├── .gitignore ├── LICENSE ├── README.md ├── ch01 ├── dtree.py └── test.py ├── ch02 ├── alphabets.txt ├── dtree.py ├── sample.txt └── test.py ├── ch03 ├── dtree.py └── test.py ├── ch04 ├── census.csv ├── dtree.py └── test.py ├── ch05 ├── dtree.py ├── forest.py ├── test.py └── titanic.py ├── ch06 ├── dtree.py └── test.py └── ch07 ├── dtree.py ├── forest.py └── test.py /.gitignore: -------------------------------------------------------------------------------- 1 | **/__pycache__/* 2 | .idea/* 3 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | Apache License 2 | Version 2.0, January 2004 3 | http://www.apache.org/licenses/ 4 | 5 | TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION 6 | 7 | 1. Definitions. 8 | 9 | "License" shall mean the terms and conditions for use, reproduction, 10 | and distribution as defined by Sections 1 through 9 of this document. 11 | 12 | "Licensor" shall mean the copyright owner or entity authorized by 13 | the copyright owner that is granting the License. 14 | 15 | "Legal Entity" shall mean the union of the acting entity and all 16 | other entities that control, are controlled by, or are under common 17 | control with that entity. For the purposes of this definition, 18 | "control" means (i) the power, direct or indirect, to cause the 19 | direction or management of such entity, whether by contract or 20 | otherwise, or (ii) ownership of fifty percent (50%) or more of the 21 | outstanding shares, or (iii) beneficial ownership of such entity. 22 | 23 | "You" (or "Your") shall mean an individual or Legal Entity 24 | exercising permissions granted by this License. 25 | 26 | "Source" form shall mean the preferred form for making modifications, 27 | including but not limited to software source code, documentation 28 | source, and configuration files. 29 | 30 | "Object" form shall mean any form resulting from mechanical 31 | transformation or translation of a Source form, including but 32 | not limited to compiled object code, generated documentation, 33 | and conversions to other media types. 34 | 35 | "Work" shall mean the work of authorship, whether in Source or 36 | Object form, made available under the License, as indicated by a 37 | copyright notice that is included in or attached to the work 38 | (an example is provided in the Appendix below). 39 | 40 | "Derivative Works" shall mean any work, whether in Source or Object 41 | form, that is based on (or derived from) the Work and for which the 42 | editorial revisions, annotations, elaborations, or other modifications 43 | represent, as a whole, an original work of authorship. For the purposes 44 | of this License, Derivative Works shall not include works that remain 45 | separable from, or merely link (or bind by name) to the interfaces of, 46 | the Work and Derivative Works thereof. 47 | 48 | "Contribution" shall mean any work of authorship, including 49 | the original version of the Work and any modifications or additions 50 | to that Work or Derivative Works thereof, that is intentionally 51 | submitted to Licensor for inclusion in the Work by the copyright owner 52 | or by an individual or Legal Entity authorized to submit on behalf of 53 | the copyright owner. For the purposes of this definition, "submitted" 54 | means any form of electronic, verbal, or written communication sent 55 | to the Licensor or its representatives, including but not limited to 56 | communication on electronic mailing lists, source code control systems, 57 | and issue tracking systems that are managed by, or on behalf of, the 58 | Licensor for the purpose of discussing and improving the Work, but 59 | excluding communication that is conspicuously marked or otherwise 60 | designated in writing by the copyright owner as "Not a Contribution." 61 | 62 | "Contributor" shall mean Licensor and any individual or Legal Entity 63 | on behalf of whom a Contribution has been received by Licensor and 64 | subsequently incorporated within the Work. 65 | 66 | 2. Grant of Copyright License. Subject to the terms and conditions of 67 | this License, each Contributor hereby grants to You a perpetual, 68 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 69 | copyright license to reproduce, prepare Derivative Works of, 70 | publicly display, publicly perform, sublicense, and distribute the 71 | Work and such Derivative Works in Source or Object form. 72 | 73 | 3. Grant of Patent License. Subject to the terms and conditions of 74 | this License, each Contributor hereby grants to You a perpetual, 75 | worldwide, non-exclusive, no-charge, royalty-free, irrevocable 76 | (except as stated in this section) patent license to make, have made, 77 | use, offer to sell, sell, import, and otherwise transfer the Work, 78 | where such license applies only to those patent claims licensable 79 | by such Contributor that are necessarily infringed by their 80 | Contribution(s) alone or by combination of their Contribution(s) 81 | with the Work to which such Contribution(s) was submitted. If You 82 | institute patent litigation against any entity (including a 83 | cross-claim or counterclaim in a lawsuit) alleging that the Work 84 | or a Contribution incorporated within the Work constitutes direct 85 | or contributory patent infringement, then any patent licenses 86 | granted to You under this License for that Work shall terminate 87 | as of the date such litigation is filed. 88 | 89 | 4. Redistribution. You may reproduce and distribute copies of the 90 | Work or Derivative Works thereof in any medium, with or without 91 | modifications, and in Source or Object form, provided that You 92 | meet the following conditions: 93 | 94 | (a) You must give any other recipients of the Work or 95 | Derivative Works a copy of this License; and 96 | 97 | (b) You must cause any modified files to carry prominent notices 98 | stating that You changed the files; and 99 | 100 | (c) You must retain, in the Source form of any Derivative Works 101 | that You distribute, all copyright, patent, trademark, and 102 | attribution notices from the Source form of the Work, 103 | excluding those notices that do not pertain to any part of 104 | the Derivative Works; and 105 | 106 | (d) If the Work includes a "NOTICE" text file as part of its 107 | distribution, then any Derivative Works that You distribute must 108 | include a readable copy of the attribution notices contained 109 | within such NOTICE file, excluding those notices that do not 110 | pertain to any part of the Derivative Works, in at least one 111 | of the following places: within a NOTICE text file distributed 112 | as part of the Derivative Works; within the Source form or 113 | documentation, if provided along with the Derivative Works; or, 114 | within a display generated by the Derivative Works, if and 115 | wherever such third-party notices normally appear. The contents 116 | of the NOTICE file are for informational purposes only and 117 | do not modify the License. You may add Your own attribution 118 | notices within Derivative Works that You distribute, alongside 119 | or as an addendum to the NOTICE text from the Work, provided 120 | that such additional attribution notices cannot be construed 121 | as modifying the License. 122 | 123 | You may add Your own copyright statement to Your modifications and 124 | may provide additional or different license terms and conditions 125 | for use, reproduction, or distribution of Your modifications, or 126 | for any such Derivative Works as a whole, provided Your use, 127 | reproduction, and distribution of the Work otherwise complies with 128 | the conditions stated in this License. 129 | 130 | 5. Submission of Contributions. Unless You explicitly state otherwise, 131 | any Contribution intentionally submitted for inclusion in the Work 132 | by You to the Licensor shall be under the terms and conditions of 133 | this License, without any additional terms or conditions. 134 | Notwithstanding the above, nothing herein shall supersede or modify 135 | the terms of any separate license agreement you may have executed 136 | with Licensor regarding such Contributions. 137 | 138 | 6. Trademarks. This License does not grant permission to use the trade 139 | names, trademarks, service marks, or product names of the Licensor, 140 | except as required for reasonable and customary use in describing the 141 | origin of the Work and reproducing the content of the NOTICE file. 142 | 143 | 7. Disclaimer of Warranty. Unless required by applicable law or 144 | agreed to in writing, Licensor provides the Work (and each 145 | Contributor provides its Contributions) on an "AS IS" BASIS, 146 | WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 147 | implied, including, without limitation, any warranties or conditions 148 | of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A 149 | PARTICULAR PURPOSE. You are solely responsible for determining the 150 | appropriateness of using or redistributing the Work and assume any 151 | risks associated with Your exercise of permissions under this License. 152 | 153 | 8. Limitation of Liability. In no event and under no legal theory, 154 | whether in tort (including negligence), contract, or otherwise, 155 | unless required by applicable law (such as deliberate and grossly 156 | negligent acts) or agreed to in writing, shall any Contributor be 157 | liable to You for damages, including any direct, indirect, special, 158 | incidental, or consequential damages of any character arising as a 159 | result of this License or out of the use or inability to use the 160 | Work (including but not limited to damages for loss of goodwill, 161 | work stoppage, computer failure or malfunction, or any and all 162 | other commercial damages or losses), even if such Contributor 163 | has been advised of the possibility of such damages. 164 | 165 | 9. Accepting Warranty or Additional Liability. While redistributing 166 | the Work or Derivative Works thereof, You may choose to offer, 167 | and charge a fee for, acceptance of support, warranty, indemnity, 168 | or other liability obligations and/or rights consistent with this 169 | License. However, in accepting such obligations, You may act only 170 | on Your own behalf and on Your sole responsibility, not on behalf 171 | of any other Contributor, and only if You agree to indemnify, 172 | defend, and hold each Contributor harmless for any liability 173 | incurred by, or claims asserted against, such Contributor by reason 174 | of your accepting any such warranty or additional liability. 175 | 176 | END OF TERMS AND CONDITIONS 177 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # Tree-based Machine Learning Algorithms 2 | Source code from the book Tree-based Machine Learning Algorithms by Clinton Sheppard 3 | 4 | Description 5 | === 6 | 7 | Tree-based Machine Learning Algorithms cover 8 | Get a hands-on introduction to building and using decision trees and random forests. Tree-based machine learning algorithms are used to categorize data based by known outcomes in order to facilitate predicting outcomes in new situations. 9 | 10 | You will learn not only how to use decision trees and random forests for classification and regression, and their respective limitations, but also how the algorithms that build them work. Each chapter introduces a new data concern and then walks you through modifying the code, thus building the engine just-in-time. Along the way you will gain experience making decision trees and random forests work for you. 11 | 12 | Available from major stores including Amazon and Barnes & Noble, in paperback and Kindle formats. 13 | 14 | - https://www.amazon.com/Tree-based-Machine-Learning-Algorithms-Decision/dp/1975860977 ISBN-13: 978-1-975-86097-4 (paperback) 15 | - https://www.amazon.com/dp/B0756FGJCP/ (Kindle Edition) 16 | 17 | Table of Contents 18 | === 19 | 20 | A brief introduction to decision trees 21 | 22 | Chapter 1: Branching 23 | - uses a greedy algorithm to build a decision tree from data that can be split on a single attribute. 