├── .gitignore
├── CNN Deeplearning v2.ipynb
├── CNN Deeplearning.ipynb
├── CNN_Deeplearning_mattijs.ipynb
├── CNN_Deeplearning_walkforward.ipynb
├── CoinBase.ipynb
├── DL_for_LOB_Sirignano.ipynb
├── Data_Filtering.ipynb
├── DepthAnalysis.ipynb
├── LICENSE
├── README.md
├── WalkForwardV2.ipynb
├── Walkthrough.ipynb
└── paper
    └── main.tex


/.gitignore:
--------------------------------------------------------------------------------
  1 | # Byte-compiled / optimized / DLL files
  2 | __pycache__/
  3 | *.py[cod]
  4 | *$py.class
  5 | 
  6 | # C extensions
  7 | *.so
  8 | 
  9 | # Distribution / packaging
 10 | .Python
 11 | build/
 12 | develop-eggs/
 13 | dist/
 14 | downloads/
 15 | eggs/
 16 | .eggs/
 17 | lib/
 18 | lib64/
 19 | parts/
 20 | sdist/
 21 | var/
 22 | wheels/
 23 | *.egg-info/
 24 | .installed.cfg
 25 | *.egg
 26 | MANIFEST
 27 | 
 28 | # PyInstaller
 29 | #  Usually these files are written by a python script from a template
 30 | #  before PyInstaller builds the exe, so as to inject date/other infos into it.
 31 | *.manifest
 32 | *.spec
 33 | 
 34 | # Installer logs
 35 | pip-log.txt
 36 | pip-delete-this-directory.txt
 37 | 
 38 | # Unit test / coverage reports
 39 | htmlcov/
 40 | .tox/
 41 | .coverage
 42 | .coverage.*
 43 | .cache
 44 | nosetests.xml
 45 | coverage.xml
 46 | *.cover
 47 | .hypothesis/
 48 | .pytest_cache/
 49 | 
 50 | # Translations
 51 | *.mo
 52 | *.pot
 53 | 
 54 | # Django stuff:
 55 | *.log
 56 | local_settings.py
 57 | db.sqlite3
 58 | 
 59 | # Flask stuff:
 60 | instance/
 61 | .webassets-cache
 62 | 
 63 | # Scrapy stuff:
 64 | .scrapy
 65 | 
 66 | # Sphinx documentation
 67 | docs/_build/
 68 | 
 69 | # PyBuilder
 70 | target/
 71 | 
 72 | # Jupyter Notebook
 73 | .ipynb_checkpoints
 74 | 
 75 | # pyenv
 76 | .python-version
 77 | 
 78 | # celery beat schedule file
 79 | celerybeat-schedule
 80 | 
 81 | # SageMath parsed files
 82 | *.sage.py
 83 | 
 84 | # Environments
 85 | .env
 86 | .venv
 87 | env/
 88 | venv/
 89 | ENV/
 90 | env.bak/
 91 | venv.bak/
 92 | 
 93 | # Spyder project settings
 94 | .spyderproject
 95 | .spyproject
 96 | 
 97 | # Rope project settings
 98 | .ropeproject
 99 | 
100 | # mkdocs documentation
101 | /site
102 | 
103 | # mypy
104 | .mypy_cache/
105 | 
106 | # Latex
107 | *.aux
108 | *.fdb_latexmk
109 | *.fls
110 | paper/*.pdf
111 | *synctex*
112 | *.toc
113 | *.dvi
114 | 
115 | scraped_data*


--------------------------------------------------------------------------------
/CNN Deeplearning v2.ipynb:
--------------------------------------------------------------------------------
  1 | {
  2 |  "cells": [
  3 |   {
  4 |    "cell_type": "code",
  5 |    "execution_count": 1,
  6 |    "metadata": {},
  7 |    "outputs": [],
  8 |    "source": [
  9 |     "import os\n",
 10 |     "import pandas as pd\n",
 11 |     "import numpy as np\n",
 12 |     "%matplotlib inline\n",
 13 |     "import matplotlib.pyplot as plt\n",
 14 |     "import seaborn as sns\n",
 15 |     "import gc\n",
 16 |     "from tqdm import tqdm\n",
 17 |     "from sklearn.preprocessing import StandardScaler\n",
 18 |     "from sklearn.metrics import classification_report"
 19 |    ]
 20 |   },
 21 |   {
 22 |    "cell_type": "code",
 23 |    "execution_count": 2,
 24 |    "metadata": {},
 25 |    "outputs": [],
 26 |    "source": [
 27 |     "paths = 'coinbase_btc_usd/coinbase/btc_usd/l2_snapshots/100ms/'"
 28 |    ]
 29 |   },
 30 |   {
 31 |    "cell_type": "code",
 32 |    "execution_count": 3,
 33 |    "metadata": {},
 34 |    "outputs": [],
 35 |    "source": [
 36 |     "register = {}"
 37 |    ]
 38 |   },
 39 |   {
 40 |    "cell_type": "code",
 41 |    "execution_count": 4,
 42 |    "metadata": {},
 43 |    "outputs": [],
 44 |    "source": [
 45 |     "l2_snapshot = pd.DataFrame()"
 46 |    ]
 47 |   },
 48 |   {
 49 |    "cell_type": "code",
 50 |    "execution_count": 5,
 51 |    "metadata": {},
 52 |    "outputs": [
 53 |     {
 54 |      "name": "stdout",
 55 |      "output_type": "stream",
 56 |      "text": [
 57 |       "(566480, 200)\n",
 58 |       "(533024, 200)\n",
 59 |       "(517431, 200)\n",
 60 |       "(529814, 200)\n",
 61 |       "(537967, 200)\n",
 62 |       "(511850, 200)\n",
 63 |       "(545960, 200)\n",
 64 |       "(561058, 200)\n",
 65 |       "(535563, 200)\n"
 66 |      ]
 67 |     }
 68 |    ],
 69 |    "source": [
 70 |     "count = 0\n",
 71 |     "day = pd.DataFrame()\n",
 72 |     "i = 0\n",
 73 |     "for x in os.listdir(paths):\n",
 74 |     "    path = paths+x\n",
 75 |     "    temp = pd.read_parquet(path)\n",
 76 |     "    if count%24 == 0:\n",
 77 |     "        day = pd.read_parquet(path)\n",
 78 |     "    else:\n",
