109 |     		.on('mouseover', function(d) {
110 |     			if(!d.children) {
111 |     				d3.select(this).style('fill', function(d){
112 |     					return d.color;
113 |     				});
114 |     				recurse(d, d.color);
115 |     			}
116 |     		})
117 |             // when mouseout, recover everything
118 |     		.on('mouseout', function(d) {
119 |     			d3.select(this).style('fill', function(d) {
120 |     				node_color_scale.range(['#fff', d.color]);
121 |     				return node_color_scale(d.impurity);
122 |     			}).attr('r', function(d){
123 |                     // when the node is collapesd, increase the size
124 |                     return d._children? 20 : 10;
125 |                 });
126 |     			d3.selectAll('.link')
127 |                     .style('stroke-width', '2')
128 |                     .style('stroke', '#ccc');
129 |     		});
130 | 
131 |     // add text on tree nodes
132 |     var text = nodeEnter.append('text')
133 |     		.attr('x', 13)
134 |     		.attr('dy', '-0.5em');
135 | 
136 |     text.append('tspan')
137 |     		.style('font-weight', 'bold')
138 |     		.text(function(d) {
139 |                 return d.name;
140 |     		});
141 | 
142 |     text.append('tspan')
143 |     		.attr('x', 13)
144 |     		.attr('y', 8)
145 |     		.text(function(d) {
146 |     			return 'Values: [' + d.value + ']';
147 |     		});
148 | 
149 |     text.append('tspan')
150 |     		.attr('x', 13)
151 |     		.attr('y', 22)
152 |     		.text(function(d) {
153 |     			return 'Predict: ' + d.predict;
154 |     		});
155 | 
156 |     nodeEnter.select('text').style('fill-opacity', 0);
157 | 
158 |     // transform tree nodes to their initial positions
159 |     var nodeUpdate = node.transition()
160 |     		.duration(duration)
161 |     		.attr('transform', function(d) {
162 |     			return 'translate(' + d.y + ',' + d.x + ')';
163 |     		});
164 | 
165 |     nodeUpdate.select('circle')
166 |     		.attr('r', function(d){
167 |     			return d._children? 20 : 10;
168 |     		})
169 |     		.style('stroke', function(d) {
170 |     			return d.color;
171 |     		})
172 |     		.style('fill', function(d) {
173 |     			node_color_scale.range(['#fff', d.color]);
174 |     			return node_color_scale(d.impurity);
175 |     		});
176 | 
177 |     nodeUpdate.select('text').style('fill-opacity', 1);
178 | 
179 |     // transition exiting nodes to the parent's new position
180 |     var nodeExit = node.exit().transition()
181 |     		.duration(duration)
182 |     		.attr('transform', function(d) {
183 |     			return 'translate(' + source.y + ',' + source.x + ')';
184 |     		})
185 |     		.remove();
186 | 
187 |     // hide nodes and texts
188 |     nodeExit.select('circle').attr('r', 1e-6);
189 |     nodeExit.select('text').style('fill-opacity', 0);
190 | 
191 |     // draw the links
192 |     var link = svg.selectAll('path.link')
193 |     		.data(links, function(d) {
194 |     			return d.target.id;
195 |     		});
196 | 
197 |     // enter any new links at the parent's previous position
198 |     link.enter().insert('path', 'g')
199 |     		.attr('class', 'link')
200 |     		.attr('d', function(d) {
201 |     			var o = {x: source.x0, y : source.y0};
202 |     			return diagonal({source:o, target: o});
203 |     		});
204 | 
205 |     // transition links to their new position
206 |     link.transition()
207 |     		.duration(duration)
208 |     		.attr('d', diagonal);
209 | 
210 |     // transition the existing links to the parent's new position
211 |     link.exit().transition()
212 |     		.duration(duration)
213 |     		.attr('d', function(d) {
214 |     			var o = {x: source.x, y: source.y};
215 |     			return diagonal({source: o, target: o});
216 |     		}).remove();
217 | 
218 |     // stash the old positions for transition
219 |     nodes.forEach(function(d) {
220 |     	d.x0 = d.x;
221 |     	d.y0 = d.y;
222 |     });
223 | }
224 | 
225 | // recursively color the links when hovering on a tree node
226 | function recurse(d, color) {
227 | 	d3.selectAll('.link').each(function(l) {
228 | 		if(l.target === d) {
229 | 			d3.select(this).style('stroke-width', 4).style('stroke', color);
230 | 		}
231 | 	});
232 | 	if(d.parent) {
233 | 		recurse(d.parent, color);
234 | 	}
235 | }
236 | 
237 | function mouseover(d) {
238 |     // if this node has no children nodes
239 | 	if(!d.children) {
240 | 		var rule = rules[d.self];
241 | 		var append = ['Predict: ' + d.predict, 'Impurity: ' + d.impurity, ''];
242 | 		rule = append.concat(rule);
243 | 
244 | 		var len = rule.length;
245 | 		var info = d3.select(this).append('g').attr('class', 'hover');
246 | 		var y0 = 35;
247 | 		var rules_text = info.append('g').append('text').attr('y', y0);
248 | 
249 | 		for(var i = 1; i <= len; i++) {
250 | 			rules_text.append('tspan')
251 | 					.attr('x', 0)
252 | 					.attr('y', y0 + i * 15)
253 | 					.attr('text-anchor', 'left')
254 | 					.style('font-weight', function(){
255 | 						if(i <= 3) {
256 | 							return 'bold';
257 | 						}else{
258 | 							return null;
259 | 						}
260 | 					})
261 | 					.style('fill', function(){
262 | 						if(i <= 3) {
263 | 							return '#000';
264 | 						}else{
265 | 							return d.color;
266 | 						}
267 | 					})
268 | 					.text(function() {
269 | 						return rule[i - 1];
270 | 					});
271 | 		}
272 | 
273 |         // draw the box
274 | 		var bbox = info.node().getBBox();
275 | 		var padding = 10;
276 | 		var rect = d3.select(this).insert('rect', 'g')
277 | 				.attr('class', 'info_bg')
278 | 				.attr('x', bbox.x - padding)
279 | 				.attr('y', bbox.y - padding)
280 | 				.attr('width', bbox.width + padding * 2)
281 | 				.attr('height', bbox.height + padding * 2)
282 | 				.style('fill', 'white')
283 | 				.style('stroke', d.color)
284 | 				.style('stroke-width', 2);
285 | 
286 | 		// make the rule infomation on the first layer
287 | 		this.parentNode.appendChild(this);		
288 | 	}
289 | }
290 | 
291 | function mouseout(d) {
292 | 	d3.select('.hover').remove();
293 | 	d3.select('.info_bg').remove();
294 | }
295 | 
296 | function click(d) {
297 | 	if(d.children) {
298 | 		d._children = d.children;
299 | 		d.children = null;
300 | 	}else {
301 | 		d.children = d._children;
302 | 		d._children = null;
303 | 	}
304 | 	update(d);
305 | }
306 | 
307 | </script>
308 | </body>
309 | </html>


--------------------------------------------------------------------------------
/tree/simple_tree_output/simple_tree_Titanic_Tree.html:
--------------------------------------------------------------------------------
  1 | <!DOCTYPE html>
  2 | <html lang='en'>
  3 | <head>
  4 | <meta charset='utf-8'>
  5 | <style>
  6 | .node {
  7 | 	cursor: pointer;
  8 | }
  9 | 
 10 | .node circle {
 11 | 	stroke-width: 3px;
 12 | }
 13 | 
 14 | .node text {
 15 | 	font: 12px helvetica;
 16 | }
 17 | 
 18 | .link {
 19 | 	fill: none;
 20 | 	stroke: #ccc;
 21 | 	stroke-width: 2px;
 22 | }
 23 | 
 24 | </style>
 25 | </head>
 26 | 
 27 | <body>
 28 | <script src="https://d3js.org/d3.v3.min.js"></script>
 29 | <script>
 30 | 
 31 | var margin = {top: 50, right: 100, bottom: 150, left: 50};
 32 | 
 33 | // collect all inputs
 34 | var tree_depth = 8;