24 | 25 | Chapter 2: Multiple Branches 26 | - examines several ways to split data in order to generate multi-level decision trees. 27 | 28 | Chapter 3: Continuous Attributes 29 | - adds the ability to split numeric attributes using greater-than. 30 | 31 | Chapter 4: Pruning 32 | - explore ways of reducing the amount of error encoded in the tree. 33 | 34 | Chapter 5: Random Forests 35 | - introduces ensemble learning and feature engineering. 36 | 37 | Chapter 6: Regression Trees 38 | - investigates numeric predictions, like age, price, and miles per gallon. 39 | 40 | Chapter 7: Boosting 41 | - adjusts the voting power of the randomly selected decision trees in the random forest in order to improve its ability to predict outcomes. 42 | 43 | -------------------------------------------------------------------------------- /ch01/dtree.py: -------------------------------------------------------------------------------- 1 | # File: dtree.py 2 | # from chapter 1 of _Tree-based Machine Learning Algorithms_ 3 | # 4 | # Author: Clinton Sheppard 5 | # Copyright (c) 2017 Clinton Sheppard 6 | # 7 | # Licensed under the Apache License, Version 2.0 (the "License"). 8 | # You may not use this file except in compliance with the License. 9 | # You may obtain a copy of the License at 10 | # http://www.apache.org/licenses/LICENSE-2.0 11 | # 12 | # Unless required by applicable law or agreed to in writing, software 13 | # distributed under the License is distributed on an "AS IS" BASIS, 14 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 15 | # implied. See the License for the specific language governing 16 | # permissions and limitations under the License. 17 | 18 | from collections import Counter 19 | 20 | 21 | def build(data, outcomeLabel): 22 | attrIndexes = [index for index, label in enumerate(data[0]) if 23 | label != outcomeLabel] 24 | outcomeIndex = data[0].index(outcomeLabel) 25 | 26 | nodes = [] 27 | lastNodeNumber = 0 28 | 29 | workQueue = [(-1, lastNodeNumber, set(i for i in range(1, len(data))))] 30 | while len(workQueue) > 0: 31 | parentNodeId, nodeId, dataRowIndexes = workQueue.pop() 32 | uniqueOutcomes = set(data[i][outcomeIndex] for i in dataRowIndexes) 33 | if len(uniqueOutcomes) == 1: 34 | nodes.append((nodeId, uniqueOutcomes.pop())) 35 | continue 36 | attrValueResults = [] 37 | for attrIndex in attrIndexes: 38 | for rowIndex in dataRowIndexes: 39 | row = data[rowIndex] 40 | value = row[attrIndex] 41 | attrValueResults.append((attrIndex, value)) 42 | potentials = [i for i in Counter(attrValueResults).most_common(1)] 43 | attrIndex, attrValue = potentials[0][0] 44 | matches = {rowIndex for rowIndex in dataRowIndexes if 45 | data[rowIndex][attrIndex] == attrValue} 46 | nonMatches = dataRowIndexes - matches 47 | lastNodeNumber += 1 48 | matchId = lastNodeNumber 49 | workQueue.append((nodeId, matchId, matches)) 50 | lastNodeNumber += 1 51 | nonMatchId = lastNodeNumber 52 | workQueue.append((nodeId, nonMatchId, nonMatches)) 53 | nodes.append((nodeId, attrIndex, attrValue, matchId, nonMatchId)) 54 | nodes = sorted(nodes, key=lambda n: n[0]) 55 | return DTree(nodes, data[0]) 56 | 57 | 58 | class DTree: 59 | def __init__(self, nodes, attrNames): 60 | self._nodes = nodes 61 | self._attrNames = attrNames 62 | 63 | @staticmethod 64 | def _is_leaf(node): 65 | return len(node) == 2 66 | 67 | def __str__(self): 68 | s = '' 69 | for node in self._nodes: 70 | if self._is_leaf(node): 71 | s += '{}: {}\n'.format(node[0], node[1]) 72 | else: 73 | nodeId, attrIndex, attrValue, nodeIdIfMatch, \ 74 | nodeIdIfNonMatch = node 75 | s += '{}: {}={}, Yes->{}, No->{}\n'.format( 76 | nodeId, self._attrNames[attrIndex], attrValue, 77 | nodeIdIfMatch, nodeIdIfNonMatch) 78 | return s 79 | 80 | def get_prediction(self, data): 81 | currentNode = self._nodes[0] 82 | while True: 83 | if self._is_leaf(currentNode): 84 | return currentNode[1] 85 | nodeId, attrIndex, attrValue, nodeIdIfMatch, \ 86 | nodeIdIfNonMatch = currentNode 87 | currentNode = self._nodes[nodeIdIfMatch if 88 | data[attrIndex] == attrValue else nodeIdIfNonMatch] 89 | -------------------------------------------------------------------------------- /ch01/test.py: -------------------------------------------------------------------------------- 1 | # File: test.py 2 | # from chapter 1 of _Tree-based Machine Learning Algorithms_ 3 | # 4 | # Author: Clinton Sheppard 5 | # Copyright (c) 2017 Clinton Sheppard 6 | # 7 | # Licensed under the Apache License, Version 2.0 (the "License"). 8 | # You may not use this file except in compliance with the License. 9 | # You may obtain a copy of the License at 10 | # http://www.apache.org/licenses/LICENSE-2.0 11 | # 12 | # Unless required by applicable law or agreed to in writing, software 13 | # distributed under the License is distributed on an "AS IS" BASIS, 14 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 15 | # implied. See the License for the specific language governing 16 | # permissions and limitations under the License. 17 | 18 | import dtree 19 | 20 | data = [['Name', 'Gender', 'Born'], 21 | ['William', 'male', 'Germany'], 22 | ['Louise', 'female', 'Texas'], 23 | ['Minnie', 'female', 'Texas'], 24 | ['Emma', 'female', 'Texas'], 25 | ['Henry', 'male', 'Germany'], 26 | ] 27 | 28 | outcomeLabel = 'Born' 29 | 30 | tree = dtree.build(data, outcomeLabel) 31 | print(tree) 32 | 33 | testData = ["Alice", "female"] 34 | predicted = tree.get_prediction(testData) 35 | print("predicted: {}".format(predicted)) 36 | -------------------------------------------------------------------------------- /ch02/alphabets.txt: -------------------------------------------------------------------------------- 1 | # source: https://github.com/handcraftsman/TreeBasedMachineLearningAlgorithms/blob/master/ch02/alphabets.txt 2 | Albanian A a B b C c Ç ç D d Dh dh E e Ë ë F f G g Gj gj H h I i J j K k L l Ll ll M m N n Nj nj O o P p Q q R r Rr rr S s Sh sh T t Th th U u V v X x Xh xh Y y Z z Zh zh 3 | Croatian A a B b C c Č č Ć ć D d Dž dž Đ đ E e F f G g H h I i J j K k L l Lj lj M m N n Nj nj O o P p R r S s Š š T t U u V v Z z Ž ž 4 | Czech A a Á á B b C c Č č D d Ď ď E e É é Ě ě F f G g H h Ch ch I i Í í J j K k L l M m N n Ň ň O o Ó ó P p Q q R r Ř ř S s Š š T t Ť ť U u Ú ú Ů ů V v W w X x Y y Ý ý Z z Ž ž 5 | Danish A a B b C c D d E e F f G g H h I i J j K k L l M m N n O o P p Q q R r S s T t U u V v W w X x Y y Z z Æ æ Ø ø Å å 6 | Dutch A a B b, C c, D d, E e F f, G g, H h, I i J j, K k, L l, M m, N n, O o P p, Q q, R r, S s, T t, U u V v, W w, X x, Y y Z z Á á, Ä ä É é, Ë ë Í í, Ï ï, IJ ij Ó ó, Ö ö Ú ú, Ü ü Ý ý 7 | English A a B b C c D d E e F f G g H h I i J j K k L l M m N n O o P p Q q R r S s T t U u V v W w X x Y y Z z 8 | Estonian A a B b C c D d E e F f G g H h I i J j K k L l M m N n O o P p Q q R r S s Š š Z z Ž ž T t U u V v W w Õ õ Ä ä Ö ö Ü ü X x Y y 9 | Finnish A a B b C c D d E e F f G g H h I i J j K k L l M m N n O o P p Q q R r S s T t U u V v W w X x Y y Z z Å å Ä ä Ö ö 10 | French A a B b C c D d E e F f G g H h I i J j K k L l M m N n O o P p Q q R r S s T t U u V v W w X x Y y Z z à â ç è é ë ò ô ö ù 11 | German A a B b C c D d E e F f G g H h I i J j K k L l M m N n O o P p Q q R r S s T t U u V v W w X x Y y Z z Ä ä Ö ö Ü ü ß 12 | Hungarian A a B b C c Cs cs D d Dz dz Dzs dzs E e F f G g Gy gy H h I i J j K k L l Ly ly M m N n Ny ny O o P p Q q R r S s Sz sz T t Ty ty U u V v W w X x Y y Z z Zs zs Á á É é Í í Ó ó Ö ö Ú ú Ü ü Ő ő Ű ű 13 | Icelandic A a Á á B b D d Ð ð E e É é F f G g H h I i Í í J j K k L l M m N n O o Ó ó P p R r S s T t U u Ú ú V v X x Y y Ý ý Þ þ Æ æ Ö ö 14 | Irish A a B b C c D d E e F f G g H h I i L l M m N n O o P p R r S s T t U u Á á É é Í í Ó ó Ú ú 15 | Italian A a B b C c D d E e F f G g H h I i L l M m N n O o P p Q q R r S s T t U u V v Z z à è é ì í ò ó ù ú 16 | Latvian A a Ā ā B b C c Č č D d E e Ē ē F f G g Ģ ģ H h I i Ī ī J j K k Ķ ķ L l Ļ ļ M m N n Ņ ņ O o P p R r S s Š š T t U u Ū ū V v Z z Ž ž 17 | Lithuanian A a Ą ą B b C c Č č D d E e Ę ę Ė ė F f G g H h I i Į į Y y J j K k L l M m N n O o P p R r S s Š š T t U u Ų ų Ū ū V v Z z Ž ž 18 | Luxembourgish A a B b C c D d E e F f G g H h I i J j K k L l M m N n O o P p Q q R r S s T t U u V v W w X x Y y Z z É é Ä ä Ë ë 19 | Maltese A a B b Ċ ċ D d E e F f Ġ ġ G g Għ għ H h Ħ ħ I i Ie ie J j K k L l M m N n O o P p Q q R r S s T t U u V v W w X x Ż ż Z z 20 | Norwegian A a B b C c D d E e F f G g H h I i J j K k L l M m N n O o P p Q q R r S s T t U u V v W w X x Y y Z z Æ æ Ø ø Å å 21 | Polish A a Ą ą B b C c Ć ć D d E e Ę ę F f G g H h I i J j K k L l Ł ł M m N n Ń ń O o Ó ó P p R r S s Ś ś T t U u W w Y y Z z Ź ź Ż ż 22 | Portuguese A a B b C c D d E e F f G g H h I i J j K k L l M m N n O o P p Q q R r S s T t U u V v W w X x Y y Z z Á á À à Â â Ã ã Ç ç É é Ê ê Í í Ó ó Ô ô Õ õ Ú ú 23 | Romanian A a Ă ă Â â B b C c D d E e F f G g H h I i Î î J j K k L l M m N n O o P p Q q R r S s Ș ș T t Ț ț U u V v W w X x Y y Z z 24 | Slovak A a Á á Ä ä B b C c Č č D d Ď ď Dz dz Dž dž E e É é F f G g H h Ch ch I i Í í J j K k L l Ĺ ĺ Ľ ľ M m N n Ň ň O o Ó ó Ô ô P p Q q R r Ŕ ŕ S s Š š T t Ť ť U u Ú ú V v W w X x Y y Ý ý Z z Ž ž 25 | Slovenian A a B b C c Č č D d E e F f G g H h I i J j K k L l M m N n O o P p R r S s Š š T t U u V v Z z Ž ž 26 | Spanish A a B b C c D d E e F f G g H h I i J j K k L l M m N n Ñ ñ O o P p Q q R r S s T t U u V v W w X x Y y Z z Á á É é Í í Ó ó Ú ú Ü ü 27 | Swedish A a B b C c D d E e F f G g H h I i J j K k L l M m N n O o P p Q q R r S s T t U u V v W w X x Y y Z z Å å Ä ä Ö ö 28 | Turkish A a B b C c Ç ç D d E e F f G g Ğ ğ H h I ı İ i J j K k L l M m N n O o Ö ö P p R r S s Ş ş T t U u Ü ü V v Y y Z z -------------------------------------------------------------------------------- /ch02/dtree.py: -------------------------------------------------------------------------------- 1 | # File: dtree.py 2 | # from chapter 2 of _Tree-based Machine Learning Algorithms_ 3 | # 4 | # Author: Clinton Sheppard 5 | # Copyright (c) 2017 Clinton Sheppard 6 | # 7 | # Licensed under the Apache License, Version 2.0 (the "License"). 8 | # You may not use this file except in compliance with the License. 9 | # You may obtain a copy of the License at 10 | # http://www.apache.org/licenses/LICENSE-2.0 11 | # 12 | # Unless required by applicable law or agreed to in writing, software 13 | # distributed under the License is distributed on an "AS IS" BASIS, 14 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 15 | # implied. See the License for the specific language governing 16 | # permissions and limitations under the License. 