 79 |     "        day = pd.concat([day,temp])\n",
 80 |     "  \n",
 81 |     "    i+=1\n",
 82 |     "    count +=1\n",
 83 |     "\n",
 84 |     "    if count%24 == 0:\n",
 85 |     "        flag = not register\n",
 86 |     "        day = day.dropna()\n",
 87 |     "        print(day.shape)\n",
 88 |     "        result = []\n",
 89 |     "        for cols in day.columns:\n",
 90 |     "            values = day[cols].to_numpy()\n",
 91 |     "            mean = np.mean(values)\n",
 92 |     "            std = np.std(values)\n",
 93 |     "            result.append([cols,[mean,std]])\n",
 94 |     "        register[count/24] = result\n",
 95 |     "        if not flag:\n",
 96 |     "            for l in range(200):\n",
 97 |     "                prev_stat = register[(count/24)-1]\n",
 98 |     "                col_name = prev_stat[l][0]\n",
 99 |     "                col_mean = prev_stat[l][1][0]\n",
100 |     "                col_std = prev_stat[l][1][1]\n",
101 |     "                values = day[col_name].to_numpy()\n",
102 |     "                values = (values - col_mean)/col_std\n",
103 |     "                day[col_name] = values\n",
104 |     "            l2_snapshot = pd.concat([day,l2_snapshot])\n",
105 |     "        del day\n",
106 |     "  \n",
107 |     "    if i>240:\n",
108 |     "        break"
109 |    ]
110 |   },
111 |   {
112 |    "cell_type": "code",
113 |    "execution_count": 6,
114 |    "metadata": {},
115 |    "outputs": [],
116 |    "source": [
117 |     "l2_snapshot_ = l2_snapshot[['b1','b2','b3','b4','b5','b6','b7','b8', 'b9', 'b10', 'a1','a2','a3','a4','a5','a6','a7','a8', 'a9', 'a10', 'bq1','bq2','bq3','bq4','bq5','bq6','bq7','bq8', 'bq9', 'bq10', 'aq1','aq2','aq3','aq4','aq5','aq6','aq7','aq8', 'aq9', 'aq10']]"
118 |    ]
119 |   },
120 |   {
121 |    "cell_type": "code",
122 |    "execution_count": 7,
123 |    "metadata": {},
124 |    "outputs": [],
125 |    "source": [
126 |     "def generate_features_labels(df,ask, bid,k=20, alpha = 10e-5):\n",
127 |     "    df_ = df.copy()\n",
128 |     "    df_['mid_price'] = (df_[ask].to_numpy()+df_[bid].to_numpy())/2\n",
129 |     "    df_['target'] = 1\n",
130 |     "    index = df_.columns.get_loc('mid_price')\n",
131 |     "    target_index = df_.columns.get_loc('target')\n",
132 |     "    shape = df_.shape[0]\n",
133 |     "    y = df_['target'].to_numpy()\n",
134 |     "    for i in tqdm(range(k,shape-k)):\n",
135 |     "        if i==k:\n",
136 |     "            m_b = np.mean(df_.iloc[(i-k):i, index].to_numpy())\n",
137 |     "            m_a = np.mean(df_.iloc[i+1:(i+k+1), index].to_numpy())\n",
138 |     "            val = df_.iloc[i-k,index]\n",
139 |     "            valB = df_.iloc[i+1, index]\n",
140 |     "        else:\n",
141 |     "            curr_val = df_.iloc[i-1,index]\n",
142 |     "            curr_valB = df_.iloc[i+k, index]\n",
143 |     "            m_b = (m_b*k+(curr_val)-(val))/k\n",
144 |     "            m_a = (m_a*k+(curr_valB - valB))/k\n",
145 |     "            val = df_.iloc[(i-k),index]\n",
146 |     "            valB = df_.iloc[(i+1), index]\n",
147 |     "\n",
148 |     "        if (m_b > m_a*(1+alpha)):\n",
149 |     "            y[i] = 2\n",
150 |     "        if (m_b < m_a*(1-alpha)):\n",
151 |     "            y[i] = 0\n",
152 |     "\n",
153 |     "    y = y[k:shape-k]\n",
154 |     "    X = df.iloc[k:shape-k,:].to_numpy()\n",
155 |     "    return X,y"
156 |    ]
157 |   },
158 |   {
159 |    "cell_type": "code",
160 |    "execution_count": 8,
161 |    "metadata": {},
162 |    "outputs": [
163 |     {
164 |      "name": "stderr",
165 |      "output_type": "stream",
166 |      "text": [
167 |       "100%|██████████| 4272627/4272627 [02:53<00:00, 24654.56it/s]\n",
168 |       "100%|██████████| 4272627/4272627 [02:55<00:00, 24407.77it/s]\n",
169 |       "100%|██████████| 4272627/4272627 [03:02<00:00, 23402.77it/s]\n",
170 |       "100%|██████████| 4272627/4272627 [03:02<00:00, 23407.06it/s]\n",
171 |       "100%|██████████| 4272627/4272627 [02:57<00:00, 24138.81it/s]\n",
172 |       "100%|██████████| 4272627/4272627 [02:56<00:00, 24214.54it/s]\n",
173 |       "100%|██████████| 4272627/4272627 [02:54<00:00, 24419.84it/s]\n",
174 |       "100%|██████████| 4272627/4272627 [02:55<00:00, 24315.06it/s]\n",
175 |       "100%|██████████| 4272627/4272627 [03:00<00:00, 23713.46it/s]\n",
176 |       "100%|██████████| 4272627/4272627 [02:55<00:00, 24276.69it/s]\n"
177 |      ]
178 |     }
179 |    ],
180 |    "source": [
181 |     "# GridSearch\n",
182 |     "min_ = float('inf')\n",
183 |     "minI = 0\n",
184 |     "for i in np.logspace(0,-8,10, endpoint = True):\n",
185 |     "    X,y = generate_features_labels(l2_snapshot, ask='a1', bid = 'b1', alpha=i)\n",