 35 | var rules = {"0": [], "1": ["Sex: 0"], "2": ["Sex: 1"], "3": ["Sex: 0", "Pclass <= 2"], "4": ["Sex: 0", "Pclass > 2"], "5": ["Sex: 1", "Age <= 6.5"], "6": ["Sex: 1", "Age > 6.5"], "7": ["Sex: 1", "Age > 6.5", "Pclass <= 1"], "8": ["Sex: 1", "Age > 6.5", "Pclass > 1"], "9": ["Sex: 0", "Pclass > 2", "Fare <= 23.35"], "10": ["Sex: 0", "Pclass > 2", "Fare > 23.35"], "11": ["Sex: 1", "Age <= 6.5", "SibSp <= 2"], "12": ["Sex: 1", "SibSp > 2", "Age <= 6.5"], "13": ["Sex: 0", "Pclass > 2", "Age <= 36.5", "Fare <= 23.35"], "14": ["Sex: 0", "Age > 36.5", "Pclass > 2", "Fare <= 23.35"], "15": ["Sex: 1", "Age > 6.5", "Pclass <= 1", "Fare <= 26.144"], "16": ["Sex: 1", "Age > 6.5", "Pclass <= 1", "Fare > 26.144"], "17": ["Sex: 1", "Pclass <= 1", "Fare > 26.144", "6.5 < Age <= 53.0"], "18": ["Sex: 1", "Age > 53.0", "Pclass <= 1", "Fare > 26.144"], "19": ["Sex: 1", "Pclass <= 1", "6.5 < Age <= 53.0", "26.144 < Fare <= 27.135"], "20": ["Sex: 1", "Pclass <= 1", "Fare > 27.135", "6.5 < Age <= 53.0"], "21": ["Sex: 0", "Pclass > 2", "Age <= 36.5", "Fare <= 23.35", "Embarked_S: 0"], "22": ["Sex: 0", "Pclass > 2", "Age <= 36.5", "Fare <= 23.35", "Embarked_S: 1"], "23": ["Sex: 0", "Pclass > 2", "Age <= 29.75", "Fare <= 23.35", "Embarked_S: 0"], "24": ["Sex: 0", "Pclass > 2", "Fare <= 23.35", "Embarked_S: 0", "29.75 < Age <= 36.5"], "25": ["Sex: 0", "Pclass > 2", "Age <= 29.75", "Embarked_S: 0", "Fare <= 15.373"], "26": ["Sex: 0", "Pclass > 2", "Age <= 29.75", "Embarked_S: 0", "15.373 < Fare <= 23.35"], "27": ["Sex: 0", "Pclass > 2", "Age <= 29.75", "Embarked_S: 0", "Fare <= 13.935"], "28": ["Sex: 0", "Pclass > 2", "Age <= 29.75", "Embarked_S: 0", "13.935 < Fare <= 15.373"], "29": ["Sex: 0", "Pclass > 2", "Fare <= 23.35", "36.5 < Age <= 55.0"], "30": ["Sex: 0", "Age > 55.0", "Pclass > 2", "Fare <= 23.35"], "31": ["Sex: 0", "Parch <= 0", "Pclass > 2", "Fare > 23.35"], "32": ["Sex: 0", "Parch > 0", "Pclass > 2", "Fare > 23.35"], "33": ["Sex: 0", "Pclass > 2", "Age <= 36.5", "Embarked_S: 1", "Fare <= 10.825"], "34": ["Sex: 0", "Pclass > 2", "Age <= 36.5", "Embarked_S: 1", "10.825 < Fare <= 23.35"], "35": ["Sex: 0", "Pclass > 2", "Age <= 36.5", "Embarked_S: 1", "10.825 < Fare <= 17.6"], "36": ["Sex: 0", "Pclass > 2", "Age <= 36.5", "Embarked_S: 1", "17.6 < Fare <= 23.35"], "37": ["Sex: 0", "Pclass > 2", "Age <= 36.5", "Fare <= 7.762", "Embarked_S: 1"], "38": ["Sex: 0", "Pclass > 2", "Age <= 36.5", "Embarked_S: 1", "7.762 < Fare <= 10.825"]};
 36 | var num_node = 39;
 37 | var root = {"name": "Titanic_Tree", "parent": "null", "predict": "Not Survived", "self": 0, "pos": "null", "value": [549, 342], "sample": 891, "color": "#aec7e8", "children": [{"name": "Sex: 0", "parent": 0, "predict": "Survived", "self": 1, "pos": "left", "value": [81, 233], "sample": 314, "color": "#ff7f0e", "children": [{"name": "Pclass <= 2", "parent": 1, "predict": "Survived", "self": 3, "pos": "left", "value": [9, 161], "sample": 170, "color": "#ff7f0e", "impurity": 0.1}, {"name": "Pclass > 2", "parent": 1, "predict": "Not Survived", "self": 4, "pos": "right", "value": [72, 72], "sample": 144, "color": "#aec7e8", "children": [{"name": "Fare <= 23.35", "parent": 4, "predict": "Survived", "self": 9, "pos": "left", "value": [48, 69], "sample": 117, "color": "#ff7f0e", "children": [{"name": "Age <= 36.5", "parent": 9, "predict": "Survived", "self": 13, "pos": "left", "value": [42, 68], "sample": 110, "color": "#ff7f0e", "children": [{"name": "Embarked_S: 0", "parent": 13, "predict": "Survived", "self": 21, "pos": "left", "value": [15, 38], "sample": 53, "color": "#ff7f0e", "children": [{"name": "Age <= 29.75", "parent": 21, "predict": "Survived", "self": 23, "pos": "left", "value": [13, 38], "sample": 51, "color": "#ff7f0e", "children": [{"name": "Fare <= 15.373", "parent": 23, "predict": "Survived", "self": 25, "pos": "left", "value": [13, 27], "sample": 40, "color": "#ff7f0e", "children": [{"name": "Fare <= 13.935", "parent": 25, "predict": "Survived", "self": 27, "pos": "left", "value": [6, 25], "sample": 31, "color": "#ff7f0e", "impurity": 0.312}, {"name": "Fare > 13.935", "parent": 25, "predict": "Not Survived", "self": 28, "pos": "right", "value": [7, 2], "sample": 9, "color": "#aec7e8", "impurity": 0.346}], "impurity": 0.439}, {"name": "Fare > 15.373", "parent": 23, "predict": "Survived", "self": 26, "pos": "right", "value": [0, 11], "sample": 11, "color": "#ff7f0e", "impurity": 0.0}], "impurity": 0.38}, {"name": "Age > 29.75", "parent": 21, "predict": "Not Survived", "self": 24, "pos": "right", "value": [2, 0], "sample": 2, "color": "#aec7e8", "impurity": 0.0}], "impurity": 0.406}, {"name": "Embarked_S: 1", "parent": 13, "predict": "Survived", "self": 22, "pos": "right", "value": [27, 30], "sample": 57, "color": "#ff7f0e", "children": [{"name": "Fare <= 10.825", "parent": 22, "predict": "Not Survived", "self": 33, "pos": "left", "value": [19, 14], "sample": 33, "color": "#aec7e8", "children": [{"name": "Fare <= 7.762", "parent": 33, "predict": "Survived", "self": 37, "pos": "left", "value": [1, 5], "sample": 6, "color": "#ff7f0e", "impurity": 0.278}, {"name": "Fare > 7.762", "parent": 33, "predict": "Not Survived", "self": 38, "pos": "right", "value": [18, 9], "sample": 27, "color": "#aec7e8", "impurity": 0.444}], "impurity": 0.489}, {"name": "Fare > 10.825", "parent": 22, "predict": "Survived", "self": 34, "pos": "right", "value": [8, 16], "sample": 24, "color": "#ff7f0e", "children": [{"name": "Fare <= 17.6", "parent": 34, "predict": "Survived", "self": 35, "pos": "left", "value": [2, 13], "sample": 15, "color": "#ff7f0e", "impurity": 0.231}, {"name": "Fare > 17.6", "parent": 34, "predict": "Not Survived", "self": 36, "pos": "right", "value": [6, 3], "sample": 9, "color": "#aec7e8", "impurity": 0.444}], "impurity": 0.444}], "impurity": 0.499}], "impurity": 0.472}, {"name": "Age > 36.5", "parent": 9, "predict": "Not Survived", "self": 14, "pos": "right", "value": [6, 1], "sample": 7, "color": "#aec7e8", "children": [{"name": "Age <= 55.0", "parent": 14, "predict": "Not Survived", "self": 29, "pos": "left", "value": [6, 0], "sample": 6, "color": "#aec7e8", "impurity": 0.0}, {"name": "Age > 55.0", "parent": 14, "predict": "Survived", "self": 30, "pos": "right", "value": [0, 1], "sample": 1, "color": "#ff7f0e", "impurity": 0.0}], "impurity": 0.245}], "impurity": 0.484}, {"name": "Fare > 23.35", "parent": 4, "predict": "Not Survived", "self": 10, "pos": "right", "value": [24, 3], "sample": 27, "color": "#aec7e8", "children": [{"name": "Parch <= 0", "parent": 10, "predict": "Survived", "self": 31, "pos": "left", "value": [0, 1], "sample": 1, "color": "#ff7f0e", "impurity": 0.0}, {"name": "Parch > 0", "parent": 10, "predict": "Not Survived", "self": 32, "pos": "right", "value": [24, 2], "sample": 26, "color": "#aec7e8", "impurity": 0.142}], "impurity": 0.198}], "impurity": 0.5}], "impurity": 0.383}, {"name": "Sex: 1", "parent": 0, "predict": "Not Survived", "self": 2, "pos": "right", "value": [468, 109], "sample": 577, "color": "#aec7e8", "children": [{"name": "Age <= 6.5", "parent": 2, "predict": "Survived", "self": 5, "pos": "left", "value": [8, 16], "sample": 24, "color": "#ff7f0e", "children": [{"name": "SibSp <= 2", "parent": 5, "predict": "Survived", "self": 11, "pos": "left", "value": [0, 15], "sample": 15, "color": "#ff7f0e", "impurity": 0.0}, {"name": "SibSp > 2", "parent": 5, "predict": "Not Survived", "self": 12, "pos": "right", "value": [8, 1], "sample": 9, "color": "#aec7e8", "impurity": 0.198}], "impurity": 0.444}, {"name": "Age > 6.5", "parent": 2, "predict": "Not Survived", "self": 6, "pos": "right", "value": [460, 93], "sample": 553, "color": "#aec7e8", "children": [{"name": "Pclass <= 1", "parent": 6, "predict": "Not Survived", "self": 7, "pos": "left", "value": [77, 43], "sample": 