17 | 18 | def _get_bias(avPair, dataRowIndexes, data, outcomeIndex): 19 | attrIndex, attrValue = avPair 20 | matchIndexes = {i for i in dataRowIndexes if 21 | data[i][attrIndex] == attrValue} 22 | nonMatchIndexes = dataRowIndexes - matchIndexes 23 | matchOutcomes = {data[i][outcomeIndex] for i in matchIndexes} 24 | nonMatchOutcomes = {data[i][outcomeIndex] for i in nonMatchIndexes} 25 | numPureRows = (len(matchIndexes) if len(matchOutcomes) == 1 else 0) \ 26 | + (len(nonMatchIndexes) if len(nonMatchOutcomes) == 1 27 | else 0) 28 | percentPure = numPureRows / len(dataRowIndexes) 29 | 30 | numNonPureRows = len(dataRowIndexes) - numPureRows 31 | percentNonPure = 1 - percentPure 32 | split = 1 - abs(len(matchIndexes) - len(nonMatchIndexes)) / len( 33 | dataRowIndexes) - .001 34 | splitBias = split * percentNonPure if numNonPureRows > 0 else 0 35 | return splitBias + percentPure 36 | 37 | 38 | def build(data, outcomeLabel): 39 | attrIndexes = [index for index, label in enumerate(data[0]) if 40 | label != outcomeLabel] 41 | outcomeIndex = data[0].index(outcomeLabel) 42 | 43 | nodes = [] 44 | lastNodeNumber = 0 45 | 46 | workQueue = [(-1, lastNodeNumber, set(i for i in range(1, len(data))))] 47 | while len(workQueue) > 0: 48 | parentNodeId, nodeId, dataRowIndexes = workQueue.pop() 49 | uniqueOutcomes = set(data[i][outcomeIndex] for i in dataRowIndexes) 50 | if len(uniqueOutcomes) == 1: 51 | nodes.append((nodeId, uniqueOutcomes.pop())) 52 | continue 53 | uniqueAttributeValuePairs = {(attrIndex, data[rowIndex][attrIndex]) 54 | for attrIndex in attrIndexes 55 | for rowIndex in dataRowIndexes} 56 | potentials = sorted((-_get_bias(avPair, dataRowIndexes, data, 57 | outcomeIndex), avPair[0], avPair[1]) 58 | for avPair in uniqueAttributeValuePairs) 59 | attrIndex, attrValue = potentials[0][1:] 60 | matches = {rowIndex for rowIndex in dataRowIndexes if 61 | data[rowIndex][attrIndex] == attrValue} 62 | nonMatches = dataRowIndexes - matches 63 | lastNodeNumber += 1 64 | matchId = lastNodeNumber 65 | workQueue.append((nodeId, matchId, matches)) 66 | lastNodeNumber += 1 67 | nonMatchId = lastNodeNumber 68 | workQueue.append((nodeId, nonMatchId, nonMatches)) 69 | nodes.append((nodeId, attrIndex, attrValue, matchId, nonMatchId, 70 | len(matches), len(nonMatches))) 71 | nodes = sorted(nodes, key=lambda n: n[0]) 72 | return DTree(nodes, data[0]) 73 | 74 | 75 | class DTree: 76 | def __init__(self, nodes, attrNames): 77 | self._nodes = nodes 78 | self._attrNames = attrNames 79 | 80 | @staticmethod 81 | def _is_leaf(node): 82 | return len(node) == 2 83 | 84 | def __str__(self): 85 | s = '' 86 | for node in self._nodes: 87 | if self._is_leaf(node): 88 | s += '{}: {}\n'.format(node[0], node[1]) 89 | else: 90 | nodeId, attrIndex, attrValue, nodeIdIfMatch, \ 91 | nodeIdIfNonMatch, matchCount, nonMatchCount = node 92 | s += '{0}: {1}={2}, {5} Yes->{3}, {6} No->{4}\n'.format( 93 | nodeId, self._attrNames[attrIndex], attrValue, 94 | nodeIdIfMatch, nodeIdIfNonMatch, matchCount, 95 | nonMatchCount) 96 | return s 97 | 98 | def get_prediction(self, data): 99 | currentNode = self._nodes[0] 100 | while True: 101 | if self._is_leaf(currentNode): 102 | return currentNode[1] 103 | nodeId, attrIndex, attrValue, nodeIdIfMatch, \ 104 | nodeIdIfNonMatch = currentNode[:5] 105 | currentNode = self._nodes[nodeIdIfMatch if 106 | data[attrIndex] == attrValue else nodeIdIfNonMatch] 107 | -------------------------------------------------------------------------------- /ch02/sample.txt: -------------------------------------------------------------------------------- 1 | štkýck ľudia učia radostné správ -------------------------------------------------------------------------------- /ch02/test.py: -------------------------------------------------------------------------------- 1 | # File: test.py 2 | # from chapter 2 of _Tree-based Machine Learning Algorithms_ 3 | # 4 | # Author: Clinton Sheppard 5 | # Copyright (c) 2017 Clinton Sheppard 6 | # 7 | # Licensed under the Apache License, Version 2.0 (the "License"). 8 | # You may not use this file except in compliance with the License. 9 | # You may obtain a copy of the License at 10 | # http://www.apache.org/licenses/LICENSE-2.0 11 | # 12 | # Unless required by applicable law or agreed to in writing, software 13 | # distributed under the License is distributed on an "AS IS" BASIS, 14 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 15 | # implied. See the License for the specific language governing 16 | # permissions and limitations under the License. 17 | 18 | import dtree 19 | 20 | data = [['Name', 'Gender', 'Marital Status', 'Born'], 21 | ['William', 'male', 'Married', 'Germany'], 22 | ['Louise', 'female', 'Single', 'Texas'], 23 | ['Minnie', 'female', 'Single', 'Texas'], 24 | ['Emma', 'female', 'Single', 'Texas'], 25 | ['Henry', 'male', 'Single', 'Germany'], 26 | ['Theo', 'male', 'Single', 'Texas'], 27 | ] 28 | 29 | outcomeLabel = 'Born' 30 | 31 | tree = dtree.build(data, outcomeLabel) 32 | print(tree) 33 | 34 | testData = ['Sophie', 'female', 'Single'] 35 | predicted = tree.get_prediction(testData) 36 | print("predicted: {}".format(predicted)) 37 | -------------------------------------------------------------------------------- /ch03/dtree.py: -------------------------------------------------------------------------------- 1 | # File: dtree.py 2 | # from chapter 3 of _Tree-based Machine Learning Algorithms_ 3 | # 4 | # Author: Clinton Sheppard 5 | # Copyright (c) 2017 Clinton Sheppard 6 | # 7 | # Licensed under the Apache License, Version 2.0 (the "License"). 8 | # You may not use this file except in compliance with the License. 9 | # You may obtain a copy of the License at 10 | # http://www.apache.org/licenses/LICENSE-2.0 11 | # 12 | # Unless required by applicable law or agreed to in writing, software 13 | # distributed under the License is distributed on an "AS IS" BASIS, 14 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 15 | # implied. See the License for the specific language governing 16 | # permissions and limitations under the License. 17 | 18 | from numbers import Number 19 | import operator 20 | import math 21 | 22 | 23 | def _get_bias(avPair, dataRowIndexes, data, outcomeIndex): 24 | attrIndex, attrValue, isMatch = avPair 25 | matchIndexes = {i for i in dataRowIndexes if 26 | isMatch(data[i][attrIndex], attrValue)} 27 | nonMatchIndexes = dataRowIndexes - matchIndexes 28 | matchOutcomes = {data[i][outcomeIndex] for i in matchIndexes} 29 | nonMatchOutcomes = {data[i][outcomeIndex] for i in nonMatchIndexes} 30 | numPureRows = (len(matchIndexes) if len(matchOutcomes) == 1 else 0) \ 31 | + (len(nonMatchIndexes) if len(nonMatchOutcomes) == 1 32 | else 0) 33 | percentPure = numPureRows / len(dataRowIndexes) 34 | 35 | numNonPureRows = len(dataRowIndexes) - numPureRows 36 | percentNonPure = 1 - percentPure 37 | split = 1 - abs(len(matchIndexes) - len(nonMatchIndexes)) / len( 38 | dataRowIndexes) - .001 39 | splitBias = split * percentNonPure if numNonPureRows > 0 else 0 40 | return splitBias + percentPure 41 | 42 | 43 | def build(data, outcomeLabel, continuousAttributes=None): 44 | attrIndexes = [index for index, label in enumerate(data[0]) if 45 | label != outcomeLabel] 46 | outcomeIndex = data[0].index(outcomeLabel) 47 | continuousAttrIndexes = set() 48 | if continuousAttributes is not None: 49 | continuousAttrIndexes = {data[0].index(label) for label in 50 | continuousAttributes} 51 | if len(continuousAttrIndexes) != len(continuousAttributes): 52 | raise Exception( 53 | 'One or more continuous column names are duplicates.') 54 | else: 55 | for attrIndex in attrIndexes: 56 | uniqueValues = {row[attrIndex] for rowIndex, row in 57 | enumerate(data) if rowIndex > 0} 58 | numericValues = {value for value in uniqueValues if 59 | isinstance(value, Number)} 60 | if len(uniqueValues) == len(numericValues): 61 | continuousAttrIndexes.add(attrIndex) 62 | 63 | nodes = [] 64 | lastNodeNumber = 0 65 | 66 | workQueue = [(-1, lastNodeNumber, set(i for i in range(1, len(data))))] 67 | while len(workQueue) > 0: 68 | parentNodeId, nodeId, dataRowIndexes = workQueue.pop() 69 | uniqueOutcomes = set(data[i][outcomeIndex] for i in dataRowIndexes) 70 | if len(uniqueOutcomes) == 1: 71 | nodes.append((nodeId, uniqueOutcomes.pop())) 72 | continue 73 | potentials = _get_potentials(attrIndexes, continuousAttrIndexes, 74 | data, dataRowIndexes, outcomeIndex) 75 | attrIndex, attrValue, isMatch = potentials[0][1:] 76 | matches = {rowIndex for rowIndex in dataRowIndexes if 77 | isMatch(data[rowIndex][attrIndex], attrValue)} 78 | nonMatches = dataRowIndexes - matches 79 | lastNodeNumber += 1 80 | matchId = lastNodeNumber 81 | workQueue.append((nodeId, matchId, matches)) 82 | lastNodeNumber += 1 83 | nonMatchId = lastNodeNumber 84 | workQueue.append((nodeId, nonMatchId, nonMatches)) 85 | nodes.append((nodeId, attrIndex, attrValue, isMatch, matchId, 86 | nonMatchId, len(matches), len(nonMatches))) 87 | nodes = sorted(nodes, key=lambda n: n[0]) 88 | return DTree(nodes, data[0]) 89 | 90 | 91 | def _get_potentials(attrIndexes, continuousAttrIndexes, data, 92 | dataRowIndexes, outcomeIndex): 93 | uniqueAttributeValuePairs = { 94 | (attrIndex, data[rowIndex][attrIndex], operator.eq) 95 | for attrIndex in attrIndexes 96 | if attrIndex not in continuousAttrIndexes 97 | for rowIndex in dataRowIndexes} 98 | continuousAttributeValuePairs = _get_continuous_av_pairs( 99 | continuousAttrIndexes, data, dataRowIndexes) 100 | uniqueAttributeValuePairs |= continuousAttributeValuePairs 101 | potentials = sorted((-_get_bias(avPair, dataRowIndexes, data, 102 | outcomeIndex), 103 | avPair[0], avPair[1], avPair[2]) 104 | for avPair in uniqueAttributeValuePairs) 105 | return potentials 106 | 107 | 108 | def _get_continuous_av_pairs(continuousAttrIndexes, data, dataRowIndexes): 109 | avPairs = set() 110 | for attrIndex in continuousAttrIndexes: 111 | sortedAttrValues = [i for i in sorted( 112 | data[rowIndex][attrIndex] for rowIndex in dataRowIndexes)] 113 | indexes = _get_discontinuity_indexes( 114 | sortedAttrValues, 115 | max(math.sqrt( 116 | len(sortedAttrValues)), 117 | min(10, 118 | len(sortedAttrValues)))) 119 | for index in indexes: 120 | avPairs.add((attrIndex, sortedAttrValues[index], operator.gt)) 121 | return avPairs 122 | 123 | 124 | def _get_discontinuity_indexes(sortedAttrValues, maxIndexes): 125 | indexes = [] 126 | for i in _generate_discontinuity_indexes_center_out(sortedAttrValues): 127 | indexes.append(i) 128 | if len(indexes) >= maxIndexes: 129 | break 130 | return indexes 131 | 132 | 133 | def _generate_discontinuity_indexes_center_out(sortedAttrValues): 134 | center = len(sortedAttrValues) // 2 135 | left = center - 1 136 | right = center + 1 137 | while left >= 0 or right < len(sortedAttrValues): 138 | if left >= 0: 139 | if sortedAttrValues[left] != sortedAttrValues[left + 1]: 140 | yield left 141 | left -= 1 142 | if right < len(sortedAttrValues): 143 | if sortedAttrValues[right - 1] != sortedAttrValues[right]: 144 | yield right - 1 145 | right += 1 146 | 147 | 148 | class DTree: 149 | def __init__(self, nodes, attrNames): 150 | self._nodes = nodes 151 | self._attrNames = attrNames 152 | 153 | @staticmethod 154 | def _is_leaf(node): 155 | return len(node) == 2 156 | 157 | def __str__(self): 158 | s = '' 159 | for node in self._nodes: 160 | if self._is_leaf(node): 161 | s += '{}: {}\n'.format(node[0], node[1]) 162 | else: 163 | nodeId, attrIndex, attrValue, isMatch, nodeIdIfMatch, \ 164 | nodeIdIfNonMatch, matchCount, nonMatchCount = node 165 | s += '{0}: {1}{7}{2}, {5} Yes->{3}, {6} No->{4}\n'.format( 166 | nodeId, self._attrNames[attrIndex], attrValue, 167 | nodeIdIfMatch, nodeIdIfNonMatch, matchCount, 168 | nonMatchCount, '=' if isMatch == operator.eq else '>') 169 | return s 170 | 171 | def get_prediction(self, data): 172 | currentNode = self._nodes[0] 173 | while True: 174 | if self._is_leaf(currentNode): 175 | return currentNode[1] 176 | nodeId, attrIndex, attrValue, isMatch, nodeIdIfMatch, \ 177 | nodeIdIfNonMatch = currentNode[:6] 178 | currentNode = self._nodes[nodeIdIfMatch if 179 | isMatch(data[attrIndex], attrValue) else nodeIdIfNonMatch] 180 | -------------------------------------------------------------------------------- /ch03/test.py: -------------------------------------------------------------------------------- 1 | # File: test.py 2 | # from chapter 3 of _Tree-based Machine Learning Algorithms_ 3 | # 4 | # Author: Clinton Sheppard 5 | # Copyright (c) 2017 Clinton Sheppard 6 | # 7 | # Licensed under the Apache License, Version 2.0 (the "License"). 8 | # You may not use this file except in compliance with the License. 9 | # You may obtain a copy of the License at 10 | # http://www.apache.org/licenses/LICENSE-2.0 11 | # 12 | # Unless required by applicable law or agreed to in writing, software 13 | # distributed under the License is distributed on an "AS IS" BASIS, 14 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 15 | # implied. See the License for the specific language governing 16 | # permissions and limitations under the License. 