186 |     "    y_ = pd.Series(y)\n",
187 |     "    temp = y_.value_counts()\n",
188 |     "    a,b,c = temp[0], temp[1], temp[2]\n",
189 |     "    sum_ = a+b+c\n",
190 |     "    l2 = ((1/3-(a/sum_))*100)**2 + ((1/3-(b/sum_))*100)**2 + ((1/3-(c/sum_))*100)**2\n",
191 |     "    if l2 < min_:\n",
192 |     "        min_ = l2\n",
193 |     "        minI = i\n",
194 |     "    del X,y,temp"
195 |    ]
196 |   },
197 |   {
198 |    "cell_type": "code",
199 |    "execution_count": 9,
200 |    "metadata": {},
201 |    "outputs": [
202 |     {
203 |      "name": "stdout",
204 |      "output_type": "stream",
205 |      "text": [
206 |       "7.742636826811277e-08 347.51453285264193\n"
207 |      ]
208 |     }
209 |    ],
210 |    "source": [
211 |     "print(minI, min_)"
212 |    ]
213 |   },
214 |   {
215 |    "cell_type": "code",
216 |    "execution_count": 10,
217 |    "metadata": {},
218 |    "outputs": [
219 |     {
220 |      "name": "stderr",
221 |      "output_type": "stream",
222 |      "text": [
223 |       "100%|██████████| 4272627/4272627 [02:55<00:00, 24396.68it/s]\n"
224 |      ]
225 |     }
226 |    ],
227 |    "source": [
228 |     "X,y = generate_features_labels(l2_snapshot_, ask='a1', bid = 'b1', alpha=minI)"
229 |    ]
230 |   },
231 |   {
232 |    "cell_type": "code",
233 |    "execution_count": 11,
234 |    "metadata": {
235 |     "scrolled": true
236 |    },
237 |    "outputs": [],
238 |    "source": [
239 |     "y = y.astype('int')"
240 |    ]
241 |   },
242 |   {
243 |    "cell_type": "code",
244 |    "execution_count": 12,
245 |    "metadata": {},
246 |    "outputs": [
247 |     {
248 |      "data": {
249 |       "text/plain": [
250 |        "<matplotlib.axes._subplots.AxesSubplot at 0x7f982a437d30>"
251 |       ]
252 |      },
253 |      "execution_count": 12,
254 |      "metadata": {},
255 |      "output_type": "execute_result"
256 |     },
257 |     {
258 |      "data": {
259 |       "image/png": "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\n",
260 |       "text/plain": [
261 |        "<Figure size 432x288 with 1 Axes>"
262 |       ]
263 |      },
264 |      "metadata": {
265 |       "needs_background": "light"
266 |      },
267 |      "output_type": "display_data"
268 |     }
269 |    ],
270 |    "source": [
271 |     "sns.countplot(x = y)"
272 |    ]
273 |   },
274 |   {
275 |    "cell_type": "code",
276 |    "execution_count": 13,
277 |    "metadata": {},
278 |    "outputs": [],
279 |    "source": [
280 |     "T = 100\n",
281 |     "D = 40\n",
282 |     "N = len(X) - T"
283 |    ]
284 |   },
285 |   {
286 |    "cell_type": "code",
287 |    "execution_count": 14,
288 |    "metadata": {},
289 |    "outputs": [],
290 |    "source": [
291 |     "trainPart = 3176046"
292 |    ]
293 |   },
294 |   {
295 |    "cell_type": "code",
296 |    "execution_count": 15,
297 |    "metadata": {},
298 |    "outputs": [],
299 |    "source": [
300 |     "X_train = np.zeros((trainPart, T, D), dtype = 'float16')\n",
301 |     "y_train = np.zeros(trainPart, dtype = 'float16')\n",
302 |     "\n",
303 |     "#Preparing the time series data using timestep of 100 and no of features = 200\n",
304 |     "for t in range(trainPart):\n",
305 |     "    X_train[t, :, :] = X[t:t+T]\n",
306 |     "    y_train[t] = y[t+T] \n"
307 |    ]
308 |   },
309 |   {
310 |    "cell_type": "code",
311 |    "execution_count": 16,
312 |    "metadata": {},
313 |    "outputs": [
314 |     {
315 |      "name": "stdout",
316 |      "output_type": "stream",
317 |      "text": [
318 |       "X_train - Before: 23.663386702537537 GB float16\n"
319 |      ]
320 |     }
321 |    ],
322 |    "source": [
323 |     "print('X_train - Before: {} GB'.format(X_train.nbytes/1024**3), X_train.dtype)"
324 |    ]
325 |   },
326 |   {
327 |    "cell_type": "code",
328 |    "execution_count": 17,
329 |    "metadata": {},
330 |    "outputs": [
331 |     {
332 |      "name": "stdout",
333 |      "output_type": "stream",
334 |      "text": [
335 |       "y_train - Before: 0.005915846675634384 GB float16\n"
336 |      ]
337 |     }
338 |    ],
339 |    "source": [
340 |     "print('y_train - Before: {} GB'.format(y_train.nbytes/1024**3), y_train.dtype)"
341 |    ]
342 |   },
343 |   {
344 |    "cell_type": "code",
345 |    "execution_count": 18,
346 |    "metadata": {},
347 |    "outputs": [
348 |     {
349 |      "name": "stdout",
350 |      "output_type": "stream",
351 |      "text": [
352 |       "X_test - Before: 8.169420063495636 GB float16\n",
353 |       "y_test - Before: 0.002042355015873909 GB float16\n"
354 |      ]
355 |     }
356 |    ],
357 |    "source": [
358 |     "X_test = np.zeros((N - trainPart, T, D), dtype='float16')\n",