120, "color": "#aec7e8", "children": [{"name": "Fare <= 26.144", "parent": 7, "predict": "Not Survived", "self": 15, "pos": "left", "value": [10, 0], "sample": 10, "color": "#aec7e8", "impurity": 0.0}, {"name": "Fare > 26.144", "parent": 7, "predict": "Not Survived", "self": 16, "pos": "right", "value": [67, 43], "sample": 110, "color": "#aec7e8", "children": [{"name": "Age <= 53.0", "parent": 16, "predict": "Not Survived", "self": 17, "pos": "left", "value": [49, 40], "sample": 89, "color": "#aec7e8", "children": [{"name": "Fare <= 27.135", "parent": 17, "predict": "Survived", "self": 19, "pos": "left", "value": [2, 11], "sample": 13, "color": "#ff7f0e", "impurity": 0.26}, {"name": "Fare > 27.135", "parent": 17, "predict": "Not Survived", "self": 20, "pos": "right", "value": [47, 29], "sample": 76, "color": "#aec7e8", "impurity": 0.472}], "impurity": 0.495}, {"name": "Age > 53.0", "parent": 16, "predict": "Not Survived", "self": 18, "pos": "right", "value": [18, 3], "sample": 21, "color": "#aec7e8", "impurity": 0.245}], "impurity": 0.476}], "impurity": 0.46}, {"name": "Pclass > 1", "parent": 6, "predict": "Not Survived", "self": 8, "pos": "right", "value": [383, 50], "sample": 433, "color": "#aec7e8", "impurity": 0.204}], "impurity": 0.28}], "impurity": 0.306}], "impurity": 0.473};
 38 | var width = 1500, height = 1000;
 39 | var n_classes = 2;
 40 | 
 41 | // calculate tree gap
 42 | var tree_gap = (width - margin['left'] - margin['right']) * 1.0 / tree_depth;
 43 | 
 44 | // generate tree layout (horizonal)
 45 | var tree = d3.layout.tree().size([height, width]);
 46 | var diagonal = d3.svg.diagonal()
 47 |         .projection(function(d){
 48 | 			return [d.y, d.x];
 49 | 		});
 50 | 
 51 | var svg = d3.select('body').append('svg')
 52 | 		.attr('width', width + margin.right + margin.left)
 53 | 		.attr('height', height + margin.top + margin.bottom)
 54 | 		.append('g')
 55 | 		.attr('transform', 'translate(' + margin.left + ',' + margin.top + ')');
 56 | 
 57 | // color scale based on node impurity
 58 | var node_color_scale = d3.scale.linear()
 59 |     .domain([1.0 / n_classes + 0.1, 0.])
 60 |     .clamp(true);
 61 | 
 62 | var i = 0, duration = 750;
 63 | root.x0 = height / 2;
 64 | root.y0 = 0;
 65 | 
 66 | // update tree
 67 | update(root);
 68 | 
 69 | 
 70 | function update(source) {
 71 | 
 72 |     // compute the tree layout
 73 |     var nodes = tree.nodes(root).reverse(),
 74 |     links = tree.links(nodes);
 75 | 
 76 |     // normalize for fixed-depth
 77 |     nodes.forEach(function(d) {
 78 |     	d.y = d.depth * tree_gap;
 79 |     });
 80 | 
 81 |     // draw the nodes
 82 |     var node = svg.selectAll('g.node')
 83 |     		.data(nodes, function(d) {
 84 |     			return d.id || (d.id = ++i);
 85 |     		});
 86 | 
 87 |     // enter any new nodes at the parent's previous position
 88 |     var nodeEnter = node.enter().append('g')
 89 |     		.attr('class', 'node')
 90 |     		.on('click', click)
 91 |     		.attr('transform', function(d) {
 92 |     			return 'translate(' + source.y0 + ',' + source.x0 + ')';
 93 |     		})
 94 |     		.on('click', click)
 95 |     		.on('mouseover', mouseover)
 96 |     		.on('mouseout', mouseout);
 97 | 
 98 |     // draw tree nodes
 99 |     nodeEnter.append('circle')
100 |     		.attr('r', 1e-6)
101 |     		.style('fill', function(d) {
102 |                 // d.color is the target color
103 |     			node_color_scale.range(['#fff', d.color]);
104 |     			return node_color_scale(d.impurity);
105 |     		})
106 |     		.style('stroke', function(d) {
107 |     			return d.color;
108 |     		})
109 |     		.on('mouseover', function(d) {
110 |     			if(!d.children) {
111 |     				d3.select(this).style('fill', function(d){
112 |     					return d.color;
113 |     				});
114 |     				recurse(d, d.color);
115 |     			}
116 |     		})
117 |             // when mouseout, recover everything
118 |     		.on('mouseout', function(d) {
119 |     			d3.select(this).style('fill', function(d) {
120 |     				node_color_scale.range(['#fff', d.color]);
121 |     				return node_color_scale(d.impurity);
122 |     			}).attr('r', function(d){
123 |                     // when the node is collapesd, increase the size
124 |                     return d._children? 20 : 10;
125 |                 });
126 |     			d3.selectAll('.link')
127 |                     .style('stroke-width', '2')
128 |                     .style('stroke', '#ccc');
129 |     		});
130 | 
131 |     // add text on tree nodes
132 |     var text = nodeEnter.append('text')
133 |     		.attr('x', 13)
134 |     		.attr('dy', '-0.5em');
135 | 
136 |     text.append('tspan')
137 |     		.style('font-weight', 'bold')
138 |     		.text(function(d) {
139 |                 return d.name;
140 |     		});
141 | 
142 |     text.append('tspan')
143 |     		.attr('x', 13)
144 |     		.attr('y', 8)
145 |     		.text(function(d) {
146 |     			return 'Values: [' + d.value + ']';
147 |     		});
148 | 
149 |     text.append('tspan')
150 |     		.attr('x', 13)
151 |     		.attr('y', 22)
152 |     		.text(function(d) {
153 |     			return 'Predict: ' + d.predict;
154 |     		});
155 | 
156 |     nodeEnter.select('text').style('fill-opacity', 0);
157 | 
158 |     // transform tree nodes to their initial positions
159 |     var nodeUpdate = node.transition()
160 |     		.duration(duration)
161 |     		.attr('transform', function(d) {
162 |     			return 'translate(' + d.y + ',' + d.x + ')';
163 |     		});
164 | 
165 |     nodeUpdate.select('circle')
166 |     		.attr('r', function(d){
167 |     			return d._children? 20 : 10;
168 |     		})
169 |     		.style('stroke', function(d) {
170 |     			return d.color;
171 |     		})
172 |     		.style('fill', function(d) {
173 |     			node_color_scale.range(['#fff', d.color]);
174 |     			return node_color_scale(d.impurity);
175 |     		});
176 | 
177 |     nodeUpdate.select('text').style('fill-opacity', 1);
178 | 
179 |     // transition exiting nodes to the parent's new position
180 |     var nodeExit = node.exit().transition()
181 |     		.duration(duration)
182 |     		.attr('transform', function(d) {
183 |     			return 'translate(' + source.y + ',' + source.x + ')';
184 |     		})
185 |     		.remove();
186 | 
187 |     // hide nodes and texts
188 |     nodeExit.select('circle').attr('r', 1e-6);
189 |     nodeExit.select('text').style('fill-opacity', 0);
190 | 
191 |     // draw the links
192 |     var link = svg.selectAll('path.link')
193 |     		.data(links, function(d) {
194 |     			return d.target.id;
195 |     		});
196 | 
197 |     // enter any new links at the parent's previous position
198 |     link.enter().insert('path', 'g')
199 |     		.attr('class', 'link')
200 |     		.attr('d', function(d) {
201 |     			var o = {x: source.x0, y : source.y0};
202 |     			return diagonal({source:o, target: o});
203 |     		});
204 | 
205 |     // transition links to their new position
206 |     link.transition()
207 |     		.duration(duration)
208 |     		.attr('d', diagonal);
209 | 
210 |     // transition the existing links to the parent's new position
211 |     link.exit().transition()
212 |     		.duration(duration)
213 |     		.attr('d', function(d) {
214 |     			var o = {x: source.x, y: source.y};
215 |     			return diagonal({source: o, target: o});
216 |     		}).remove();
217 | 
218 |     // stash the old positions for transition
219 |     nodes.forEach(function(d) {
220 |     	d.x0 = d.x;
221 |     	d.y0 = d.y;
222 |     });
223 | }