17 | 18 | import dtree 19 | 20 | data = [['Name', 'Gender', 'Marital Status', 'Age', 'Born'], 21 | ['William', 'male', 'Married', 37, 'Germany'], 22 | ['Louise', 'female', 'Single', 18, 'Germany'], 23 | ['Minnie', 'female', 'Single', 16, 'Texas'], 24 | ['Emma', 'female', 'Single', 14, 'Texas'], 25 | ['Henry', 'male', 'Married', 47, 'Germany'], 26 | ['Theo', 'male', 'Single', 17, 'Texas'], 27 | ] 28 | 29 | outcomeLabel = 'Born' 30 | 31 | tree = dtree.build(data, outcomeLabel) 32 | print(tree) 33 | 34 | testData = ['Sophie', 'female', 'Single', 17] 35 | predicted = tree.get_prediction(testData) 36 | print("predicted: {}".format(predicted)) 37 | -------------------------------------------------------------------------------- /ch04/census.csv: -------------------------------------------------------------------------------- 1 | Name,Gender,Marital Status,Age,Relationship,Born 2 | August,male,Married,32,Head,Germany 3 | Minnie,female,Married,28,Wife,Texas 4 | Emma,female,Single,9,Daughter,Texas 5 | Theo,male,Single,3,Son,Texas 6 | William,male,Married,37,Head,Germany 7 | Sophie,female,Married,22,Wife,Germany 8 | Louise,female,Single,4,Daughter,Texas 9 | Minnie,female,Single,2,Daughter,Texas 10 | Emma,female,Single,1,Daughter,Texas 11 | Henry,male,Married,33,Head,Germany 12 | Henrietta,female,Married,28,Wife,Germany 13 | Henry,male,Single,9,Son,Texas 14 | Frank,male,Single,7,Son,Texas 15 | Hermann,male,Single,4,Son,Texas 16 | Louise,female,Single,3,Daughter,Texas 17 | Charles,male,Single,1,Son,Texas 18 | Hermann,male,Married,39,Head,Germany 19 | Dora,female,Married,31,Wife,Germany 20 | Hennie,female,Single,8,Daughter,Texas 21 | Lisette,female,Single,5,Daughter,Texas 22 | Fritz,male,Single,3,Son,Texas 23 | Minnie,female,Single,3,Daughter,Texas 24 | Charles,male,Married,68,Head,Germany 25 | Louise,female,Married,64,Wife,Germany 26 | Katie,female,Single,21,Daughter,Germany 27 | Charles,male,Single,18,Son,Germany 28 | Henry,male,Single,2,Nephew,Texas 29 | Horace,male,Married,27,Head,Texas 30 | Lucy,female,Married,25,Wife,Texas 31 | Henry,male,Married,61,Head,Germany 32 | Louise,female,Married,51,Wife,Germany 33 | Fritz,male,Single,18,Son,Germany 34 | Otto,male,Single,16,Son,Texas 35 | Bertha,female,Single,15,Daughter,Texas 36 | Nathlie,female,Single,10,Daughter,Texas 37 | Elsa,female,Single,8,Daughter,Texas 38 | August,male,Single,6,Son,Texas 39 | Henry,male,Single,2,Nephew,Texas 40 | William,male,Married,66,Head,Germany 41 | Minnie,female,Married,89,Wife,Germany 42 | Hermann,male,Married,43,Head,Germany 43 | Emily,female,Married,47,Wife,Germany 44 | Henry,male,Single,19,Son,Texas 45 | Olga,female,Single,18,Daughter,Texas 46 | Paul,male,Single,16,Son,Texas 47 | Ernst,male,Single,15,Son,Texas 48 | Emil,male,Single,12,Son,Texas 49 | Ed,male,Single,11,Son,Texas 50 | Otto,male,Single,9,Son,Texas 51 | Ella,female,Single,7,Daughter,Texas 52 | William,male,Married,47,Head,Germany 53 | Emily,female,Married,42,Wife,Germany 54 | Lena,female,Single,15,Daughter,Texas 55 | Christian,male,Single,14,Son,Texas 56 | Bertha,female,Single,12,Daughter,Texas 57 | Ella,female,Single,9,Daughter,Texas 58 | Mollie,female,Single,6,Daughter,Texas 59 | Hettie,female,Single,1,Daughter,Texas -------------------------------------------------------------------------------- /ch04/dtree.py: -------------------------------------------------------------------------------- 1 | # File: dtree.py 2 | # from chapter 4 of _Tree-based Machine Learning Algorithms_ 3 | # 4 | # Author: Clinton Sheppard 5 | # Copyright (c) 2017 Clinton Sheppard 6 | # 7 | # Licensed under the Apache License, Version 2.0 (the "License"). 8 | # You may not use this file except in compliance with the License. 9 | # You may obtain a copy of the License at 10 | # http://www.apache.org/licenses/LICENSE-2.0 11 | # 12 | # Unless required by applicable law or agreed to in writing, software 13 | # distributed under the License is distributed on an "AS IS" BASIS, 14 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 15 | # implied. See the License for the specific language governing 16 | # permissions and limitations under the License. 17 | 18 | from collections import Counter 19 | from numbers import Number 20 | import operator 21 | import math 22 | import csv 23 | import random 24 | 25 | 26 | def _get_bias(avPair, dataRowIndexes, data, outcomeIndex, minimumSubsetSize, 27 | validationRowIndexes): 28 | attrIndex, attrValue, isMatch = avPair 29 | if len(validationRowIndexes) > 0: 30 | validationMatchIndexes = {i for i in validationRowIndexes if 31 | isMatch(data[i][attrIndex], attrValue)} 32 | validationNonMatchIndexes = validationRowIndexes - \ 33 | validationMatchIndexes 34 | if len(validationMatchIndexes) == 0 or len( 35 | validationNonMatchIndexes) == 0: 36 | return -2 37 | matchIndexes = {i for i in dataRowIndexes if 38 | isMatch(data[i][attrIndex], attrValue)} 39 | nonMatchIndexes = dataRowIndexes - matchIndexes 40 | if len(matchIndexes) < minimumSubsetSize or len( 41 | nonMatchIndexes) < minimumSubsetSize: 42 | return -1 43 | matchOutcomes = {data[i][outcomeIndex] for i in matchIndexes} 44 | nonMatchOutcomes = {data[i][outcomeIndex] for i in nonMatchIndexes} 45 | numPureRows = (len(matchIndexes) if len(matchOutcomes) == 1 else 0) \ 46 | + (len(nonMatchIndexes) if len(nonMatchOutcomes) == 1 47 | else 0) 48 | percentPure = numPureRows / len(dataRowIndexes) 49 | 50 | numNonPureRows = len(dataRowIndexes) - numPureRows 51 | percentNonPure = 1 - percentPure 52 | split = 1 - abs(len(matchIndexes) - len(nonMatchIndexes)) / len( 53 | dataRowIndexes) - .001 54 | splitBias = split * percentNonPure if numNonPureRows > 0 else 0 55 | return splitBias + percentPure 56 | 57 | 58 | def build(data, outcomeLabel, continuousAttributes=None, 59 | minimumSubsetSizePercentage=0, validationPercentage=0): 60 | if validationPercentage > 0: 61 | validationPercentage /= 100 62 | validationCount = int(validationPercentage * len(data)) 63 | if minimumSubsetSizePercentage > 0: 64 | minimumSubsetSizePercentage /= 100 65 | minimumSubsetSize = int(minimumSubsetSizePercentage * len(data)) 66 | attrIndexes = [index for index, label in enumerate(data[0]) if 67 | label != outcomeLabel] 68 | outcomeIndex = data[0].index(outcomeLabel) 69 | continuousAttrIndexes = set() 70 | if continuousAttributes is not None: 71 | continuousAttrIndexes = {data[0].index(label) for label in 72 | continuousAttributes} 73 | if len(continuousAttrIndexes) != len(continuousAttributes): 74 | raise Exception( 75 | 'One or more continuous column names are duplicates.') 76 | else: 77 | for attrIndex in attrIndexes: 78 | uniqueValues = {row[attrIndex] for rowIndex, row in 79 | enumerate(data) if rowIndex > 0} 80 | numericValues = {value for value in uniqueValues if 81 | isinstance(value, Number)} 82 | if len(uniqueValues) == len(numericValues): 83 | continuousAttrIndexes.add(attrIndex) 84 | 85 | nodes = [] 86 | lastNodeNumber = 0 87 | dataIndexes = {i for i in range(1, len(data))} 88 | validationIndexes = set() 89 | if validationCount > 0: 90 | validationIndexes = set( 91 | random.sample(range(1, len(data)), validationCount)) 92 | dataIndexes -= validationIndexes 93 | workQueue = [(-1, lastNodeNumber, dataIndexes, validationIndexes)] 94 | while len(workQueue) > 0: 95 | parentNodeId, nodeId, dataRowIndexes, validationRowIndexes = \ 96 | workQueue.pop() 97 | uniqueOutcomes = Counter( 98 | data[i][outcomeIndex] for i in dataRowIndexes).most_common() 99 | if len(uniqueOutcomes) == 1: 100 | nodes.append((nodeId, uniqueOutcomes.pop(0)[0])) 101 | continue 102 | potentials = _get_potentials(attrIndexes, continuousAttrIndexes, 103 | data, dataRowIndexes, outcomeIndex, 104 | minimumSubsetSize, 105 | validationRowIndexes) 106 | if len(potentials) == 0 or potentials[0][0] > 0: 107 | nodes.append((nodeId, [(n[0], n[1] / len(dataRowIndexes)) 108 | for n in uniqueOutcomes])) 109 | continue 110 | attrIndex, attrValue, isMatch = potentials[0][1:] 111 | matches = {rowIndex for rowIndex in dataRowIndexes if 112 | isMatch(data[rowIndex][attrIndex], attrValue)} 113 | nonMatches = dataRowIndexes - matches 114 | validationMatches = { 115 | rowIndex for rowIndex in validationRowIndexes if 116 | isMatch(data[rowIndex][attrIndex], attrValue)} 117 | nonValidationMatches = validationRowIndexes - validationMatches 118 | lastNodeNumber += 1 119 | matchId = lastNodeNumber 120 | workQueue.append((nodeId, matchId, matches, validationMatches)) 121 | lastNodeNumber += 1 122 | nonMatchId = lastNodeNumber 123 | workQueue.append((nodeId, nonMatchId, nonMatches, 124 | nonValidationMatches)) 125 | nodes.append((nodeId, attrIndex, attrValue, isMatch, matchId, 126 | nonMatchId, len(matches), len(nonMatches))) 127 | nodes = sorted(nodes, key=lambda n: n[0]) 128 | return DTree(nodes, data[0]) 129 | 130 | 131 | def _get_potentials(attrIndexes, continuousAttrIndexes, data, 132 | dataRowIndexes, outcomeIndex, minimumSubsetSize, 133 | validationRowIndexes): 134 | uniqueAttributeValuePairs = { 135 | (attrIndex, data[rowIndex][attrIndex], operator.eq) 136 | for attrIndex in attrIndexes 137 | if attrIndex not in continuousAttrIndexes 138 | for rowIndex in dataRowIndexes} 139 | continuousAttributeValuePairs = _get_continuous_av_pairs( 140 | continuousAttrIndexes, data, dataRowIndexes) 141 | uniqueAttributeValuePairs |= continuousAttributeValuePairs 142 | potentials = sorted((-_get_bias(avPair, dataRowIndexes, data, 143 | outcomeIndex, minimumSubsetSize, 144 | validationRowIndexes), 145 | avPair[0], avPair[1], avPair[2]) 146 | for avPair in uniqueAttributeValuePairs) 147 | return potentials 148 | 149 | def _get_continuous_av_pairs(continuousAttrIndexes, data, dataRowIndexes): 150 | avPairs = set() 151 | for attrIndex in continuousAttrIndexes: 152 | sortedAttrValues = [i for i in sorted( 153 | data[rowIndex][attrIndex] for rowIndex in dataRowIndexes)] 154 | indexes = _get_discontinuity_indexes( 155 | sortedAttrValues, 156 | max(math.sqrt( 157 | len(sortedAttrValues)), 158 | min(10, 159 | len(sortedAttrValues)))) 160 | for index in indexes: 161 | avPairs.add((attrIndex, sortedAttrValues[index], operator.gt)) 162 | return avPairs 163 | 164 | 165 | def _get_discontinuity_indexes(sortedAttrValues, maxIndexes): 166 | indexes = [] 167 | for i in _generate_discontinuity_indexes_center_out(sortedAttrValues): 168 | indexes.append(i) 169 | if len(indexes) >= maxIndexes: 170 | break 171 | return indexes 172 | 173 | 174 | def _generate_discontinuity_indexes_center_out(sortedAttrValues): 175 | center = len(sortedAttrValues) // 2 176 | left = center - 1 177 | right = center + 1 178 | while left >= 0 or right < len(sortedAttrValues): 179 | if left >= 0: 180 | if sortedAttrValues[left] != sortedAttrValues[left + 1]: 181 | yield left 182 | left -= 1 183 | if right < len(sortedAttrValues): 184 | if sortedAttrValues[right - 1] != sortedAttrValues[right]: 185 | yield right - 1 186 | right += 1 187 | 188 | 189 | def read_csv(filepath): 190 | with open(filepath, 'r') as f: 191 | reader = csv.reader(f) 192 | data = list(reader) 193 | return data 194 | 195 | 196 | def prepare_data(data, numericColumnLabels=None): 197 | if numericColumnLabels is not None and len(numericColumnLabels) > 0: 198 | numericColumnIndexes = [data[0].index(label) for label in 199 | numericColumnLabels] 200 | for rowIndex, row in enumerate(data): 201 | if rowIndex == 0: 202 | continue 203 | for numericIndex in numericColumnIndexes: 204 | f = float(data[rowIndex][numericIndex]) if len( 205 | data[rowIndex][numericIndex]) > 0 else 0 206 | i = int(f) 207 | data[rowIndex][numericIndex] = i if i == f else f 208 | return data 209 | 210 | 211 | class DTree: 212 | def __init__(self, nodes, attrNames): 213 | self._nodes = nodes 214 | self._attrNames = attrNames 215 | 216 | @staticmethod 217 | def _is_leaf(node): 218 | return len(node) == 2 219 | 220 | def __str__(self): 221 | s = '' 222 | for node in self._nodes: 223 | if self._is_leaf(node): 224 | s += '{}: {}\n'.format(node[0], node[1]) 225 | else: 226 | nodeId, attrIndex, attrValue, isMatch, nodeIdIfMatch, \ 227 | nodeIdIfNonMatch, matchCount, nonMatchCount = node 228 | s += '{0}: {1}{7}{2}, {5} Yes->{3}, {6} No->{4}\n'.format( 229 | nodeId, self._attrNames[attrIndex], attrValue, 230 | nodeIdIfMatch, nodeIdIfNonMatch, matchCount, 231 | nonMatchCount, '=' if isMatch == operator.eq else '>') 232 | return s 233 | 234 | def get_prediction(self, data): 235 | currentNode = self._nodes[0] 236 | while True: 237 | if self._is_leaf(currentNode): 238 | node = currentNode[1] 239 | if type(node) is not list: 240 | return node 241 | randPercent = random.uniform(0, 1) 242 | total = 0 243 | for outcome, percentage in node: 244 | total += percentage 245 | if total > randPercent: 246 | return outcome 247 | return node[-1][0] 248 | nodeId, attrIndex, attrValue, isMatch, nodeIdIfMatch, \ 249 | nodeIdIfNonMatch = currentNode[:6] 250 | currentNode = self._nodes[nodeIdIfMatch if 251 | isMatch(data[attrIndex], attrValue) else nodeIdIfNonMatch] 252 | -------------------------------------------------------------------------------- /ch04/test.py: -------------------------------------------------------------------------------- 1 | # File: test.py 2 | # from chapter 4 of _Tree-based Machine Learning Algorithms_ 3 | # 4 | # Author: Clinton Sheppard 5 | # Copyright (c) 2017 Clinton Sheppard 6 | # 7 | # Licensed under the Apache License, Version 2.0 (the "License"). 