359 |     "y_test = np.zeros(N - trainPart, dtype='float16')\n",
360 |     "\n",
361 |     "#Preparing the time series data using timestep of 100 and no of features = 40\n",
362 |     "for k in range(N - trainPart):\n",
363 |     "    t = k + trainPart\n",
364 |     "    X_test[k, :, :] = X[t:t+T]\n",
365 |     "    y_test[k] = y[t+T]\n",
366 |     "\n",
367 |     "print('X_test - Before: {} GB'.format(X_test.nbytes/1024**3), X_test.dtype)\n",
368 |     "print('y_test - Before: {} GB'.format(y_test.nbytes/1024**3), y_test.dtype)"
369 |    ]
370 |   },
371 |   {
372 |    "cell_type": "code",
373 |    "execution_count": 19,
374 |    "metadata": {},
375 |    "outputs": [
376 |     {
377 |      "name": "stdout",
378 |      "output_type": "stream",
379 |      "text": [
380 |       "2.1.0\n"
381 |      ]
382 |     }
383 |    ],
384 |    "source": [
385 |     "import tensorflow as tf\n",
386 |     "print(tf.__version__)"
387 |    ]
388 |   },
389 |   {
390 |    "cell_type": "code",
391 |    "execution_count": 20,
392 |    "metadata": {},
393 |    "outputs": [],
394 |    "source": [
395 |     "from tensorflow.keras.layers import Input, Conv1D, Conv2D, Flatten, MaxPooling1D, MaxPooling2D, Dense, LeakyReLU, Bidirectional\n",
396 |     "from tensorflow.keras.models import Model\n",
397 |     "from tensorflow.keras.optimizers import SGD, Adam\n",
398 |     "from tensorflow.compat.v1.keras.layers import CuDNNLSTM"
399 |    ]
400 |   },
401 |   {
402 |    "cell_type": "code",
403 |    "execution_count": 21,
404 |    "metadata": {},
405 |    "outputs": [],
406 |    "source": [
407 |     "X_train = np.expand_dims(X_train, -1)\n",
408 |     "X_test = np.expand_dims(X_test, -1)"
409 |    ]
410 |   },
411 |   {
412 |    "cell_type": "code",
413 |    "execution_count": 22,
414 |    "metadata": {},
415 |    "outputs": [],
416 |    "source": [
417 |     "i = Input(shape=X_train[0].shape)\n",
418 |     "x = Conv2D(16, (4,D), activation=LeakyReLU(alpha=0.01))(i)\n",
419 |     "x = tf.keras.layers.Reshape(target_shape=(T-3,16))(x)\n",
420 |     "x = Conv1D(16, 4, activation=LeakyReLU(alpha=0.01))(x)\n",
421 |     "x = tf.keras.layers.BatchNormalization()(x)\n",
422 |     "x = MaxPooling1D(2)(x)\n",
423 |     "x = Conv1D(32, 3, activation=LeakyReLU(alpha=0.01))(x)\n",
424 |     "x = tf.keras.layers.BatchNormalization()(x)\n",
425 |     "x = MaxPooling1D(2)(x)\n",
426 |     "x = Bidirectional(CuDNNLSTM(64, return_sequences = False))(x)\n",
427 |     "x = Dense(32, activation=LeakyReLU(alpha=0.01))(x)\n",
428 |     "x = Dense(32, activation=LeakyReLU(alpha=0.01))(x)\n",
429 |     "x = Dense(3, activation='softmax')(x)\n",
430 |     "\n",
431 |     "model = Model(i,x)"
432 |    ]
433 |   },
434 |   {
435 |    "cell_type": "code",
436 |    "execution_count": 23,
437 |    "metadata": {},
438 |    "outputs": [
439 |     {
440 |      "name": "stdout",
441 |      "output_type": "stream",
442 |      "text": [
443 |       "Model: \"model\"\n",
444 |       "_________________________________________________________________\n",
445 |       "Layer (type)                 Output Shape              Param #   \n",
446 |       "=================================================================\n",
447 |       "input_1 (InputLayer)         [(None, 100, 40, 1)]      0         \n",
448 |       "_________________________________________________________________\n",
449 |       "conv2d (Conv2D)              (None, 97, 1, 16)         2576      \n",
450 |       "_________________________________________________________________\n",
451 |       "reshape (Reshape)            (None, 97, 16)            0         \n",
452 |       "_________________________________________________________________\n",
453 |       "conv1d (Conv1D)              (None, 94, 16)            1040      \n",
454 |       "_________________________________________________________________\n",
455 |       "batch_normalization (BatchNo (None, 94, 16)            64        \n",
456 |       "_________________________________________________________________\n",
457 |       "max_pooling1d (MaxPooling1D) (None, 47, 16)            0         \n",
458 |       "_________________________________________________________________\n",
459 |       "conv1d_1 (Conv1D)            (None, 45, 32)            1568      \n",
460 |       "_________________________________________________________________\n",
461 |       "batch_normalization_1 (Batch (None, 45, 32)            128       \n",
462 |       "_________________________________________________________________\n",
463 |       "max_pooling1d_1 (MaxPooling1 (None, 22, 32)            0         \n",
464 |       "_________________________________________________________________\n",