224 | 
225 | // recursively color the links when hovering on a tree node
226 | function recurse(d, color) {
227 | 	d3.selectAll('.link').each(function(l) {
228 | 		if(l.target === d) {
229 | 			d3.select(this).style('stroke-width', 4).style('stroke', color);
230 | 		}
231 | 	});
232 | 	if(d.parent) {
233 | 		recurse(d.parent, color);
234 | 	}
235 | }
236 | 
237 | function mouseover(d) {
238 |     // if this node has no children nodes
239 | 	if(!d.children) {
240 | 		var rule = rules[d.self];
241 | 		var append = ['Predict: ' + d.predict, 'Impurity: ' + d.impurity, ''];
242 | 		rule = append.concat(rule);
243 | 
244 | 		var len = rule.length;
245 | 		var info = d3.select(this).append('g').attr('class', 'hover');
246 | 		var y0 = 35;
247 | 		var rules_text = info.append('g').append('text').attr('y', y0);
248 | 
249 | 		for(var i = 1; i <= len; i++) {
250 | 			rules_text.append('tspan')
251 | 					.attr('x', 0)
252 | 					.attr('y', y0 + i * 15)
253 | 					.attr('text-anchor', 'left')
254 | 					.style('font-weight', function(){
255 | 						if(i <= 3) {
256 | 							return 'bold';
257 | 						}else{
258 | 							return null;
259 | 						}
260 | 					})
261 | 					.style('fill', function(){
262 | 						if(i <= 3) {
263 | 							return '#000';
264 | 						}else{
265 | 							return d.color;
266 | 						}
267 | 					})
268 | 					.text(function() {
269 | 						return rule[i - 1];
270 | 					});
271 | 		}
272 | 
273 |         // draw the box
274 | 		var bbox = info.node().getBBox();
275 | 		var padding = 10;
276 | 		var rect = d3.select(this).insert('rect', 'g')
277 | 				.attr('class', 'info_bg')
278 | 				.attr('x', bbox.x - padding)
279 | 				.attr('y', bbox.y - padding)
280 | 				.attr('width', bbox.width + padding * 2)
281 | 				.attr('height', bbox.height + padding * 2)
282 | 				.style('fill', 'white')
283 | 				.style('stroke', d.color)
284 | 				.style('stroke-width', 2);
285 | 
286 | 		// make the rule infomation on the first layer
287 | 		this.parentNode.appendChild(this);		
288 | 	}
289 | }
290 | 
291 | function mouseout(d) {
292 | 	d3.select('.hover').remove();
293 | 	d3.select('.info_bg').remove();
294 | }
295 | 
296 | function click(d) {
297 | 	if(d.children) {
298 | 		d._children = d.children;
299 | 		d.children = null;
300 | 	}else {
301 | 		d.children = d._children;
302 | 		d._children = null;
303 | 	}
304 | 	update(d);
305 | }
306 | 
307 | </script>
308 | </body>
309 | </html>


--------------------------------------------------------------------------------
/tree/src/generate_tree.py:
--------------------------------------------------------------------------------
  1 | 
  2 | import numpy as np
  3 | 
  4 | import matplotlib.pyplot as plt
  5 | import matplotlib
  6 | 
  7 | import copy
  8 | import sys
  9 | import os
 10 | 
 11 | import json
 12 | import jinja2
 13 | 
 14 | 
 15 | def _get_tree_info(X, tree_model, target_names, target_colors, tree_title, color_map):
 16 | 	'''
 17 | 	get useful information of the tree
 18 | 
 19 | 	:param X: pandas DataFrame
 20 | 		dataset model was fitted on
 21 | 	:param tree_model: a fitted sklearn Decision Tree Classifier
 22 | 	:param target_names: list
 23 | 		list of names for targets
 24 | 	:param target_colors: list, default=None
 25 | 		list of colors for targets
 26 | 	:param tree_title: string
 27 | 		name of the tree
 28 | 	:param color_map: string, default=None
 29 | 		matplotlib color map name, like 'Vega20'
 30 | 	:return:
 31 | 		dictionary of useful information
 32 | 	'''
 33 | 	# classify features into 3 types: binary, float and int
 34 | 	binary_features = []
 35 | 	for col in X.columns.values:
 36 | 		if list(sorted(np.unique(X[col].values))) == [0, 1]:
 37 | 			binary_features.append(col)
 38 | 
 39 | 	int_features = []
 40 | 	for col in list(set(X.columns.values) - set(binary_features)):
 41 | 		if list(X[col].map(int).values) == list(X[col].values):
 42 | 			int_features.append(col)
 43 | 
 44 | 	# get feature names
 45 | 	feature_names = X.columns.values
 46 | 
 47 | 	# check target names
 48 | 	if type(target_names) != list or len(target_names) != tree_model.tree_.n_classes:
 49 | 		raise ValueError("target_names should be a list of length %d." % (tree_model.tree_.n_classes))
 50 | 
 51 | 	# color mapping for targets
 52 | 	if target_colors is None:
 53 | 		if color_map is not None:
 54 | 			cm = plt.get_cmap(color_map)
 55 | 		else:
 56 | 			cm = plt.get_cmap('Vega20')
 57 | 		target_colors = []
 58 | 		for n in range(tree_model.tree_.n_classes):
 59 | 			target_colors.append(str(matplotlib.colors.rgb2hex(cm(n + 1))))
 60 | 
 61 | 	tree_info = {
 62 | 		'tree_model': tree_model,
 63 | 		'features': [feature_names[i] for i in tree_model.tree_.feature],
 64 | 		'tree_title': tree_title,
 65 | 		'binary_features': binary_features,
 66 | 		'int_features': int_features,
 67 | 		'target_names': target_names,
 68 | 		'target_colors': target_colors
 69 | 	}
 70 | 	return tree_info
 71 | 
 72 | 
 73 | def _parse_tree(node_id, parent, pos, tree_info):
 74 | 	'''
 75 | 	parse the tree structure
 76 | 
 77 | 	:param node_id: int
 78 | 		node id
 79 | 	:param parent: int
 80 | 	 	parent node id
 81 | 	:param pos: string
 82 | 		position of the node
 83 | 	:param tree_info: dict
 84 | 		information of the tree model
 85 | 	:return:
 86 | 		complete tree structure
 87 | 	'''
 88 | 	tree_model = tree_info['tree_model']
 89 | 	features = tree_info['features']
 90 | 	tree_title = tree_info['tree_title']
 91 | 	binary_features = tree_info['binary_features']
 92 | 	int_features = tree_info['int_features']
 93 | 	target_names = tree_info['target_names']
 94 | 
 95 | 	node = {}
 96 | 	if parent == 'null':
 97 | 		node['name'] = tree_title
 98 | 	else:
 99 | 		feature = features[parent]
100 | 		if pos == 'left':
101 | 			if feature in binary_features:
102 | 				node['name'] = feature + ': 0'
103 | 			elif feature in int_features:
104 | 				node['name'] = feature + " <= " + str(int(tree_model.tree_.threshold[parent]))
105 | 			else:
106 | 				node['name'] = feature + " <= " + str(round(tree_model.tree_.threshold[parent], 3))
107 | 		else:
108 | 			if feature in binary_features:
109 | 				node['name'] = feature + ': 1'
110 | 			elif feature in int_features:
111 | 				node['name'] = feature + " > " + str(int(tree_model.tree_.threshold[parent]))
112 | 			else:
113 | 				node['name'] = feature + " > " + str(round(tree_model.tree_.threshold[parent], 3))
114 | 	try:
115 | 		node['parent'] = int(parent)
116 | 	except:
117 | 		node['parent'] = parent
118 | 
119 | 	node['self'] = int(node_id)
120 | 	node['sample'] = int(tree_model.tree_.n_node_samples[node_id])
121 | 	node['impurity'] = round(tree_model.tree_.impurity[node_id], 3)
122 | 	node['value'] = [int(v) for v in tree_model.tree_.value[node_id][0]]
123 | 	node['predict'] = target_names[np.argmax(node['value'])]
124 | 	node['color'] = tree_info['target_colors'][np.argmax(node['value'])]
125 | 	node['pos'] = pos
126 | 
127 | 	if tree_model.tree_.children_left[node_id] != -1 or tree_model.tree_.children_right[node_id] != -1:
128 | 		node['children'] = []
129 | 		if tree_model.tree_.children_left[node_id] != -1:
130 | 			child = tree_model.tree_.children_left[node_id]
131 | 			node['children'].append(_parse_tree(child, node_id, 'left', tree_info))
132 | 		if tree_model.tree_.children_right[node_id] != -1:
133 | 			child = tree_model.tree_.children_right[node_id]
134 | 			node['children'].append(_parse_tree(child, node_id, 'right', tree_info))
135 | 	return node