8 | # You may not use this file except in compliance with the License. 9 | # You may obtain a copy of the License at 10 | # http://www.apache.org/licenses/LICENSE-2.0 11 | # 12 | # Unless required by applicable law or agreed to in writing, software 13 | # distributed under the License is distributed on an "AS IS" BASIS, 14 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 15 | # implied. See the License for the specific language governing 16 | # permissions and limitations under the License. 17 | 18 | import dtree 19 | 20 | data = dtree.read_csv('census.csv') 21 | data = dtree.prepare_data(data, ['Age']) 22 | 23 | outcomeLabel = 'Born' 24 | 25 | tree = dtree.build(data, outcomeLabel, validationPercentage=6) 26 | print(tree) 27 | 28 | testData = ['Elizabeth', 'female', 'Married', 19, 'Daughter'] 29 | predicted = tree.get_prediction(testData) 30 | print("predicted: {}".format(predicted)) 31 | -------------------------------------------------------------------------------- /ch05/dtree.py: -------------------------------------------------------------------------------- 1 | # File: dtree.py 2 | # from chapter 5 of _Tree-based Machine Learning Algorithms_ 3 | # 4 | # Author: Clinton Sheppard 5 | # Copyright (c) 2017 Clinton Sheppard 6 | # 7 | # Licensed under the Apache License, Version 2.0 (the "License"). 8 | # You may not use this file except in compliance with the License. 9 | # You may obtain a copy of the License at 10 | # http://www.apache.org/licenses/LICENSE-2.0 11 | # 12 | # Unless required by applicable law or agreed to in writing, software 13 | # distributed under the License is distributed on an "AS IS" BASIS, 14 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 15 | # implied. See the License for the specific language governing 16 | # permissions and limitations under the License. 17 | 18 | from collections import Counter 19 | from numbers import Number 20 | import operator 21 | import math 22 | import csv 23 | import random 24 | 25 | 26 | def _get_bias(avPair, dataRowIndexes, data, outcomeIndex, minimumSubsetSize, 27 | validationRowIndexes): 28 | attrIndex, attrValue, isMatch = avPair 29 | if len(validationRowIndexes) > 0: 30 | validationMatchIndexes = {i for i in validationRowIndexes if 31 | isMatch(data[i][attrIndex], attrValue)} 32 | validationNonMatchIndexes = validationRowIndexes - \ 33 | validationMatchIndexes 34 | if len(validationMatchIndexes) == 0 or len( 35 | validationNonMatchIndexes) == 0: 36 | return -2 37 | matchIndexes = {i for i in dataRowIndexes if 38 | isMatch(data[i][attrIndex], attrValue)} 39 | nonMatchIndexes = dataRowIndexes - matchIndexes 40 | if len(matchIndexes) < minimumSubsetSize or len( 41 | nonMatchIndexes) < minimumSubsetSize: 42 | return -1 43 | matchOutcomes = {data[i][outcomeIndex] for i in matchIndexes} 44 | nonMatchOutcomes = {data[i][outcomeIndex] for i in nonMatchIndexes} 45 | numPureRows = (len(matchIndexes) if len(matchOutcomes) == 1 else 0) \ 46 | + (len(nonMatchIndexes) if len(nonMatchOutcomes) == 1 47 | else 0) 48 | percentPure = numPureRows / len(dataRowIndexes) 49 | 50 | numNonPureRows = len(dataRowIndexes) - numPureRows 51 | percentNonPure = 1 - percentPure 52 | split = 1 - abs(len(matchIndexes) - len(nonMatchIndexes)) / len( 53 | dataRowIndexes) - .001 54 | splitBias = split * percentNonPure if numNonPureRows > 0 else 0 55 | return splitBias + percentPure 56 | 57 | 58 | def build(data, outcomeLabel, continuousAttributes=None, 59 | minimumSubsetSizePercentage=0, validationPercentage=0, 60 | dataIndexes=None, attrIndexes=None): 61 | if validationPercentage > 0: 62 | validationPercentage /= 100 63 | validationCount = int(validationPercentage * 64 | (len(data) if dataIndexes is None else len( 65 | dataIndexes))) 66 | if minimumSubsetSizePercentage > 0: 67 | minimumSubsetSizePercentage /= 100 68 | minimumSubsetSize = int(minimumSubsetSizePercentage * 69 | (len(data) if dataIndexes is None else len( 70 | dataIndexes))) 71 | if attrIndexes is None: 72 | attrIndexes = [index for index, label in enumerate(data[0]) if 73 | label != outcomeLabel] 74 | outcomeIndex = data[0].index(outcomeLabel) 75 | continuousAttrIndexes = set() 76 | if continuousAttributes is not None: 77 | continuousAttrIndexes = {data[0].index(label) for label in 78 | continuousAttributes} 79 | if len(continuousAttrIndexes) != len(continuousAttributes): 80 | raise Exception( 81 | 'One or more continuous column names are duplicates.') 82 | else: 83 | for attrIndex in attrIndexes: 84 | uniqueValues = {row[attrIndex] for rowIndex, row in 85 | enumerate(data) if rowIndex > 0} 86 | numericValues = {value for value in uniqueValues if 87 | isinstance(value, Number)} 88 | if len(uniqueValues) == len(numericValues): 89 | continuousAttrIndexes.add(attrIndex) 90 | 91 | nodes = [] 92 | lastNodeNumber = 0 93 | if dataIndexes is None: 94 | dataIndexes = {i for i in range(1, len(data))} 95 | elif not isinstance(dataIndexes, set): 96 | dataIndexes = {i for i in dataIndexes} 97 | validationIndexes = set() 98 | if validationCount > 0: 99 | validationIndexes = set( 100 | random.sample([i for i in dataIndexes], validationCount)) 101 | dataIndexes -= validationIndexes 102 | workQueue = [(-1, lastNodeNumber, dataIndexes, validationIndexes)] 103 | while len(workQueue) > 0: 104 | parentNodeId, nodeId, dataRowIndexes, validationRowIndexes = \ 105 | workQueue.pop() 106 | uniqueOutcomes = Counter( 107 | data[i][outcomeIndex] for i in dataRowIndexes).most_common() 108 | if len(uniqueOutcomes) == 1: 109 | nodes.append((nodeId, uniqueOutcomes.pop(0)[0])) 110 | continue 111 | potentials = _get_potentials(attrIndexes, continuousAttrIndexes, 112 | data, dataRowIndexes, outcomeIndex, 113 | minimumSubsetSize, 114 | validationRowIndexes) 115 | if len(potentials) == 0 or potentials[0][0] > 0: 116 | nodes.append((nodeId, [(n[0], n[1] / len(dataRowIndexes)) 117 | for n in uniqueOutcomes])) 118 | continue 119 | attrIndex, attrValue, isMatch = potentials[0][1:] 120 | matches = {rowIndex for rowIndex in dataRowIndexes if 121 | isMatch(data[rowIndex][attrIndex], attrValue)} 122 | nonMatches = dataRowIndexes - matches 123 | validationMatches = { 124 | rowIndex for rowIndex in validationRowIndexes if 125 | isMatch(data[rowIndex][attrIndex], attrValue)} 126 | nonValidationMatches = validationRowIndexes - validationMatches 127 | lastNodeNumber += 1 128 | matchId = lastNodeNumber 129 | workQueue.append((nodeId, matchId, matches, validationMatches)) 130 | lastNodeNumber += 1 131 | nonMatchId = lastNodeNumber 132 | workQueue.append((nodeId, nonMatchId, nonMatches, 133 | nonValidationMatches)) 134 | nodes.append((nodeId, attrIndex, attrValue, isMatch, matchId, 135 | nonMatchId, len(matches), len(nonMatches))) 136 | nodes = sorted(nodes, key=lambda n: n[0]) 137 | return DTree(nodes, data[0]) 138 | 139 | 140 | def _get_potentials(attrIndexes, continuousAttrIndexes, data, 141 | dataRowIndexes, outcomeIndex, minimumSubsetSize, 142 | validationRowIndexes): 143 | uniqueAttributeValuePairs = { 144 | (attrIndex, data[rowIndex][attrIndex], operator.eq) 145 | for attrIndex in attrIndexes 146 | if attrIndex not in continuousAttrIndexes 147 | for rowIndex in dataRowIndexes} 148 | continuousAttributeValuePairs = _get_continuous_av_pairs( 149 | continuousAttrIndexes, data, dataRowIndexes) 150 | uniqueAttributeValuePairs |= continuousAttributeValuePairs 151 | potentials = sorted((-_get_bias(avPair, dataRowIndexes, data, 152 | outcomeIndex, minimumSubsetSize, 153 | validationRowIndexes), 154 | avPair[0], avPair[1], avPair[2]) 155 | for avPair in uniqueAttributeValuePairs) 156 | return potentials 157 | 158 | def _get_continuous_av_pairs(continuousAttrIndexes, data, dataRowIndexes): 159 | avPairs = set() 160 | for attrIndex in continuousAttrIndexes: 161 | sortedAttrValues = [i for i in sorted( 162 | data[rowIndex][attrIndex] for rowIndex in dataRowIndexes)] 163 | indexes = _get_discontinuity_indexes( 164 | sortedAttrValues, 165 | max(math.sqrt( 166 | len(sortedAttrValues)), 167 | min(10, 168 | len(sortedAttrValues)))) 169 | for index in indexes: 170 | avPairs.add((attrIndex, sortedAttrValues[index], operator.gt)) 171 | return avPairs 172 | 173 | 174 | def _get_discontinuity_indexes(sortedAttrValues, maxIndexes): 175 | indexes = [] 176 | for i in _generate_discontinuity_indexes_center_out(sortedAttrValues): 177 | indexes.append(i) 178 | if len(indexes) >= maxIndexes: 179 | break 180 | return indexes 181 | 182 | 183 | def _generate_discontinuity_indexes_center_out(sortedAttrValues): 184 | center = len(sortedAttrValues) // 2 185 | left = center - 1 186 | right = center + 1 187 | while left >= 0 or right < len(sortedAttrValues): 188 | if left >= 0: 189 | if sortedAttrValues[left] != sortedAttrValues[left + 1]: 190 | yield left 191 | left -= 1 192 | if right < len(sortedAttrValues): 193 | if sortedAttrValues[right - 1] != sortedAttrValues[right]: 194 | yield right - 1 195 | right += 1 196 | 197 | 198 | def read_csv(filepath): 199 | with open(filepath, 'r') as f: 200 | reader = csv.reader(f) 201 | data = list(reader) 202 | return data 203 | 204 | 205 | def prepare_data(data, numericColumnLabels=None): 206 | if numericColumnLabels is not None and len(numericColumnLabels) > 0: 207 | numericColumnIndexes = [data[0].index(label) for label in 208 | numericColumnLabels] 209 | for rowIndex, row in enumerate(data): 210 | if rowIndex == 0: 211 | continue 212 | for numericIndex in numericColumnIndexes: 213 | f = float(data[rowIndex][numericIndex]) if len( 214 | data[rowIndex][numericIndex]) > 0 else 0 215 | i = int(f) 216 | data[rowIndex][numericIndex] = i if i == f else f 217 | return data 218 | 219 | 220 | class DTree: 221 | def __init__(self, nodes, attrNames): 222 | self._nodes = nodes 223 | self._attrNames = attrNames 224 | 225 | @staticmethod 226 | def _is_leaf(node): 227 | return len(node) == 2 228 | 229 | def __str__(self): 230 | s = '' 231 | for node in self._nodes: 232 | if self._is_leaf(node): 233 | s += '{}: {}\n'.format(node[0], node[1]) 234 | else: 235 | nodeId, attrIndex, attrValue, isMatch, nodeIdIfMatch, \ 236 | nodeIdIfNonMatch, matchCount, nonMatchCount = node 237 | s += '{0}: {1}{7}{2}, {5} Yes->{3}, {6} No->{4}\n'.format( 238 | nodeId, self._attrNames[attrIndex], attrValue, 239 | nodeIdIfMatch, nodeIdIfNonMatch, matchCount, 240 | nonMatchCount, '=' if isMatch == operator.eq else '>') 241 | return s 242 | 243 | def get_prediction(self, data): 244 | currentNode = self._nodes[0] 245 | while True: 246 | if self._is_leaf(currentNode): 247 | node = currentNode[1] 248 | if type(node) is not list: 249 | return node 250 | randPercent = random.uniform(0, 1) 251 | total = 0 252 | for outcome, percentage in node: 253 | total += percentage 254 | if total > randPercent: 255 | return outcome 256 | return node[-1][0] 257 | nodeId, attrIndex, attrValue, isMatch, nodeIdIfMatch, \ 258 | nodeIdIfNonMatch = currentNode[:6] 259 | currentNode = self._nodes[nodeIdIfMatch if 260 | isMatch(data[attrIndex], attrValue) else nodeIdIfNonMatch] 261 | -------------------------------------------------------------------------------- /ch05/forest.py: -------------------------------------------------------------------------------- 1 | # File: forest.py 2 | # from chapter 5 of _Tree-based Machine Learning Algorithms_ 3 | # 4 | # Author: Clinton Sheppard 5 | # Copyright (c) 2017 Clinton Sheppard 6 | # 7 | # Licensed under the Apache License, Version 2.0 (the "License"). 8 | # You may not use this file except in compliance with the License. 