465 |       "bidirectional (Bidirectional (None, 128)               50176     \n",
466 |       "_________________________________________________________________\n",
467 |       "dense (Dense)                (None, 32)                4128      \n",
468 |       "_________________________________________________________________\n",
469 |       "dense_1 (Dense)              (None, 32)                1056      \n",
470 |       "_________________________________________________________________\n",
471 |       "dense_2 (Dense)              (None, 3)                 99        \n",
472 |       "=================================================================\n",
473 |       "Total params: 60,835\n",
474 |       "Trainable params: 60,739\n",
475 |       "Non-trainable params: 96\n",
476 |       "_________________________________________________________________\n"
477 |      ]
478 |     }
479 |    ],
480 |    "source": [
481 |     "model.summary()"
482 |    ]
483 |   },
484 |   {
485 |    "cell_type": "code",
486 |    "execution_count": 24,
487 |    "metadata": {},
488 |    "outputs": [],
489 |    "source": [
490 |     "model.compile(optimizer='adam',\n",
491 |     "              loss='sparse_categorical_crossentropy',\n",
492 |     "              metrics=['accuracy'])"
493 |    ]
494 |   },
495 |   {
496 |    "cell_type": "code",
497 |    "execution_count": 25,
498 |    "metadata": {},
499 |    "outputs": [
500 |     {
501 |      "name": "stdout",
502 |      "output_type": "stream",
503 |      "text": [
504 |       "Train on 3176046 samples, validate on 1096481 samples\n",
505 |       "Epoch 1/4\n",
506 |       "3176046/3176046 [==============================] - 1623s 511us/sample - loss: 0.4721 - accuracy: 0.8082 - val_loss: 0.4457 - val_accuracy: 0.8195\n",
507 |       "Epoch 2/4\n",
508 |       "3176046/3176046 [==============================] - 1616s 509us/sample - loss: 0.4309 - accuracy: 0.8272 - val_loss: 0.4443 - val_accuracy: 0.8223\n",
509 |       "Epoch 3/4\n",
510 |       "3176046/3176046 [==============================] - 1624s 511us/sample - loss: 0.4118 - accuracy: 0.8345 - val_loss: 0.4485 - val_accuracy: 0.8240\n",
511 |       "Epoch 4/4\n",
512 |       "3176046/3176046 [==============================] - 1621s 511us/sample - loss: 0.3971 - accuracy: 0.8399 - val_loss: 0.4598 - val_accuracy: 0.8224\n"
513 |      ]
514 |     }
515 |    ],
516 |    "source": [
517 |     "EPOCHS = 4\n",
518 |     "BATCH_SIZE = 32\n",
519 |     "r = model.fit(\n",
520 |     "  X_train, y_train,\n",
521 |     "  batch_size=BATCH_SIZE,\n",
522 |     "  epochs=EPOCHS,\n",
523 |     "  validation_data=(X_test, y_test),\n",
524 |     ")"
525 |    ]
526 |   },
527 |   {
528 |    "cell_type": "code",
529 |    "execution_count": 26,
530 |    "metadata": {
531 |     "scrolled": true
532 |    },
533 |    "outputs": [
534 |     {
535 |      "data": {
536 |       "text/plain": [
537 |        "<matplotlib.legend.Legend at 0x7f977e25f940>"
538 |       ]
539 |      },
540 |      "execution_count": 26,
541 |      "metadata": {},
542 |      "output_type": "execute_result"
543 |     },
544 |     {
545 |      "data": {
546 |       "image/png": "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\n",
547 |       "text/plain": [
548 |        "<Figure size 432x288 with 1 Axes>"
549 |       ]
550 |      },
551 |      "metadata": {
552 |       "needs_background": "light"
553 |      },
554 |      "output_type": "display_data"
555 |     }
556 |    ],
557 |    "source": [
558 |     "sns.lineplot(x=range(1,EPOCHS+1),y=r.history['loss'])\n",
559 |     "sns.lineplot(x=range(1,EPOCHS+1),y=r.history['val_loss'])\n",
560 |     "plt.title('Model Cross Entropy Loss')\n",
561 |     "plt.ylabel('Cross Entropy Loss')\n",
562 |     "plt.xlabel('Epochs')\n",
563 |     "plt.legend(['Training', 'Testing'], loc='upper left')"
564 |    ]
565 |   },
566 |   {
567 |    "cell_type": "code",
568 |    "execution_count": 27,
569 |    "metadata": {},
570 |    "outputs": [],
571 |    "source": [
572 |     "y_pred = model.predict(X_test).argmax(axis=1)"
573 |    ]
574 |   },
575 |   {
576 |    "cell_type": "code",
577 |    "execution_count": 28,
578 |    "metadata": {},
579 |    "outputs": [
580 |     {
581 |      "data": {
582 |       "image/png": "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\n",
583 |       "text/plain": [
584 |        "<Figure size 432x288 with 2 Axes>"
585 |       ]
586 |      },
587 |      "metadata": {
588 |       "needs_background": "light"
589 |      },
590 |      "output_type": "display_data"
591 |     }
592 |    ],
593 |    "source": [
594 |     "from sklearn.metrics import confusion_matrix\n",
595 |     "\n",
596 |     "cm = confusion_matrix(y_pred=y_pred, y_true=y_test)\n",