136 | 
137 | 
138 | def _extract_rules(node_id, parent, pos, tree_rules, tree_info):
139 | 	'''
140 | 	extract rules for each tree node
141 | 
142 | 	:param node_id: int
143 | 		tree node id
144 | 	:param parent: int
145 | 		parent node id
146 | 	:param pos: string
147 | 		position of the node
148 | 	:param tree_rules: dict
149 | 		key: node_id, value: rule
150 | 	:param tree_info: dict
151 | 		information of the tree model
152 | 	:return:
153 | 		complete tree_rules
154 | 	'''
155 | 	features = tree_info['features']
156 | 	tree_model = tree_info['tree_model']
157 | 
158 | 	tree_rules[node_id] = {}
159 | 	tree_rules[node_id]['features'] = {}
160 | 
161 | 	if parent != "null":
162 | 		previous = copy.deepcopy(tree_rules[parent]['features'])
163 | 		tree_rules[node_id]['features'] = previous
164 | 		feat = features[parent]
165 | 		thre = tree_model.tree_.threshold[parent]
166 | 		if feat not in previous.keys():
167 | 			tree_rules[node_id]['features'][feat] = [-sys.maxint, sys.maxint]
168 | 		if pos == "left":
169 | 			origin = tree_rules[node_id]['features'][feat][1]
170 | 			tree_rules[node_id]['features'][feat][1] = np.min([thre, origin])
171 | 		if pos == "right":
172 | 			origin = tree_rules[node_id]['features'][feat][0]
173 | 			tree_rules[node_id]['features'][feat][0] = np.max([thre, origin])
174 | 
175 | 	if tree_model.tree_.children_left[node_id] != -1:
176 | 		child = tree_model.tree_.children_left[node_id]
177 | 		_extract_rules(child, node_id, "left", tree_rules, tree_info)
178 | 
179 | 	if tree_model.tree_.children_right[node_id] != -1:
180 | 		child = tree_model.tree_.children_right[node_id]
181 | 		_extract_rules(child, node_id, "right", tree_rules, tree_info)
182 | 
183 | 	return tree_rules
184 | 
185 | 
186 | def _clean_rules(tree_rules, tree_info):
187 | 	'''
188 | 	clean up the rules for each branch
189 | 
190 | 	:param tree_rules: dict
191 | 		key: node_id, value: rule
192 | 	:param tree_info: dict
193 | 		information of the tree model
194 | 	:return:
195 | 		cleaned rules with the sample structure as tree_rules
196 | 	'''
197 | 	tree_rules_clean = {}
198 | 	for key in tree_rules.keys():
199 | 		key = int(key)
200 | 		node = copy.deepcopy(tree_rules[key])
201 | 		rules = []
202 | 		if node['features'].keys():
203 | 			for k in node['features'].keys():
204 | 				feat = node['features'][k]
205 | 				if k in tree_info['binary_features']:
206 | 					if feat[0] == -sys.maxint:
207 | 						rule = k + ': 0'
208 | 					else:
209 | 						rule = k + ': 1'
210 | 				elif k in tree_info['int_features']:
211 | 					if feat[0] == -sys.maxint:
212 | 						rule = k + ' <= ' + str(int(feat[1]))
213 | 					elif feat[1] == sys.maxint:
214 | 						rule = k + ' > ' + str(int(feat[0]))
215 | 					else:
216 | 						rule = str(int(feat[0])) + ' < ' + k + ' <= ' + str(int(feat[1]))
217 | 				else:
218 | 					if feat[0] == -sys.maxint:
219 | 						rule = k + ' <= ' + str(round(feat[1], 3))
220 | 					elif feat[1] == sys.maxint:
221 | 						rule = k + ' > ' + str(round(feat[0], 3))
222 | 					else:
223 | 						rule = str(round(feat[0], 3)) + ' < ' + k + ' <= ' + str(round(feat[1], 3))
224 | 				rules.append(rule)
225 | 			rules = sorted(rules, key= lambda x : len(x))
226 | 		tree_rules_clean[key] = rules
227 | 	return tree_rules_clean
228 | 
229 | 
230 | def generate_simple_tree(tree_title, tree_model, X, target_names,
231 | 						 target_colors=None, color_map=None, width=None, height=None):
232 | 	'''
233 | 	visualize a sklearn Decision Tree Classifier
234 | 
235 | 	:param tree_title: string
236 | 		name of the tree
237 | 	:param tree_model: a fitted sklearn Decision Tree Classifier
238 | 	:param X: pandas DataFrame
239 | 		dataset model was fitted on
240 | 	:param target_names: list
241 | 		list of names for targets
242 | 	:param target_colors: list, default=None
243 | 		list of colors for targets
244 | 	:param color_map: string, default=None
245 | 		matplotlib color map name, like 'Vega20'
246 | 	:param width: int
247 | 		width of the html page
248 | 	:param height: int
249 | 		height of the html page
250 | 	'''
251 | 
252 | 	# get tree information
253 | 	tree_info = _get_tree_info(X, tree_model, target_names, target_colors, tree_title, color_map)
254 | 
255 | 	# get the tree structure
256 | 	final_tree = _parse_tree(0, "null", "null", tree_info)
257 | 
258 | 	# extract tree rules
259 | 	tree_rules = {}
260 | 	tree_rules = _extract_rules(0, "null", "null", tree_rules, tree_info)
261 | 
262 | 	# clean up rules
263 | 	tree_rules_clean = _clean_rules(tree_rules, tree_info)
264 | 
265 | 	# get template
266 | 	temp = open('src/simple_tree_template.html').read()
267 | 	template = jinja2.Template(temp)
268 | 
269 | 	# create the output root if it is not exits
270 | 	if not os.path.exists('simple_tree_output'):
271 | 		os.mkdir('simple_tree_output')
272 | 
273 | 	# generate output html
274 | 	with open('simple_tree_output/simple_tree_%s.html' %(tree_title), 'wb') as fh:
275 | 		render_result = {
276 | 			'tree': json.dumps(final_tree), 'rule': json.dumps(tree_rules_clean),
277 | 			'num_node': tree_info['tree_model'].tree_.capacity,
278 | 			'tree_depth': tree_info['tree_model'].tree_.max_depth,
279 | 			'width': width, 'height': height, 'n_classes': tree_info['tree_model'].n_classes_
280 | 		}
281 | 		fh.write(template.render(render_result))
282 | 		print('The output is in simple_tree_output/simple_tree_%s.html. Enjoy!' %(tree_title))
283 | 
284 | 
285 | def generate_sankey_tree(tree_title, tree_model, X, target_names,
286 | 						 target_colors=None, color_map=None, width=None, height=None):
287 | 	'''
288 | 	visualize a sklearn Decision Tree Classifier
289 | 
290 | 	:param tree_title: string
291 | 		name of the tree
292 | 	:param tree_model: a fitted sklearn Decision Tree Classifier
293 | 	:param X: pandas DataFrame
294 | 		dataset model was fitted on
295 | 	:param target_names: list
296 | 		list of names for targets
297 | 	:param target_colors: list, default=None
298 | 		list of colors for targets
299 | 	:param color_map: string, default=None
300 | 		matplotlib color map name, like 'Vega20'
301 | 	:param width: int
302 | 		width of the html page
303 | 	:param height: int
304 | 		height of the html page
305 | 	'''
306 | 
307 | 	# get tree information
308 | 	tree_info = _get_tree_info(X, tree_model, target_names, target_colors, tree_title, color_map)
309 | 
310 | 	# get the tree structure
311 | 	final_tree = _parse_tree(0, "null", "null", tree_info)
312 | 
313 | 	# extract tree rules
314 | 	tree_rules = {}
315 | 	tree_rules = _extract_rules(0, "null", "null", tree_rules, tree_info)
316 | 
317 | 	# clean up rules
318 | 	tree_rules_clean = _clean_rules(tree_rules, tree_info)
319 | 
320 | 	# get template
321 | 	temp = open('src/sankey_tree_template.html').read()
322 | 	template = jinja2.Template(temp)
323 | 
324 | 	# create the output root if it is not exits
325 | 	if not os.path.exists('sankey_tree_output'):
326 | 		os.mkdir('sankey_tree_output')
327 | 
328 | 	# generate output html
329 | 	with open('sankey_tree_output/sankey_tree_%s.html' %(tree_title), 'wb') as fh:
330 | 		render_result = {
331 | 			'tree': json.dumps(final_tree), 'rule': json.dumps(tree_rules_clean),
332 | 			'num_node': tree_info['tree_model'].tree_.capacity,
333 | 			'tree_depth': tree_info['tree_model'].tree_.max_depth,
334 | 			'width': width, 'height': height, 'target_colors': tree_info['target_colors'], 
335 | 			'max_samples': np.max(tree_info['tree_model'].tree_.n_node_samples), 