9 | # You may obtain a copy of the License at 10 | # http://www.apache.org/licenses/LICENSE-2.0 11 | # 12 | # Unless required by applicable law or agreed to in writing, software 13 | # distributed under the License is distributed on an "AS IS" BASIS, 14 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 15 | # implied. See the License for the specific language governing 16 | # permissions and limitations under the License. 17 | 18 | from collections import Counter 19 | import dtree 20 | import math 21 | import random 22 | import statistics 23 | 24 | 25 | class Forest: 26 | def __init__(self, data, outcomeLabel, continuousAttributes=None, 27 | dataRowIndexes=None, columnsNamesToIgnore=None): 28 | self.data = data 29 | self.outcomeLabel = outcomeLabel 30 | self.continuousAttributes = continuousAttributes \ 31 | if columnsNamesToIgnore is None \ 32 | else [i for i in continuousAttributes if 33 | i not in columnsNamesToIgnore] 34 | self.numRows = math.ceil(math.sqrt( 35 | len(data) if dataRowIndexes is None else len(dataRowIndexes))) 36 | self.outcomeIndex = data[0].index(outcomeLabel) 37 | columnIdsToIgnore = set() if columnsNamesToIgnore is None else set( 38 | data[0].index(s) for s in columnsNamesToIgnore) 39 | columnIdsToIgnore.add(self.outcomeIndex) 40 | self.attrIndexesExceptOutcomeIndex = [i for i in range(0, len(data[0])) 41 | if i not in columnIdsToIgnore] 42 | self.numAttributes = math.ceil( 43 | math.sqrt(len(self.attrIndexesExceptOutcomeIndex))) 44 | self.dataRowIndexes = range(1, len( 45 | data)) if dataRowIndexes is None else dataRowIndexes 46 | self.numTrees = 200 47 | self.populate() 48 | 49 | def _build_tree(self): 50 | return dtree.build(self.data, self.outcomeLabel, 51 | continuousAttributes=self.continuousAttributes, 52 | dataIndexes={i for i in random.sample( 53 | self.dataRowIndexes, self.numRows)}, 54 | attrIndexes=[ 55 | i for i in random.sample( 56 | self.attrIndexesExceptOutcomeIndex, 57 | self.numAttributes)]) 58 | 59 | def populate(self): 60 | self._trees = [self._build_tree() for _ in range(0, self.numTrees)] 61 | 62 | def get_prediction(self, dataItem): 63 | sorted_predictions = self._get_predictions(dataItem) 64 | return sorted_predictions[0][0] 65 | 66 | def _get_predictions(self, dataItem): 67 | predictions = [t.get_prediction(dataItem) for t in self._trees] 68 | return Counter(p for p in predictions).most_common() 69 | 70 | 71 | class Benchmark: 72 | @staticmethod 73 | def run(f): 74 | results = [] 75 | for i in range(100): 76 | result = f() 77 | results.append(result) 78 | if i < 10 or i % 10 == 9: 79 | mean = statistics.mean(results) 80 | print("{} {:3.2f} {:3.2f}".format( 81 | 1 + i, mean, 82 | statistics.stdev(results, mean) if i > 1 else 0)) -------------------------------------------------------------------------------- /ch05/test.py: -------------------------------------------------------------------------------- 1 | # File: test.py 2 | # from chapter 5 of _Tree-based Machine Learning Algorithms_ 3 | # 4 | # Author: Clinton Sheppard 5 | # Copyright (c) 2017 Clinton Sheppard 6 | # 7 | # Licensed under the Apache License, Version 2.0 (the "License"). 8 | # You may not use this file except in compliance with the License. 9 | # You may obtain a copy of the License at 10 | # http://www.apache.org/licenses/LICENSE-2.0 11 | # 12 | # Unless required by applicable law or agreed to in writing, software 13 | # distributed under the License is distributed on an "AS IS" BASIS, 14 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 15 | # implied. See the License for the specific language governing 16 | # permissions and limitations under the License. 17 | 18 | from collections import Counter 19 | from forest import Forest 20 | import dtree 21 | 22 | 23 | data = dtree.read_csv('..\ch04\census.csv') 24 | continuousColumns = ['Age'] 25 | data = dtree.prepare_data(data, continuousColumns) 26 | outcomeLabel = 'Born' 27 | 28 | forest = Forest(data, outcomeLabel, continuousColumns) 29 | testData = ['Elizabeth', 'female', 'Married', 16, 'Daughter'] 30 | predicted = forest.get_prediction(testData) 31 | print("predicted: {}".format(predicted)) 32 | 33 | forest = Forest(data, outcomeLabel, continuousColumns) 34 | predictions = [] 35 | for _ in range(0, 100): 36 | predictions.append(forest.get_prediction(testData)) 37 | forest.populate() 38 | counts = Counter(predictions) 39 | print("predictions: {}".format(counts.most_common())) -------------------------------------------------------------------------------- /ch05/titanic.py: -------------------------------------------------------------------------------- 1 | # File: titanic.py 2 | # from chapter 5 of _Tree-based Machine Learning Algorithms_ 3 | # 4 | # Author: Clinton Sheppard 5 | # Copyright (c) 2017 Clinton Sheppard 6 | # 7 | # Licensed under the Apache License, Version 2.0 (the "License"). 8 | # You may not use this file except in compliance with the License. 9 | # You may obtain a copy of the License at 10 | # http://www.apache.org/licenses/LICENSE-2.0 11 | # 12 | # Unless required by applicable law or agreed to in writing, software 13 | # distributed under the License is distributed on an "AS IS" BASIS, 14 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 15 | # implied. See the License for the specific language governing 16 | # permissions and limitations under the License. 17 | 18 | from forest import Forest 19 | from forest import Benchmark 20 | import dtree 21 | import random 22 | 23 | 24 | continuousColumns = ['PassengerId', 'Pclass', 'Age', 'SibSp', 'Parch', 'Fare'] 25 | data = dtree.read_csv('train.csv') 26 | data = dtree.prepare_data(data, continuousColumns) 27 | outcomeLabel = 'Survived' 28 | columnsToIgnore = ['PassengerId', 'Name', 'Ticket', 'Cabin'] 29 | 30 | 31 | def predict(): 32 | trainingRowIds = random.sample(range(1, len(data)), int(.8 * len(data))) 33 | forest = Forest(data, outcomeLabel, continuousColumns, trainingRowIds, columnsToIgnore) 34 | correct = sum(1 for rowId, row in enumerate(data) if 35 | rowId > 0 and 36 | rowId not in trainingRowIds and 37 | forest.get_prediction(row) == row[1]) 38 | return 100 * correct / (len(data) - 1 - len(trainingRowIds)) 39 | 40 | Benchmark.run(predict) -------------------------------------------------------------------------------- /ch06/dtree.py: -------------------------------------------------------------------------------- 1 | # File: dtree.py 2 | # from chapter 6 of _Tree-based Machine Learning Algorithms_ 3 | # 4 | # Author: Clinton Sheppard 5 | # Copyright (c) 2017 Clinton Sheppard 6 | # 7 | # Licensed under the Apache License, Version 2.0 (the "License"). 8 | # You may not use this file except in compliance with the License. 9 | # You may obtain a copy of the License at 10 | # http://www.apache.org/licenses/LICENSE-2.0 11 | # 12 | # Unless required by applicable law or agreed to in writing, software 13 | # distributed under the License is distributed on an "AS IS" BASIS, 14 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 15 | # implied. See the License for the specific language governing 16 | # permissions and limitations under the License. 17 | 18 | from collections import Counter 19 | from numbers import Number 20 | import operator 21 | import math 22 | import csv 23 | import random 24 | 25 | 26 | def _get_bias(avPair, dataRowIndexes, data, outcomeIndex, minimumSubsetSize, 27 | validationRowIndexes): 28 | attrIndex, attrValue, isMatch = avPair 29 | if len(validationRowIndexes) > 0: 30 | validationMatchIndexes = {i for i in validationRowIndexes if 31 | isMatch(data[i][attrIndex], attrValue)} 32 | validationNonMatchIndexes = validationRowIndexes - \ 33 | validationMatchIndexes 34 | if len(validationMatchIndexes) == 0 or len( 35 | validationNonMatchIndexes) == 0: 36 | return -2 37 | matchIndexes = {i for i in dataRowIndexes if 38 | isMatch(data[i][attrIndex], attrValue)} 39 | nonMatchIndexes = dataRowIndexes - matchIndexes 40 | if len(matchIndexes) < minimumSubsetSize or len( 41 | nonMatchIndexes) < minimumSubsetSize: 42 | return -1 43 | matchOutcomes = {data[i][outcomeIndex] for i in matchIndexes} 44 | nonMatchOutcomes = {data[i][outcomeIndex] for i in nonMatchIndexes} 45 | numPureRows = (len(matchIndexes) if len(matchOutcomes) == 1 else 0) \ 46 | + (len(nonMatchIndexes) if len(nonMatchOutcomes) == 1 47 | else 0) 48 | percentPure = numPureRows / len(dataRowIndexes) 49 | 50 | numNonPureRows = len(dataRowIndexes) - numPureRows 51 | percentNonPure = 1 - percentPure 52 | split = 1 - abs(len(matchIndexes) - len(nonMatchIndexes)) / len( 53 | dataRowIndexes) - .001 54 | splitBias = split * percentNonPure if numNonPureRows > 0 else 0 55 | return splitBias + percentPure 56 | 57 | 58 | def build(data, outcomeLabel, continuousAttributes=None, 59 | minimumSubsetSizePercentage=0, validationPercentage=0, 60 | dataIndexes=None, attrIndexes=None): 61 | if validationPercentage > 0: 62 | validationPercentage /= 100 63 | validationCount = int(validationPercentage * 64 | (len(data) if dataIndexes is None else len( 65 | dataIndexes))) 66 | if minimumSubsetSizePercentage > 0: 67 | minimumSubsetSizePercentage /= 100 68 | minimumSubsetSize = int(minimumSubsetSizePercentage * 69 | (len(data) if dataIndexes is None else len( 70 | dataIndexes))) 71 | if attrIndexes is None: 72 | attrIndexes = [index for index, label in enumerate(data[0]) if 73 | label != outcomeLabel] 74 | outcomeIndex = data[0].index(outcomeLabel) 75 | continuousAttrIndexes = set() 76 | if continuousAttributes is not None: 77 | continuousAttrIndexes = {data[0].index(label) for label in 78 | continuousAttributes} 79 | if len(continuousAttrIndexes) != len(continuousAttributes): 80 | raise Exception( 81 | 'One or more continuous column names are duplicates.') 82 | else: 83 | for attrIndex in attrIndexes: 84 | uniqueValues = {row[attrIndex] for rowIndex, row in 85 | enumerate(data) if rowIndex > 0} 86 | numericValues = {value for value in uniqueValues if 87 | isinstance(value, Number)} 88 | if len(uniqueValues) == len(numericValues): 89 | continuousAttrIndexes.add(attrIndex) 90 | 91 | if outcomeIndex in continuousAttrIndexes: 92 | continuousAttrIndexes.remove(outcomeIndex) 93 | outcomeIsContinuous = True 94 | else: 95 | outcomeIsContinuous = False 96 | 97 | nodes = [] 98 | lastNodeNumber = 0 99 | if dataIndexes is None: 100 | dataIndexes = {i for i in range(1, len(data))} 101 | elif not isinstance(dataIndexes, set): 102 | dataIndexes = {i for i in dataIndexes} 103 | validationIndexes = set() 104 | if validationCount > 0: 105 | validationIndexes = set( 106 | random.sample([i for i in dataIndexes], validationCount)) 107 | dataIndexes -= validationIndexes 108 | workQueue = [(-1, lastNodeNumber, dataIndexes, validationIndexes)] 109 | while len(workQueue) > 0: 110 | parentNodeId, nodeId, dataRowIndexes, validationRowIndexes = \ 111 | workQueue.pop() 112 | uniqueOutcomes = Counter( 113 | data[i][outcomeIndex] for i in dataRowIndexes).most_common() 114 | if len(uniqueOutcomes) == 1: 115 | nodes.append((nodeId, uniqueOutcomes.pop(0)[0])) 116 | continue 117 | potentials = _get_potentials(attrIndexes, continuousAttrIndexes, 118 | data, dataRowIndexes, outcomeIndex, 119 | minimumSubsetSize, 120 | validationRowIndexes) 121 | if len(potentials) == 0 or potentials[0][0] > 0: 122 | nodes.append((nodeId, [(n[0], n[1] / len(dataRowIndexes)) 123 | for n in uniqueOutcomes])) 124 | continue 125 | attrIndex, attrValue, isMatch = potentials[0][1:] 126 | matches = {rowIndex for rowIndex in dataRowIndexes if 127 | isMatch(data[rowIndex][attrIndex], attrValue)} 128 | nonMatches = dataRowIndexes - matches 129 | validationMatches = { 130 | rowIndex for rowIndex in validationRowIndexes if 131 | isMatch(data[rowIndex][attrIndex], attrValue)} 132 | nonValidationMatches = validationRowIndexes - validationMatches 133 | lastNodeNumber += 1 134 | matchId = lastNodeNumber 135 | workQueue.append((nodeId, matchId, matches, validationMatches)) 136 | lastNodeNumber += 1 137 | nonMatchId = lastNodeNumber 138 | workQueue.append((nodeId, nonMatchId, nonMatches, 139 | nonValidationMatches)) 140 | nodes.append((nodeId, attrIndex, attrValue, isMatch, matchId, 141 | nonMatchId, len(matches), len(nonMatches))) 142 | nodes = sorted(nodes, key=lambda n: n[0]) 143 | return DTree(nodes, data[0], outcomeIsContinuous) 144 | 145 | 146 | def _get_potentials(attrIndexes, continuousAttrIndexes, data, 147 | dataRowIndexes, outcomeIndex, minimumSubsetSize, 148 | validationRowIndexes): 149 | uniqueAttributeValuePairs = { 150 | (attrIndex, data[rowIndex][attrIndex], operator.eq) 151 | for attrIndex in attrIndexes 152 | if attrIndex not in continuousAttrIndexes 153 | for rowIndex in dataRowIndexes} 154 | continuousAttributeValuePairs = _get_continuous_av_pairs( 155 | continuousAttrIndexes, data, dataRowIndexes) 156 | uniqueAttributeValuePairs |= continuousAttributeValuePairs 157 | potentials = sorted((-_get_bias(avPair, dataRowIndexes, data, 158 | outcomeIndex, minimumSubsetSize, 159 | validationRowIndexes), 160 | avPair[0], avPair[1], avPair[2]) 161 | for avPair in uniqueAttributeValuePairs) 162 | return potentials 163 | 164 | def _get_continuous_av_pairs(continuousAttrIndexes, data, dataRowIndexes): 165 | avPairs = set() 166 | for attrIndex in continuousAttrIndexes: 167 | sortedAttrValues = [i for i in sorted( 168 | data[rowIndex][attrIndex] for rowIndex in dataRowIndexes)] 169 | indexes = _get_discontinuity_indexes( 170 | sortedAttrValues, 171 | max(math.sqrt( 172 | len(sortedAttrValues)), 173 | min(10, 174 | len(sortedAttrValues)))) 175 | for index in indexes: 176 | avPairs.add((attrIndex, sortedAttrValues[index], operator.gt)) 177 | return avPairs 178 | 179 | 180 | def _get_discontinuity_indexes(sortedAttrValues, maxIndexes): 181 | indexes = [] 182 | for i in _generate_discontinuity_indexes_center_out(sortedAttrValues): 183 | indexes.append(i) 184 | if len(indexes) >= maxIndexes: 185 | break 186 | return indexes 187 | 188 | 189 | def _generate_discontinuity_indexes_center_out(sortedAttrValues): 190 | center = len(sortedAttrValues) // 2 191 | left = center - 1 192 | right = center + 1 193 | while left >= 0 or right < len(sortedAttrValues): 194 | if left >= 0: 195 | if sortedAttrValues[left] != sortedAttrValues[left + 1]: 196 | yield left 197 | left -= 1 198 | if right < len(sortedAttrValues): 199 | if sortedAttrValues[right - 1] != sortedAttrValues[right]: 200 | yield right - 1 201 | right += 1 202 | 203 | 204 | def read_csv(filepath): 205 | with open(filepath, 'r') as f: 206 | reader = csv.reader(f) 207 | data = list(reader) 208 | return data 209 | 210 | 211 | def prepare_data(data, numericColumnLabels=None): 212 | if numericColumnLabels is not None and len(numericColumnLabels) > 0: 213 | numericColumnIndexes = [data[0].index(label) for label in 214 | numericColumnLabels] 215 | for rowIndex, row in enumerate(data): 216 | if rowIndex == 0: 217 | continue 218 | for numericIndex in numericColumnIndexes: 219 | f = float(data[rowIndex][numericIndex]) if len( 220 | data[rowIndex][numericIndex]) > 0 else 0 221 | i = int(f) 222 | data[rowIndex][numericIndex] = i if i == f else f 223 | return data 224 | 225 | 226 | class DTree: 227 | def __init__(self, nodes, attrNames, outcomeIsContinuous=False): 228 | self._nodes = nodes 229 | self._attrNames = attrNames 230 | self._outcomeIsContinuous = outcomeIsContinuous 231 | 232 | @staticmethod 233 | def _is_leaf(node): 234 | return len(node) == 2 235 | 236 | def __str__(self): 237 | s = '' 238 | for node in self._nodes: 239 | if self._is_leaf(node): 240 | s += '{}: {}\n'.format(node[0], node[1]) 241 | else: 242 | nodeId, attrIndex, attrValue, isMatch, nodeIdIfMatch, \ 243 | nodeIdIfNonMatch, matchCount, nonMatchCount = node 244 | s += '{0}: {1}{7}{2}, {5} Yes->{3}, {6} No->{4}\n'.format( 245 | nodeId, self._attrNames[attrIndex], attrValue, 246 | nodeIdIfMatch, nodeIdIfNonMatch, matchCount, 247 | nonMatchCount, '=' if isMatch == operator.eq else '>') 248 | return s 249 | 250 | def get_prediction(self, data): 251 | currentNode = self._nodes[0] 252 | while True: 253 | if self._is_leaf(currentNode): 254 | node = currentNode[1] 255 | if type(node) is not list: 256 | return node 257 | if self._outcomeIsContinuous: 258 | node = sorted(node, key=lambda n: n[0]) 259 | randPercent = .5 if self._outcomeIsContinuous else \ 260 | random.uniform(0, 1) 261 | total = 0 262 | for outcome, percentage in node: 263 | total += percentage 264 | if total > randPercent: 265 | return outcome 266 | return node[-1][0] 267 | nodeId, attrIndex, attrValue, isMatch, nodeIdIfMatch, \ 268 | nodeIdIfNonMatch = currentNode[:6] 269 | currentNode = self._nodes[nodeIdIfMatch if 270 | isMatch(data[attrIndex], attrValue) else nodeIdIfNonMatch] 271 | -------------------------------------------------------------------------------- /ch06/test.py: -------------------------------------------------------------------------------- 1 | # File: test.py 2 | # from chapter 6 of _Tree-based Machine Learning Algorithms_ 3 | # 4 | # Author: Clinton Sheppard 5 | # Copyright (c) 2017 Clinton Sheppard 6 | # 7 | # Licensed under the Apache License, Version 2.0 (the "License"). 8 | # You may not use this file except in compliance with the License. 9 | # You may obtain a copy of the License at 10 | # http://www.apache.org/licenses/LICENSE-2.0 11 | # 12 | # Unless required by applicable law or agreed to in writing, software 13 | # distributed under the License is distributed on an "AS IS" BASIS, 14 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 15 | # implied. See the License for the specific language governing 16 | # permissions and limitations under the License. 17 | 18 | import dtree 19 | 20 | 21 | continuousAttributes = ['Age'] 22 | data = dtree.read_csv('..\ch04\census.csv') 23 | data = dtree.prepare_data(data, continuousAttributes) 24 | outcomeLabel = 'Age' 25 | 26 | tree = dtree.build(data, outcomeLabel, continuousAttributes, minimumSubsetSizePercentage=6) 27 | print(tree) 28 | 29 | testData = ['Elizabeth', 'female', 'Single', -1, 'Daughter', 'Germany'] 30 | 31 | predicted = tree.get_prediction(testData) 32 | print("predicted: {}".format(predicted)) -------------------------------------------------------------------------------- /ch07/dtree.py: -------------------------------------------------------------------------------- 1 | # File: dtree.py 2 | # from chapter 7 of _Tree-based Machine Learning Algorithms_ 3 | # 4 | # Author: Clinton Sheppard 5 | # Copyright (c) 2017 Clinton Sheppard 6 | # 7 | # Licensed under the Apache License, Version 2.0 (the "License"). 8 | # You may not use this file except in compliance with the License. 9 | # You may obtain a copy of the License at 10 | # http://www.apache.org/licenses/LICENSE-2.0 11 | # 12 | # Unless required by applicable law or agreed to in writing, software 13 | # distributed under the License is distributed on an "AS IS" BASIS, 14 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 15 | # implied. See the License for the specific language governing 16 | # permissions and limitations under the License. 17 | 18 | from collections import Counter 19 | from numbers import Number 20 | import operator 21 | import math 22 | import csv 23 | import random 24 | 25 | 26 | def _get_bias(avPair, dataRowIndexes, data, outcomeIndex, minimumSubsetSize, 27 | validationRowIndexes): 28 | attrIndex, attrValue, isMatch = avPair 29 | if len(validationRowIndexes) > 0: 30 | validationMatchIndexes = {i for i in validationRowIndexes if 31 | isMatch(data[i][attrIndex], attrValue)} 32 | validationNonMatchIndexes = validationRowIndexes - \ 33 | validationMatchIndexes 34 | if len(validationMatchIndexes) == 0 or len( 35 | validationNonMatchIndexes) == 0: 36 | return -2 37 | matchIndexes = {i for i in dataRowIndexes if 38 | isMatch(data[i][attrIndex], attrValue)} 39 | nonMatchIndexes = dataRowIndexes - matchIndexes 40 | if len(matchIndexes) < minimumSubsetSize or len( 41 | nonMatchIndexes) < minimumSubsetSize: 42 | return -1 43 | matchOutcomes = {data[i][outcomeIndex] for i in matchIndexes} 44 | nonMatchOutcomes = {data[i][outcomeIndex] for i in nonMatchIndexes} 45 | numPureRows = (len(matchIndexes) if len(matchOutcomes) == 1 else 0) \ 46 | + (len(nonMatchIndexes) if len(nonMatchOutcomes) == 1 47 | else 0) 48 | percentPure = numPureRows / len(dataRowIndexes) 49 | 50 | numNonPureRows = len(dataRowIndexes) - numPureRows 51 | percentNonPure = 1 - percentPure 52 | split = 1 - abs(len(matchIndexes) - len(nonMatchIndexes)) / len( 53 | dataRowIndexes) - .001 54 | splitBias = split * percentNonPure if numNonPureRows > 0 else 0 55 | return splitBias + percentPure 56 | 57 | 58 | def build(data, outcomeLabel, continuousAttributes=None, 59 | minimumSubsetSizePercentage=0, validationPercentage=0, 60 | dataIndexes=None, attrIndexes=None): 61 | if validationPercentage > 0: 62 | validationPercentage /= 100 63 | validationCount = int(validationPercentage * 64 | (len(data) if dataIndexes is None else len( 65 | dataIndexes))) 66 | if minimumSubsetSizePercentage > 0: 67 | minimumSubsetSizePercentage /= 100 68 | minimumSubsetSize = int(minimumSubsetSizePercentage * 69 | (len(data) if dataIndexes is None else len( 70 | dataIndexes))) 71 | if attrIndexes is None: 72 | attrIndexes = [index for index, label in enumerate(data[0]) if 73 | label != outcomeLabel] 74 | outcomeIndex = data[0].index(outcomeLabel) 75 | continuousAttrIndexes = set() 76 | if continuousAttributes is not None: 77 | continuousAttrIndexes = {data[0].index(label) for label in 78 | continuousAttributes} 79 | if len(continuousAttrIndexes) != len(continuousAttributes): 80 | raise Exception( 81 | 'One or more continuous column names are duplicates.') 82 | else: 83 | for attrIndex in attrIndexes: 84 | uniqueValues = {row[attrIndex] for rowIndex, row in 85 | enumerate(data) if rowIndex > 0} 86 | numericValues = {value for value in uniqueValues if 87 | isinstance(value, Number)} 88 | if len(uniqueValues) == len(numericValues): 89 | continuousAttrIndexes.add(attrIndex) 90 | 91 | if outcomeIndex in continuousAttrIndexes: 92 | continuousAttrIndexes.remove(outcomeIndex) 93 | outcomeIsContinuous = True 94 | else: 95 | outcomeIsContinuous = False 96 | 97 | nodes = [] 98 | lastNodeNumber = 0 99 | if dataIndexes is None: 100 | dataIndexes = {i for i in range(1, len(data))} 101 | elif not isinstance(dataIndexes, set): 102 | dataIndexes = {i for i in dataIndexes} 103 | validationIndexes = set() 104 | if validationCount > 0: 105 | validationIndexes = set( 106 | random.sample([i for i in dataIndexes], validationCount)) 107 | dataIndexes -= validationIndexes 108 | workQueue = [(-1, lastNodeNumber, dataIndexes, validationIndexes)] 109 | while len(workQueue) > 0: 110 | parentNodeId, nodeId, dataRowIndexes, validationRowIndexes = \ 111 | workQueue.pop() 112 | uniqueOutcomes = Counter( 113 | data[i][outcomeIndex] for i in dataRowIndexes).most_common() 114 | if len(uniqueOutcomes) == 1: 115 | nodes.append((nodeId, uniqueOutcomes.pop(0)[0])) 116 | continue 117 | potentials = _get_potentials(attrIndexes, continuousAttrIndexes, 118 | data, dataRowIndexes, outcomeIndex, 119 | minimumSubsetSize, 120 | validationRowIndexes) 121 | if len(potentials) == 0 or potentials[0][0] > 0: 122 | nodes.append((nodeId, [(n[0], n[1] / len(dataRowIndexes)) 123 | for n in uniqueOutcomes])) 124 | continue 125 | attrIndex, attrValue, isMatch = potentials[0][1:] 126 | matches = {rowIndex for rowIndex in dataRowIndexes if 127 | isMatch(data[rowIndex][attrIndex], attrValue)} 128 | nonMatches = dataRowIndexes - matches 129 | validationMatches = { 130 | rowIndex for rowIndex in validationRowIndexes if 131 | isMatch(data[rowIndex][attrIndex], attrValue)} 132 | nonValidationMatches = validationRowIndexes - validationMatches 133 | lastNodeNumber += 1 134 | matchId = lastNodeNumber 135 | workQueue.append((nodeId, matchId, matches, validationMatches)) 136 | lastNodeNumber += 1 137 | nonMatchId = lastNodeNumber 138 | workQueue.append((nodeId, nonMatchId, nonMatches, 139 | nonValidationMatches)) 140 | nodes.append((nodeId, attrIndex, attrValue, isMatch, matchId, 141 | nonMatchId, len(matches), len(nonMatches))) 142 | nodes = sorted(nodes, key=lambda n: n[0]) 143 | return DTree(nodes, data[0], outcomeIsContinuous) 