597 |     "    \n",
598 |     "cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]\n",
599 |     "    \n",
600 |     "ax=sns.heatmap(cm, annot=True, xticklabels=[0,1,2], yticklabels=[0,1,2], cmap='Blues')\n",
601 |     "ax.set_ylim(3.0, 0)\n",
602 |     "plt.show()"
603 |    ]
604 |   },
605 |   {
606 |    "cell_type": "code",
607 |    "execution_count": 29,
608 |    "metadata": {},
609 |    "outputs": [
610 |     {
611 |      "name": "stdout",
612 |      "output_type": "stream",
613 |      "text": [
614 |       "             precision    recall  f1-score   support\n",
615 |       "\n",
616 |       "        0.0       0.85      0.87      0.86    566555\n",
617 |       "        1.0       0.81      0.95      0.87    304953\n",
618 |       "        2.0       0.73      0.53      0.62    224973\n",
619 |       "\n",
620 |       "avg / total       0.82      0.82      0.82   1096481\n",
621 |       "\n"
622 |      ]
623 |     }
624 |    ],
625 |    "source": [
626 |     "print(classification_report(y_test,y_pred))"
627 |    ]
628 |   },
629 |   {
630 |    "cell_type": "code",
631 |    "execution_count": null,
632 |    "metadata": {},
633 |    "outputs": [],
634 |    "source": []
635 |   }
636 |  ],
637 |  "metadata": {
638 |   "kernelspec": {
639 |    "display_name": "Python 3",
640 |    "language": "python",
641 |    "name": "python3"
642 |   },
643 |   "language_info": {
644 |    "codemirror_mode": {
645 |     "name": "ipython",
646 |     "version": 3
647 |    },
648 |    "file_extension": ".py",
649 |    "mimetype": "text/x-python",
650 |    "name": "python",
651 |    "nbconvert_exporter": "python",
652 |    "pygments_lexer": "ipython3",
653 |    "version": "3.5.3"
654 |   }
655 |  },
656 |  "nbformat": 4,
657 |  "nbformat_minor": 2
658 | }
659 | 


--------------------------------------------------------------------------------
/Data_Filtering.ipynb:
--------------------------------------------------------------------------------
   1 | {
   2 |  "cells": [
   3 |   {
   4 |    "cell_type": "code",
   5 |    "execution_count": 1,
   6 |    "metadata": {},
   7 |    "outputs": [
   8 |     {
   9 |      "name": "stderr",
  10 |      "output_type": "stream",
  11 |      "text": [
  12 |       "Using TensorFlow backend.\n"
  13 |      ]
  14 |     }
  15 |    ],
  16 |    "source": [
  17 |     "import keras\n",
  18 |     "import pandas as pd\n",
  19 |     "import numpy as np\n",
  20 |     "import matplotlib.pyplot as plt\n",
  21 |     "import os"
  22 |    ]
  23 |   },
  24 |   {
  25 |    "cell_type": "markdown",
  26 |    "metadata": {},
  27 |    "source": [
  28 |     "Data available: snapshot, ticker and trades\n",
  29 |     "\n",
  30 |     "Snapshot (100ms and 10s): gives the state of the orderbook at specified time interval. Includes 100 non-zero levels, 50 ask  50 bid, and their quantities.\n",
  31 |     "\n",
  32 |     "Ticker: gives the bid and ask price at each timestamp for which a trade has been made\n",
  33 |     "\n",
  34 |     "Trades: gives the time, side, price and quantity of every trade"
  35 |    ]
  36 |   },
  37 |   {
  38 |    "cell_type": "code",
  39 |    "execution_count": 181,
  40 |    "metadata": {},
  41 |    "outputs": [],
  42 |    "source": [
  43 |     "df_snapshot = pd.read_parquet(r\"Data\\coinbase_btc_usd\\coinbase\\btc_usd\\l2_snapshots\\100ms\\coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_0000_0100.parquet\")\n",
  44 |     "df_ticker = pd.read_parquet(r'Data\\coinbase_btc_usd\\coinbase\\btc_usd\\ticker\\coinbase_btc_usd_ticker_2019_11_12.parquet')\n",
  45 |     "df_trades = pd.read_parquet(r'Data\\coinbase_btc_usd\\coinbase\\btc_usd\\trades\\coinbase_btc_usd_trades_2019_11_12.parquet')"
  46 |    ]
  47 |   },
  48 |   {
  49 |    "cell_type": "code",
  50 |    "execution_count": 171,
  51 |    "metadata": {},
  52 |    "outputs": [
  53 |     {
  54 |      "data": {
  55 |       "text/html": [
  56 |        "<div>\n",
  57 |        "<style scoped>\n",
  58 |        "    .dataframe tbody tr th:only-of-type {\n",
  59 |        "        vertical-align: middle;\n",
  60 |        "    }\n",
  61 |        "\n",
  62 |        "    .dataframe tbody tr th {\n",
  63 |        "        vertical-align: top;\n",
  64 |        "    }\n",
  65 |        "\n",
  66 |        "    .dataframe thead th {\n",
  67 |        "        text-align: right;\n",
  68 |        "    }\n",
  69 |        "</style>\n",
  70 |        "<table border=\"1\" class=\"dataframe\">\n",
  71 |        "  <thead>\n",
  72 |        "    <tr style=\"text-align: right;\">\n",
  73 |        "      <th></th>\n",
  74 |        "      <th>b1</th>\n",
  75 |        "      <th>b2</th>\n",
  76 |        "      <th>b3</th>\n",