336 | 			'min_samples': np.min(tree_info['tree_model'].tree_.n_node_samples), 
337 | 		}
338 | 		fh.write(template.render(render_result))
339 | 		print('The output is in sankey_tree_output/sankey_tree_%s.html. Enjoy!' %(tree_title))
340 | 


--------------------------------------------------------------------------------
/tree/src/sankey_tree_template.html:
--------------------------------------------------------------------------------
  1 | <!DOCTYPE html>
  2 | <html lang='en'>
  3 | <head>
  4 | <meta charset='utf-8'>
  5 | <style>
  6 | .node {
  7 | 	cursor: pointer;
  8 | }
  9 | 
 10 | .node text {
 11 | 	font: 12px helvetica;
 12 | }
 13 | 
 14 | .link {
 15 | 	fill: #eee;
 16 | 	opacity: 0.8;
 17 | }
 18 | 
 19 | </style>
 20 | </head>
 21 | 
 22 | <body>
 23 | <script src="https://d3js.org/d3.v3.min.js"></script>
 24 | <script>
 25 | 
 26 | var margin = {top: 50, right: 50, bottom: 150, left: 50};
 27 | 
 28 | // collect all inputs
 29 | var tree_depth = {{ tree_depth }};
 30 | var rules = {{ rule|safe }};
 31 | var num_node = {{ num_node }};
 32 | var width = {{ width }}, height = {{ height }};
 33 | var max_samples = {{ max_samples }};
 34 | var min_samples = {{ min_samples }};
 35 | var target_colors = {{ target_colors|safe }};
 36 | var root = {{ tree|safe }};
 37 | 
 38 | // calculate tree gap
 39 | var tree_gap = (width - margin['left'] - margin['right']) * 1.0 / tree_depth;
 40 | 
 41 | // node height scale based on number of samples
 42 | var h_node = d3.scale.linear()
 43 |     .domain([parseInt(min_samples), parseInt(max_samples)])
 44 |     .clamp(true)
 45 |     .range([10, 350]);
 46 | 
 47 | // generate tree layout (horizonal)
 48 | var tree = d3.layout.tree().size([height, width])
 49 |                 .separation(function(a, b) {
 50 |                     var height = h_node(a.sample) + h_node(b.sample),
 51 |                     // horizontal distance between nodes = 16
 52 |                     distance = height / 2 + 16; 
 53 |                     return distance;
 54 |                 });
 55 | var diagonal = d3.svg.diagonal()
 56 |         .projection(function(d){
 57 |             return [d.y, d.x];
 58 |         });
 59 | 
 60 | var svg = d3.select('body').append('svg')
 61 |         .attr('width', width + margin.right + margin.left)
 62 |         .attr('height', height + margin.top + margin.bottom)
 63 |         .append('g')
 64 |         .attr('transform', 'translate(' + margin.left + ',' + margin.top + ')');
 65 | 
 66 | var i = 0, duration = 750;
 67 | var rect_width = 15;
 68 | // height scale for each tree node
 69 | var y = d3.scale.linear();
 70 | 
 71 | root.x0 = height / 2;
 72 | root.y0 = 0;
 73 | // update tree
 74 | update(root);
 75 | 
 76 | 
 77 | function update(source) {
 78 | 
 79 |     // compute the tree layout
 80 |     var nodes = tree.nodes(root).reverse(),
 81 |     links = tree.links(nodes);
 82 | 
 83 |     // normalize for fixed-depth
 84 |     nodes.forEach(function(d) {
 85 |     	d.y = d.depth * tree_gap;
 86 |     });
 87 | 
 88 |     // draw the nodes
 89 |     var node = svg.selectAll('g.node')
 90 |     		.data(nodes, function(d) {
 91 |     			return d.id || (d.id = ++i);
 92 |     		});
 93 | 
 94 |     // enter any new nodes at the parent's previous position
 95 |     var nodeEnter = node.enter().append('g')
 96 |     		.attr('class', 'node')
 97 |     		.on('click', click)
 98 |     		.attr('transform', function(d) {
 99 |     			return 'translate(' + source.y0 + ',' + source.x0 + ')';
100 |     		})
101 |     		.on('mouseover', mouseover)
102 |     		.on('mouseout', mouseout);
103 | 
104 |     // draw tree nodes as a stacked bar
105 |     nodeEnter.append("g").selectAll("rect")
106 |             .data(function(d) { 
107 |                 var y0 = 0;
108 |                 var node_samples = d.sample;
109 |                 y.rangeRound([h_node(node_samples), 0]).domain([0, node_samples]);
110 |                 d.values_info = d3.keys(target_colors).map(function(i, target_name){
111 |                     // y0 will update 
112 |                     return {target_name: target_name, y0: y(y0), y1: y(y0 += d.value[i])}
113 |                 })
114 |                 return d.values_info; 
115 |               })
116 |             .enter().append("rect")
117 |                 .attr("width", rect_width)
118 |                 .attr("y", function(d) { return d.y1; })
119 |                 .attr("height", function(d) { return d.y0 - d.y1; })
120 |                 .style("fill", function(d) { 
121 |                     return target_colors[d.target_name];
122 |                 });
123 | 
124 |     // add text on tree nodes
125 |     var text_gap = 15;
126 |     var text = nodeEnter.append('text')
127 |     		.attr('x', 18)
128 |     		.attr('y', function(d){
129 |                 return h_node(d.sample) / 2 - text_gap + 3;             
130 |     		});
131 | 
132 |     text.append('tspan')
133 |     		.style('font-weight', 'bold')
134 |     		.text(function(d) {
135 |                 return d.name;
136 |     		});
137 |     
138 |     text.append('tspan')
139 |             .attr('x', 18)
140 |             .attr('y', function(d){
141 |                 return h_node(d.sample) / 2 + 3;
142 |             })
143 |             .style('font-size', '11px')
144 |             .style('fill', '#555')
145 |             .text(function(d) {
146 |                 return 'Values: [' + d.value + ']';
147 |             });
148 | 
149 |     text.append('tspan')
150 |             .attr('x', 18)
151 |             .attr('y', function(d){
152 |                 return h_node(d.sample) / 2 + text_gap + 3;
153 |             })
154 |             .style('font-size', '11px')
155 |             .style('fill', '#555')
156 |             .text(function(d) {
157 |                 return 'Predict: ' + d.predict;
158 |             });
159 | 
160 |     nodeEnter.select('text').style('fill-opacity', 0); 
161 | 
162 |     var nodeUpdate = node.transition()
163 |     		.duration(duration)
164 |     		.attr('transform', function(d) {
165 |                 var height = h_node(+d.sample);                
166 |     			var width = rect_width;
167 |     			return 'translate(' + (d.y - width / 2) + ',' + (d.x - height / 2) + ')';
168 |     		});
169 | 
170 |     nodeUpdate.select('text').style('fill-opacity', 1);
171 | 
172 |     // transition exiting nodes to the parent's new position
173 |     var nodeExit = node.exit().transition()
174 |     		.duration(duration)
175 |     		.attr('transform', function(d) {
176 |     			return 'translate(' + source.y + ',' + source.x + ')';
177 |     		})
178 |     		.remove();
179 | 
180 |     // hide nodes and texts
181 |     nodeExit.select('rect').attr('height', 1e-6);
182 |     nodeExit.select('text').style('fill-opacity', 0);
183 | 
184 |     // draw the links
185 |     var link = svg.selectAll('path.link')
186 |     		.data(links, function(d) {
187 |     			return d.target.id;
188 |     		});
189 | 
190 |     // enter any new links at the parent's previous position
191 |     link.enter().insert('path', 'g')
192 |     		.attr('class', 'link')
193 |     		.attr('d', function(d) {
194 |                 var o = {x: source.x0, y : source.y0};
195 |                 d_new = {source: {x: o.y, y:o.x, sample:d.source.sample}, target: {x: o.y, y: o.x, sample:d.target.sample}};
196 |                 return sankey_link(d_new);
197 |     		})
198 |             .style('stroke', '#eee');
199 | 
200 |     // transition links to their new position