144 | 145 | 146 | def _get_potentials(attrIndexes, continuousAttrIndexes, data, 147 | dataRowIndexes, outcomeIndex, minimumSubsetSize, 148 | validationRowIndexes): 149 | uniqueAttributeValuePairs = { 150 | (attrIndex, data[rowIndex][attrIndex], operator.eq) 151 | for attrIndex in attrIndexes 152 | if attrIndex not in continuousAttrIndexes 153 | for rowIndex in dataRowIndexes} 154 | continuousAttributeValuePairs = _get_continuous_av_pairs( 155 | continuousAttrIndexes, data, dataRowIndexes) 156 | uniqueAttributeValuePairs |= continuousAttributeValuePairs 157 | potentials = sorted((-_get_bias(avPair, dataRowIndexes, data, 158 | outcomeIndex, minimumSubsetSize, 159 | validationRowIndexes), 160 | avPair[0], avPair[1], avPair[2]) 161 | for avPair in uniqueAttributeValuePairs) 162 | return potentials 163 | 164 | def _get_continuous_av_pairs(continuousAttrIndexes, data, dataRowIndexes): 165 | avPairs = set() 166 | for attrIndex in continuousAttrIndexes: 167 | sortedAttrValues = [i for i in sorted( 168 | data[rowIndex][attrIndex] for rowIndex in dataRowIndexes)] 169 | indexes = _get_discontinuity_indexes( 170 | sortedAttrValues, 171 | max(math.sqrt( 172 | len(sortedAttrValues)), 173 | min(10, 174 | len(sortedAttrValues)))) 175 | for index in indexes: 176 | avPairs.add((attrIndex, sortedAttrValues[index], operator.gt)) 177 | return avPairs 178 | 179 | 180 | def _get_discontinuity_indexes(sortedAttrValues, maxIndexes): 181 | indexes = [] 182 | for i in _generate_discontinuity_indexes_center_out(sortedAttrValues): 183 | indexes.append(i) 184 | if len(indexes) >= maxIndexes: 185 | break 186 | return indexes 187 | 188 | 189 | def _generate_discontinuity_indexes_center_out(sortedAttrValues): 190 | center = len(sortedAttrValues) // 2 191 | left = center - 1 192 | right = center + 1 193 | while left >= 0 or right < len(sortedAttrValues): 194 | if left >= 0: 195 | if sortedAttrValues[left] != sortedAttrValues[left + 1]: 196 | yield left 197 | left -= 1 198 | if right < len(sortedAttrValues): 199 | if sortedAttrValues[right - 1] != sortedAttrValues[right]: 200 | yield right - 1 201 | right += 1 202 | 203 | 204 | def read_csv(filepath): 205 | with open(filepath, 'r') as f: 206 | reader = csv.reader(f) 207 | data = list(reader) 208 | return data 209 | 210 | 211 | def prepare_data(data, numericColumnLabels=None): 212 | if numericColumnLabels is not None and len(numericColumnLabels) > 0: 213 | numericColumnIndexes = [data[0].index(label) for label in 214 | numericColumnLabels] 215 | for rowIndex, row in enumerate(data): 216 | if rowIndex == 0: 217 | continue 218 | for numericIndex in numericColumnIndexes: 219 | f = float(data[rowIndex][numericIndex]) if len( 220 | data[rowIndex][numericIndex]) > 0 else 0 221 | i = int(f) 222 | data[rowIndex][numericIndex] = i if i == f else f 223 | return data 224 | 225 | 226 | class DTree: 227 | def __init__(self, nodes, attrNames, outcomeIsContinuous=False): 228 | self._nodes = nodes 229 | self._attrNames = attrNames 230 | self._outcomeIsContinuous = outcomeIsContinuous 231 | 232 | @staticmethod 233 | def _is_leaf(node): 234 | return len(node) == 2 235 | 236 | def __str__(self): 237 | s = '' 238 | for node in self._nodes: 239 | if self._is_leaf(node): 240 | s += '{}: {}\n'.format(node[0], node[1]) 241 | else: 242 | nodeId, attrIndex, attrValue, isMatch, nodeIdIfMatch, \ 243 | nodeIdIfNonMatch, matchCount, nonMatchCount = node 244 | s += '{0}: {1}{7}{2}, {5} Yes->{3}, {6} No->{4}\n'.format( 245 | nodeId, self._attrNames[attrIndex], attrValue, 246 | nodeIdIfMatch, nodeIdIfNonMatch, matchCount, 247 | nonMatchCount, '=' if isMatch == operator.eq else '>') 248 | return s 249 | 250 | def get_prediction(self, data): 251 | currentNode = self._nodes[0] 252 | while True: 253 | if self._is_leaf(currentNode): 254 | node = currentNode[1] 255 | if type(node) is not list: 256 | return node 257 | if self._outcomeIsContinuous: 258 | node = sorted(node, key=lambda n: n[0]) 259 | randPercent = .5 if self._outcomeIsContinuous else \ 260 | random.uniform(0, 1) 261 | total = 0 262 | for outcome, percentage in node: 263 | total += percentage 264 | if total > randPercent: 265 | return outcome 266 | return node[-1][0] 267 | nodeId, attrIndex, attrValue, isMatch, nodeIdIfMatch, \ 268 | nodeIdIfNonMatch = currentNode[:6] 269 | currentNode = self._nodes[nodeIdIfMatch if 270 | isMatch(data[attrIndex], attrValue) else nodeIdIfNonMatch] 271 | -------------------------------------------------------------------------------- /ch07/forest.py: -------------------------------------------------------------------------------- 1 | # File: forest.py 2 | # from chapter 7 of _Tree-based Machine Learning Algorithms_ 3 | # 4 | # Author: Clinton Sheppard 5 | # Copyright (c) 2017 Clinton Sheppard 6 | # 7 | # Licensed under the Apache License, Version 2.0 (the "License"). 8 | # You may not use this file except in compliance with the License. 9 | # You may obtain a copy of the License at 10 | # http://www.apache.org/licenses/LICENSE-2.0 11 | # 12 | # Unless required by applicable law or agreed to in writing, software 13 | # distributed under the License is distributed on an "AS IS" BASIS, 14 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 15 | # implied. See the License for the specific language governing 16 | # permissions and limitations under the License. 17 | 18 | import dtree 19 | import math 20 | import random 21 | import statistics 22 | import operator 23 | 24 | class Forest: 25 | def __init__(self, data, outcomeLabel, continuousAttributes=None, 26 | dataRowIndexes=None, columnsNamesToIgnore=None, 27 | boost=False): 28 | self.data = data 29 | self.outcomeLabel = outcomeLabel 30 | self.continuousAttributes = continuousAttributes \ 31 | if columnsNamesToIgnore is None \ 32 | else [i for i in continuousAttributes if 33 | i not in columnsNamesToIgnore] 34 | self.numRows = math.ceil(math.sqrt( 35 | len(data) if dataRowIndexes is None else len(dataRowIndexes))) 36 | self.outcomeIndex = data[0].index(outcomeLabel) 37 | columnIdsToIgnore = set() if columnsNamesToIgnore is None else set( 38 | data[0].index(s) for s in columnsNamesToIgnore) 39 | columnIdsToIgnore.add(self.outcomeIndex) 40 | self.attrIndexesExceptOutcomeIndex = [i for i in range(0, len(data[0])) 41 | if i not in columnIdsToIgnore] 42 | self.numAttributes = math.ceil( 43 | math.sqrt(len(self.attrIndexesExceptOutcomeIndex))) 44 | self.dataRowIndexes = range(1, len( 45 | data)) if dataRowIndexes is None else dataRowIndexes 46 | self.numTrees = 200 47 | self.boost = boost 48 | self.weights = [.5 for _ in range(0, self.numTrees)] 49 | self.populate() 50 | 51 | def _build_tree(self): 52 | return dtree.build(self.data, self.outcomeLabel, 53 | continuousAttributes=self.continuousAttributes, 54 | dataIndexes={i for i in random.sample( 55 | self.dataRowIndexes, self.numRows)}, 56 | attrIndexes=[ 57 | i for i in random.sample( 58 | self.attrIndexesExceptOutcomeIndex, 59 | self.numAttributes)]) 60 | 61 | def populate(self): 62 | self._trees = [self._build_tree() for _ in range(0, self.numTrees)] 63 | 64 | if not self.boost: 65 | return 66 | 67 | outcomeLabelIndex = self.data[0].index(self.outcomeLabel) 68 | anyChanged = True 69 | roundsRemaining = 10 70 | while anyChanged and roundsRemaining > 0: 71 | anyChanged = False 72 | roundsRemaining -= 1 73 | for dataRowIndex in self.dataRowIndexes: 74 | dataRow = self.data[dataRowIndex] 75 | sorted_predictions, predictions = self._get_predictions( 76 | dataRow) 77 | expectedPrediction = dataRow[outcomeLabelIndex] 78 | if expectedPrediction == sorted_predictions[0][0]: 79 | continue 80 | anyChanged = True 81 | actualPrediction = sorted_predictions[0][0] 82 | lookup = dict(sorted_predictions) 83 | expectedPredictionSum = lookup.get(expectedPrediction) 84 | difference = sorted_predictions[0][1] if \ 85 | expectedPredictionSum is None else \ 86 | sorted_predictions[0][1] - expectedPredictionSum 87 | maxDifference = difference / len(self.dataRowIndexes) 88 | if maxDifference == 0: 89 | maxDifference = .5 / len(self.dataRowIndexes) 90 | for index, p in enumerate(predictions): 91 | if p == expectedPrediction: 92 | self.weights[index] = min(1, self.weights[ 93 | index] + random.uniform(0, maxDifference)) 94 | continue 95 | if p == actualPrediction: 96 | self.weights[index] = max(0, self.weights[ 97 | index] - random.uniform(0, maxDifference)) 98 | if self.weights[index] == 0: 99 | self._trees[index] = self._build_tree() 100 | self.weights[index] = 0.5 101 | 102 | def get_prediction(self, data): 103 | sorted_predictions, _ = self._get_predictions(data) 104 | return sorted_predictions[0][0] 105 | 106 | def _get_predictions(self, data): 107 | predictions = [t.get_prediction(data) for t in self._trees] 108 | counts = {p: 0 for p in set(predictions)} 109 | for index, p in enumerate(predictions): 110 | counts[p] += self.weights[index] 111 | return sorted(counts.items(), key=operator.itemgetter(1), 112 | reverse=True), \ 113 | predictions 114 | 115 | 116 | class Benchmark: 117 | @staticmethod 118 | def run(function): 119 | results = [] 120 | for i in range(100): 121 | result = function() 122 | results.append(result) 123 | if i < 10 or i % 10 == 9: 124 | mean = statistics.mean(results) 125 | print("{} {:3.2f} {:3.2f}".format( 126 | 1 + i, mean, 127 | statistics.stdev(results, mean) if i > 1 else 0)) -------------------------------------------------------------------------------- /ch07/test.py: -------------------------------------------------------------------------------- 1 | # File: test.py 2 | # from chapter 7 of _Tree-based Machine Learning Algorithms_ 3 | # 4 | # Author: Clinton Sheppard 5 | # Copyright (c) 2017 Clinton Sheppard 6 | # 7 | # Licensed under the Apache License, Version 2.0 (the "License"). 8 | # You may not use this file except in compliance with the License. 9 | # You may obtain a copy of the License at 10 | # http://www.apache.org/licenses/LICENSE-2.0 11 | # 12 | # Unless required by applicable law or agreed to in writing, software 13 | # distributed under the License is distributed on an "AS IS" BASIS, 14 | # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or 15 | # implied. See the License for the specific language governing 16 | # permissions and limitations under the License. 17 | 18 | import dtree 19 | import random 20 | import forest 21 | 22 | data = dtree.read_csv('mushrooms.csv') 23 | outcomeLabel = 'class' 24 | outcomeLabelIndex = data[0].index(outcomeLabel) 25 | continuousAttributes = [] 26 | 27 | 28 | print("-- decision tree") 29 | def predict(): 30 | trainingRowIds = random.sample(range(1, len(data)), 31 | int(.01 * len(data))) 32 | tree = dtree.build(data, outcomeLabel, continuousAttributes, 33 | dataIndexes=trainingRowIds) 34 | correct = sum(1 for rowId, row in enumerate(data) if 35 | rowId > 0 and 36 | rowId not in trainingRowIds and 37 | tree.get_prediction(row) == row[outcomeLabelIndex]) 38 | return 100 * correct / (len(data) - 1 - len(trainingRowIds)) 39 | 40 | forest.Benchmark.run(predict) 41 | 42 | 43 | print("-- random forest") 44 | def predict2(): 45 | trainingRowIds = random.sample(range(1, len(data)), 46 | int(.01 * len(data))) 47 | f = forest.Forest(data, outcomeLabel, continuousAttributes, 48 | trainingRowIds) 49 | correct = sum(1 for rowId, row in enumerate(data) if 50 | rowId > 0 and 51 | rowId not in trainingRowIds and 52 | f.get_prediction(row) == row[outcomeLabelIndex]) 53 | return 100 * correct / (len(data) - 1 - len(trainingRowIds)) 54 | 55 | forest.Benchmark.run(predict2) 56 | 57 | 58 | print("-- boosted random forest") 59 | def predict3(): 60 | trainingRowIds = random.sample(range(1, len(data)), 61 | int(.01 * len(data))) 62 | f = forest.Forest(data, outcomeLabel, continuousAttributes, 63 | trainingRowIds, boost=True) 64 | 65 | correct = sum(1 for rowId, row in enumerate(data) if 66 | rowId > 0 and 67 | rowId not in trainingRowIds and 68 | f.get_prediction(row) == row[outcomeLabelIndex]) 69 | return 100 * correct / (len(data) - 1 - len(trainingRowIds)) 70 | 71 | forest.Benchmark.run(predict3) 72 | --------------------------------------------------------------------------------