  77 |        "      <th>b4</th>\n",
  78 |        "      <th>b5</th>\n",
  79 |        "      <th>b6</th>\n",
  80 |        "      <th>b7</th>\n",
  81 |        "      <th>b8</th>\n",
  82 |        "      <th>b9</th>\n",
  83 |        "      <th>b10</th>\n",
  84 |        "      <th>...</th>\n",
  85 |        "      <th>aq41</th>\n",
  86 |        "      <th>aq42</th>\n",
  87 |        "      <th>aq43</th>\n",
  88 |        "      <th>aq44</th>\n",
  89 |        "      <th>aq45</th>\n",
  90 |        "      <th>aq46</th>\n",
  91 |        "      <th>aq47</th>\n",
  92 |        "      <th>aq48</th>\n",
  93 |        "      <th>aq49</th>\n",
  94 |        "      <th>aq50</th>\n",
  95 |        "    </tr>\n",
  96 |        "    <tr>\n",
  97 |        "      <th>timestamp</th>\n",
  98 |        "      <th></th>\n",
  99 |        "      <th></th>\n",
 100 |        "      <th></th>\n",
 101 |        "      <th></th>\n",
 102 |        "      <th></th>\n",
 103 |        "      <th></th>\n",
 104 |        "      <th></th>\n",
 105 |        "      <th></th>\n",
 106 |        "      <th></th>\n",
 107 |        "      <th></th>\n",
 108 |        "      <th></th>\n",
 109 |        "      <th></th>\n",
 110 |        "      <th></th>\n",
 111 |        "      <th></th>\n",
 112 |        "      <th></th>\n",
 113 |        "      <th></th>\n",
 114 |        "      <th></th>\n",
 115 |        "      <th></th>\n",
 116 |        "      <th></th>\n",
 117 |        "      <th></th>\n",
 118 |        "      <th></th>\n",
 119 |        "    </tr>\n",
 120 |        "  </thead>\n",
 121 |        "  <tbody>\n",
 122 |        "    <tr>\n",
 123 |        "      <th>2019-11-12 00:00:00.000</th>\n",
 124 |        "      <td>8721.53</td>\n",
 125 |        "      <td>8720.59</td>\n",
 126 |        "      <td>8719.55</td>\n",
 127 |        "      <td>8719.50</td>\n",
 128 |        "      <td>8719.0</td>\n",
 129 |        "      <td>8718.02</td>\n",
 130 |        "      <td>8717.87</td>\n",
 131 |        "      <td>8717.85</td>\n",
 132 |        "      <td>8717.56</td>\n",
 133 |        "      <td>8716.06</td>\n",
 134 |        "      <td>...</td>\n",
 135 |        "      <td>8.200000</td>\n",
 136 |        "      <td>2.203</td>\n",
 137 |        "      <td>2.600</td>\n",
 138 |        "      <td>0.005</td>\n",
 139 |        "      <td>2.500</td>\n",
 140 |        "      <td>0.896321</td>\n",
 141 |        "      <td>0.766000</td>\n",
 142 |        "      <td>0.001737</td>\n",
 143 |        "      <td>1.820000</td>\n",
 144 |        "      <td>0.933419</td>\n",
 145 |        "    </tr>\n",
 146 |        "    <tr>\n",
 147 |        "      <th>2019-11-12 00:00:00.100</th>\n",
 148 |        "      <td>8721.53</td>\n",
 149 |        "      <td>8720.59</td>\n",
 150 |        "      <td>8719.56</td>\n",
 151 |        "      <td>8719.55</td>\n",
 152 |        "      <td>8719.0</td>\n",
 153 |        "      <td>8718.02</td>\n",
 154 |        "      <td>8717.87</td>\n",
 155 |        "      <td>8717.85</td>\n",
 156 |        "      <td>8717.56</td>\n",
 157 |        "      <td>8716.06</td>\n",
 158 |        "      <td>...</td>\n",
 159 |        "      <td>8.200000</td>\n",
 160 |        "      <td>2.203</td>\n",
 161 |        "      <td>2.600</td>\n",
 162 |        "      <td>0.005</td>\n",
 163 |        "      <td>2.500</td>\n",
 164 |        "      <td>0.896321</td>\n",
 165 |        "      <td>0.766000</td>\n",
 166 |        "      <td>0.001737</td>\n",
 167 |        "      <td>1.820000</td>\n",
 168 |        "      <td>0.933419</td>\n",
 169 |        "    </tr>\n",
 170 |        "    <tr>\n",
 171 |        "      <th>2019-11-12 00:00:00.200</th>\n",
 172 |        "      <td>8721.53</td>\n",
 173 |        "      <td>8720.59</td>\n",
 174 |        "      <td>8719.61</td>\n",
 175 |        "      <td>8719.56</td>\n",
 176 |        "      <td>8719.0</td>\n",
 177 |        "      <td>8718.02</td>\n",
 178 |        "      <td>8717.87</td>\n",
 179 |        "      <td>8717.85</td>\n",
 180 |        "      <td>8716.06</td>\n",
 181 |        "      <td>8716.00</td>\n",
 182 |        "      <td>...</td>\n",
 183 |        "      <td>0.009135</td>\n",
 184 |        "      <td>8.200</td>\n",
 185 |        "      <td>2.203</td>\n",
 186 |        "      <td>2.600</td>\n",
 187 |        "      <td>0.005</td>\n",
 188 |        "      <td>2.500000</td>\n",
 189 |        "      <td>0.896321</td>\n",
 190 |        "      <td>0.766000</td>\n",
 191 |        "      <td>0.001737</td>\n",
 192 |        "      <td>1.820000</td>\n",
 193 |        "    </tr>\n",
 194 |        "    <tr>\n",
 195 |        "      <th>2019-11-12 00:00:00.300</th>\n",
 196 |        "      <td>8721.53</td>\n",