201 |     link.transition()
202 |     		.duration(duration)
203 |             .attr('d', function(d){
204 |                 var source_h = h_node(d.source.sample);
205 |                 var target_h = h_node(d.target.sample);
206 |                 // center the starting points 
207 |                 if(d.target.pos=="left") {
208 |                     var o = {x: d.source.x - (source_h - target_h) / 2., y : d.source.y};
209 |                 }else{
210 |                     var o = {x: d.source.x + (source_h - target_h) / 2., y : d.source.y};
211 |                 }
212 |                 d_new = {source: {x: o.y, y:o.x, sample:d.source.sample}, 
213 |                          target: {x: d.target.y, y: d.target.x, sample:d.target.sample}};
214 |                 return sankey_link(d_new);
215 |             });
216 | 
217 |     // transition the existing links to the parent's new position
218 |     link.exit().transition()
219 |     		.duration(duration)
220 |     		.attr('d', function(d) {
221 |     			var o = {x: source.x, y: source.y};
222 |                 d_new = {source: {x: o.y, y:o.x, sample:d.source.sample}, 
223 |                          target: {x: o.y, y: o.x, sample:d.target.sample}};
224 |                 return sankey_link(d_new);
225 |     		}).remove();
226 | 
227 |     // stash the old positions for transition
228 |     nodes.forEach(function(d) {
229 |     	d.x0 = d.x;
230 |     	d.y0 = d.y;
231 |     });
232 | }
233 | 
234 | function sankey_link(d_new) {
235 |     var source_h = h_node(d_new.source.sample);
236 |     var target_h = h_node(d_new.target.sample);
237 | 
238 |     var curvature = .5;
239 |     var x0 = d_new.source.x,
240 |         x1 = d_new.target.x,
241 |         xi = d3.interpolateNumber(x0, x1),
242 |         x2 = xi(curvature),
243 |         x3 = xi(1 - curvature),
244 |         y0 = d_new.source.y - target_h / 2.,
245 |         y1 = d_new.target.y - target_h / 2.;
246 |     return "M" + x0 + "," + y0
247 |          + "C" + x2 + "," + y0
248 |          + " " + x3 + "," + y1
249 |          + " " + x1 + "," + y1
250 |          + "L" + x1 + "," + (y1+target_h)
251 |          + "C" + x3 + "," + (y1+target_h)
252 |          + " " + x2 + "," + (y0+target_h)
253 |          + " " + x0 + "," + (y0+target_h)
254 |          + "L" + x0 + "," + y0;
255 | }
256 | 
257 | // recursively color the links when hovering on a tree node
258 | function recurse(d, color) {
259 | 	d3.selectAll('.link').each(function(l) {
260 | 		if(l.target === d) {
261 | 			d3.select(this).style('fill', color);
262 | 		}
263 | 	});
264 | 	if(d.parent) {
265 | 		recurse(d.parent, color);
266 | 	}
267 | }
268 | 
269 | function mouseover(d) {
270 |     // if this node has no children nodes
271 | 	if(!d.children) {
272 |         var rule = rules[d.self];
273 |         var append = ['Predict: ' + d.predict, 'Impurity: ' + d.impurity, ''];
274 |         rule = append.concat(rule);
275 | 
276 |         var len = rule.length;
277 |         var info = d3.select(this).append('g').attr('class', 'hover');
278 |         var y0 = h_node(d.sample) + 20;
279 |         var rules_text = info.append('g').append('text').attr('y', y0);
280 | 
281 |         for(var i = 1; i <= len; i++) {
282 |             rules_text.append('tspan')
283 |                     .attr('x', 0)
284 |                     .attr('y', y0 + i * 15)
285 |                     .attr('text-anchor', 'left')
286 |                     .style('font-weight', function(){
287 |                         if(i <= 3) {
288 |                             return 'bold';
289 |                         }else{
290 |                             return null;
291 |                         }
292 |                     })
293 |                     .style('fill', function(){
294 |                         if(i <= 3) {
295 |                             return '#000';
296 |                         }else{
297 |                             return d.color;
298 |                         }
299 |                     })
300 |                     .text(function() {
301 |                         return rule[i - 1];
302 |                     });
303 |         }
304 | 
305 |         // draw the box
306 |         var bbox = info.node().getBBox();
307 |         var padding = 10;
308 |         var rect = d3.select(this).insert('rect', 'g')
309 |                 .attr('class', 'info_bg')
310 |                 .attr('x', bbox.x - padding)
311 |                 .attr('y', bbox.y - padding)
312 |                 .attr('width', bbox.width + padding * 2)
313 |                 .attr('height', bbox.height + padding * 2)
314 |                 .style('fill', 'white')
315 |                 .style('stroke', d.color)
316 |                 .style('stroke-width', 2);
317 | 
318 |         // make the rule infomation on the first layer
319 |         this.parentNode.appendChild(this);      
320 | 		recurse(d, d.color);
321 | 	
322 | 	}
323 | }
324 | 
325 | function mouseout(d) {
326 | 	d3.select('.hover').remove();
327 | 	d3.select('.info_bg').remove();
328 | 
329 | 	d3.select(this).style('stroke', function(d) {
330 | 		if(d._children){
331 | 			var stroke_color = '#000';
332 | 		}else{
333 | 			var stroke_color = null;
334 | 		}
335 | 		d3.select(this).selectAll('rect').style('stroke-width', 2).style('stroke', stroke_color);
336 | 		return null;
337 | 	});
338 | 
339 |     d3.selectAll('.link').style('fill', '#eee');
340 | }
341 | 
342 | function click(d) {
343 | 	if(d.children) {
344 | 		d._children = d.children;
345 | 		d.children = null;
346 | 	}else {
347 | 		d.children = d._children;
348 | 		d._children = null;
349 | 	}
350 | 	update(d);
351 | }
352 | 
353 | </script>
354 | </body>
355 | </html>


--------------------------------------------------------------------------------
/tree/src/simple_tree_template.html:
--------------------------------------------------------------------------------
  1 | <!DOCTYPE html>
  2 | <html lang='en'>
  3 | <head>
  4 | <meta charset='utf-8'>
  5 | <style>
  6 | .node {
  7 | 	cursor: pointer;
  8 | }
  9 | 
 10 | .node circle {
 11 | 	stroke-width: 3px;
 12 | }
 13 | 
 14 | .node text {
 15 | 	font: 12px helvetica;
 16 | }
 17 | 
 18 | .link {
 19 | 	fill: none;
 20 | 	stroke: #ccc;
 21 | 	stroke-width: 2px;
 22 | }
 23 | 
 24 | </style>
 25 | </head>
 26 | 
 27 | <body>
 28 | <script src="https://d3js.org/d3.v3.min.js"></script>
 29 | <script>
 30 | 
 31 | var margin = {top: 50, right: 100, bottom: 150, left: 50};
 32 | 
 33 | // collect all inputs
 34 | var tree_depth = {{ tree_depth }};
 35 | var rules = {{ rule|safe }};
 36 | var num_node = {{ num_node }};
 37 | var root = {{ tree|safe }};
 38 | var width = {{ width }}, height = {{ height }};
 39 | var n_classes = {{ n_classes }};
 40 | 
 41 | // calculate tree gap
 42 | var tree_gap = (width - margin['left'] - margin['right']) * 1.0 / tree_depth;
 43 | 
 44 | // generate tree layout (horizonal)
 45 | var tree = d3.layout.tree().size([height, width]);
 46 | var diagonal = d3.svg.diagonal()
 47 |         .projection(function(d){
 48 | 			return [d.y, d.x];
 49 | 		});
 50 | 
 51 | var svg = d3.select('body').append('svg')
 52 | 		.attr('width', width + margin.right + margin.left)
 53 | 		.attr('height', height + margin.top + margin.bottom)
 54 | 		.append('g')
 55 | 		.attr('transform', 'translate(' + margin.left + ',' + margin.top + ')');
 56 | 
 57 | // color scale based on node impurity
 58 | var node_color_scale = d3.scale.linear()
 59 |     .domain([1.0 / n_classes + 0.1, 0.])
 60 |     .clamp(true);
 61 | 
 62 | var i = 0, duration = 750;
 63 | root.x0 = height / 2;
 64 | root.y0 = 0;
 65 | 
 66 | // update tree
 67 | update(root);
 68 | 
 69 | 
 70 | function update(source) {
 71 | 
 72 |     // compute the tree layout
 73 |     var nodes = tree.nodes(root).reverse(),
 74 |     links = tree.links(nodes);
 75 | 
 76 |     // normalize for fixed-depth
 77 |     nodes.forEach(function(d) {
 78 |     	d.y = d.depth * tree_gap;
 79 |     });
 80 | 
 81 |     // draw the nodes
 82 |     var node = svg.selectAll('g.node')
 83 |     		.data(nodes, function(d) {
 84 |     			return d.id || (d.id = ++i);
 85 |     		});
 86 | 
 87 |     // enter any new nodes at the parent's previous position
 88 |     var nodeEnter = node.enter().append('g')
 89 |     		.attr('class', 'node')
 90 |     		.on('click', click)
 91 |     		.attr('transform', function(d) {
 92 |     			return 'translate(' + source.y0 + ',' + source.x0 + ')';
 93 |     		})
 94 |     		.on('click', click)
 95 |     		.on('mouseover', mouseover)
 96 |     		.on('mouseout', mouseout);
 97 | 
 98 |     // draw tree nodes
 99 |     nodeEnter.append('circle')
100 |     		.attr('r', 1e-6)
101 |     		.style('fill', function(d) {
102 |                 // d.color is the target color
103 |     			node_color_scale.range(['#fff', d.color]);
104 |     			return node_color_scale(d.impurity);
105 |     		})
106 |     		.style('stroke', function(d) {
107 |     			return d.color;
108 |     		})
109 |     		.on('mouseover', function(d) {
110 |     			if(!d.children) {
111 |     				d3.select(this).style('fill', function(d){
112 |     					return d.color;
113 |     				});
114 |     				recurse(d, d.color);
115 |     			}
116 |     		})
117 |             // when mouseout, recover everything
118 |     		.on('mouseout', function(d) {
119 |     			d3.select(this).style('fill', function(d) {
120 |     				node_color_scale.range(['#fff', d.color]);
121 |     				return node_color_scale(d.impurity);
122 |     			}).attr('r', function(d){
123 |                     // when the node is collapesd, increase the size
124 |                     return d._children? 20 : 10;
125 |                 });
126 |     			d3.selectAll('.link')
127 |                     .style('stroke-width', '2')
128 |                     .style('stroke', '#ccc');
129 |     		});
130 | 
131 |     // add text on tree nodes
132 |     var text = nodeEnter.append('text')
133 |     		.attr('x', 13)
134 |     		.attr('dy', '-0.5em');
135 | 
136 |     text.append('tspan')
137 |     		.style('font-weight', 'bold')
138 |     		.text(function(d) {
139 |                 return d.name;
140 |     		});
141 | 
142 |     text.append('tspan')
143 |     		.attr('x', 13)
144 |     		.attr('y', 8)
145 |     		.text(function(d) {
146 |     			return 'Values: [' + d.value + ']';
147 |     		});
148 | 
149 |     text.append('tspan')
150 |     		.attr('x', 13)
151 |     		.attr('y', 22)
152 |     		.text(function(d) {
153 |     			return 'Predict: ' + d.predict;
154 |     		});
155 | 
156 |     nodeEnter.select('text').style('fill-opacity', 0);
157 | 
158 |     // transform tree nodes to their initial positions
159 |     var nodeUpdate = node.transition()
160 |     		.duration(duration)
161 |     		.attr('transform', function(d) {
162 |     			return 'translate(' + d.y + ',' + d.x + ')';
163 |     		});
164 | 
165 |     nodeUpdate.select('circle')
166 |     		.attr('r', function(d){
167 |     			return d._children? 20 : 10;
168 |     		})
169 |     		.style('stroke', function(d) {
170 |     			return d.color;
171 |     		})
172 |     		.style('fill', function(d) {
173 |     			node_color_scale.range(['#fff', d.color]);
174 |     			return node_color_scale(d.impurity);
175 |     		});
176 | 
177 |     nodeUpdate.select('text').style('fill-opacity', 1);
178 | 
179 |     // transition exiting nodes to the parent's new position
180 |     var nodeExit = node.exit().transition()
181 |     		.duration(duration)
182 |     		.attr('transform', function(d) {
183 |     			return 'translate(' + source.y + ',' + source.x + ')';
184 |     		})
185 |     		.remove();
186 | 
187 |     // hide nodes and texts
188 |     nodeExit.select('circle').attr('r', 1e-6);
189 |     nodeExit.select('text').style('fill-opacity', 0);
190 | 
191 |     // draw the links
192 |     var link = svg.selectAll('path.link')
193 |     		.data(links, function(d) {
194 |     			return d.target.id;
195 |     		});
196 | 
197 |     // enter any new links at the parent's previous position
198 |     link.enter().insert('path', 'g')
199 |     		.attr('class', 'link')
200 |     		.attr('d', function(d) {
201 |     			var o = {x: source.x0, y : source.y0};
202 |     			return diagonal({source:o, target: o});
203 |     		});
204 | 
205 |     // transition links to their new position
206 |     link.transition()
207 |     		.duration(duration)
208 |     		.attr('d', diagonal);
209 | 
210 |     // transition the existing links to the parent's new position
211 |     link.exit().transition()
212 |     		.duration(duration)
213 |     		.attr('d', function(d) {
214 |     			var o = {x: source.x, y: source.y};
215 |     			return diagonal({source: o, target: o});
216 |     		}).remove();
217 | 
218 |     // stash the old positions for transition
219 |     nodes.forEach(function(d) {
220 |     	d.x0 = d.x;
221 |     	d.y0 = d.y;
222 |     });
223 | }
224 | 
225 | // recursively color the links when hovering on a tree node
226 | function recurse(d, color) {
227 | 	d3.selectAll('.link').each(function(l) {
228 | 		if(l.target === d) {
229 | 			d3.select(this).style('stroke-width', 4).style('stroke', color);
230 | 		}
231 | 	});
232 | 	if(d.parent) {
233 | 		recurse(d.parent, color);
234 | 	}
235 | }
236 | 
237 | function mouseover(d) {
238 |     // if this node has no children nodes
239 | 	if(!d.children) {
240 | 		var rule = rules[d.self];
241 | 		var append = ['Predict: ' + d.predict, 'Impurity: ' + d.impurity, ''];
242 | 		rule = append.concat(rule);
243 | 
244 | 		var len = rule.length;
245 | 		var info = d3.select(this).append('g').attr('class', 'hover');
246 | 		var y0 = 35;
247 | 		var rules_text = info.append('g').append('text').attr('y', y0);
248 | 
249 | 		for(var i = 1; i <= len; i++) {
250 | 			rules_text.append('tspan')
251 | 					.attr('x', 0)
252 | 					.attr('y', y0 + i * 15)
253 | 					.attr('text-anchor', 'left')
254 | 					.style('font-weight', function(){
255 | 						if(i <= 3) {
256 | 							return 'bold';
257 | 						}else{
258 | 							return null;
259 | 						}
260 | 					})
261 | 					.style('fill', function(){
262 | 						if(i <= 3) {
263 | 							return '#000';
264 | 						}else{
265 | 							return d.color;
266 | 						}
267 | 					})
268 | 					.text(function() {
269 | 						return rule[i - 1];
270 | 					});
271 | 		}
272 | 
273 |         // draw the box
274 | 		var bbox = info.node().getBBox();
275 | 		var padding = 10;
276 | 		var rect = d3.select(this).insert('rect', 'g')
277 | 				.attr('class', 'info_bg')
278 | 				.attr('x', bbox.x - padding)
279 | 				.attr('y', bbox.y - padding)
280 | 				.attr('width', bbox.width + padding * 2)
281 | 				.attr('height', bbox.height + padding * 2)
282 | 				.style('fill', 'white')
283 | 				.style('stroke', d.color)
284 | 				.style('stroke-width', 2);
285 | 
286 | 		// make the rule infomation on the first layer
287 | 		this.parentNode.appendChild(this);		
288 | 	}
289 | }
290 | 
291 | function mouseout(d) {
292 | 	d3.select('.hover').remove();
293 | 	d3.select('.info_bg').remove();
294 | }
295 | 
296 | function click(d) {
297 | 	if(d.children) {
298 | 		d._children = d.children;
299 | 		d.children = null;
300 | 	}else {
301 | 		d.children = d._children;
302 | 		d._children = null;
303 | 	}
304 | 	update(d);
305 | }
306 | 
307 | </script>
308 | </body>
309 | </html>


--------------------------------------------------------------------------------