 197 |        "      <td>8720.59</td>\n",
 198 |        "      <td>8719.61</td>\n",
 199 |        "      <td>8719.56</td>\n",
 200 |        "      <td>8719.0</td>\n",
 201 |        "      <td>8718.02</td>\n",
 202 |        "      <td>8718.00</td>\n",
 203 |        "      <td>8717.87</td>\n",
 204 |        "      <td>8717.85</td>\n",
 205 |        "      <td>8716.60</td>\n",
 206 |        "      <td>...</td>\n",
 207 |        "      <td>0.009135</td>\n",
 208 |        "      <td>8.200</td>\n",
 209 |        "      <td>2.203</td>\n",
 210 |        "      <td>2.600</td>\n",
 211 |        "      <td>0.005</td>\n",
 212 |        "      <td>2.500000</td>\n",
 213 |        "      <td>0.896321</td>\n",
 214 |        "      <td>0.766000</td>\n",
 215 |        "      <td>0.001737</td>\n",
 216 |        "      <td>1.820000</td>\n",
 217 |        "    </tr>\n",
 218 |        "    <tr>\n",
 219 |        "      <th>2019-11-12 00:00:00.400</th>\n",
 220 |        "      <td>8721.53</td>\n",
 221 |        "      <td>8720.59</td>\n",
 222 |        "      <td>8719.61</td>\n",
 223 |        "      <td>8719.56</td>\n",
 224 |        "      <td>8719.0</td>\n",
 225 |        "      <td>8718.99</td>\n",
 226 |        "      <td>8718.02</td>\n",
 227 |        "      <td>8718.00</td>\n",
 228 |        "      <td>8717.87</td>\n",
 229 |        "      <td>8717.85</td>\n",
 230 |        "      <td>...</td>\n",
 231 |        "      <td>0.009135</td>\n",
 232 |        "      <td>8.200</td>\n",
 233 |        "      <td>2.203</td>\n",
 234 |        "      <td>2.600</td>\n",
 235 |        "      <td>0.005</td>\n",
 236 |        "      <td>2.500000</td>\n",
 237 |        "      <td>0.896321</td>\n",
 238 |        "      <td>0.766000</td>\n",
 239 |        "      <td>0.001737</td>\n",
 240 |        "      <td>1.820000</td>\n",
 241 |        "    </tr>\n",
 242 |        "  </tbody>\n",
 243 |        "</table>\n",
 244 |        "<p>5 rows × 200 columns</p>\n",
 245 |        "</div>"
 246 |       ],
 247 |       "text/plain": [
 248 |        "                              b1       b2       b3       b4      b5       b6  \\\n",
 249 |        "timestamp                                                                      \n",
 250 |        "2019-11-12 00:00:00.000  8721.53  8720.59  8719.55  8719.50  8719.0  8718.02   \n",
 251 |        "2019-11-12 00:00:00.100  8721.53  8720.59  8719.56  8719.55  8719.0  8718.02   \n",
 252 |        "2019-11-12 00:00:00.200  8721.53  8720.59  8719.61  8719.56  8719.0  8718.02   \n",
 253 |        "2019-11-12 00:00:00.300  8721.53  8720.59  8719.61  8719.56  8719.0  8718.02   \n",
 254 |        "2019-11-12 00:00:00.400  8721.53  8720.59  8719.61  8719.56  8719.0  8718.99   \n",
 255 |        "\n",
 256 |        "                              b7       b8       b9      b10  ...      aq41  \\\n",
 257 |        "timestamp                                                    ...             \n",
 258 |        "2019-11-12 00:00:00.000  8717.87  8717.85  8717.56  8716.06  ...  8.200000   \n",
 259 |        "2019-11-12 00:00:00.100  8717.87  8717.85  8717.56  8716.06  ...  8.200000   \n",
 260 |        "2019-11-12 00:00:00.200  8717.87  8717.85  8716.06  8716.00  ...  0.009135   \n",
 261 |        "2019-11-12 00:00:00.300  8718.00  8717.87  8717.85  8716.60  ...  0.009135   \n",
 262 |        "2019-11-12 00:00:00.400  8718.02  8718.00  8717.87  8717.85  ...  0.009135   \n",
 263 |        "\n",
 264 |        "                          aq42   aq43   aq44   aq45      aq46      aq47  \\\n",
 265 |        "timestamp                                                                 \n",
 266 |        "2019-11-12 00:00:00.000  2.203  2.600  0.005  2.500  0.896321  0.766000   \n",
 267 |        "2019-11-12 00:00:00.100  2.203  2.600  0.005  2.500  0.896321  0.766000   \n",
 268 |        "2019-11-12 00:00:00.200  8.200  2.203  2.600  0.005  2.500000  0.896321   \n",
 269 |        "2019-11-12 00:00:00.300  8.200  2.203  2.600  0.005  2.500000  0.896321   \n",
 270 |        "2019-11-12 00:00:00.400  8.200  2.203  2.600  0.005  2.500000  0.896321   \n",
 271 |        "\n",
 272 |        "                             aq48      aq49      aq50  \n",
 273 |        "timestamp                                              \n",
 274 |        "2019-11-12 00:00:00.000  0.001737  1.820000  0.933419  \n",
 275 |        "2019-11-12 00:00:00.100  0.001737  1.820000  0.933419  \n",
 276 |        "2019-11-12 00:00:00.200  0.766000  0.001737  1.820000  \n",
 277 |        "2019-11-12 00:00:00.300  0.766000  0.001737  1.820000  \n",
 278 |        "2019-11-12 00:00:00.400  0.766000  0.001737  1.820000  \n",
 279 |        "\n",
 280 |        "[5 rows x 200 columns]"
 281 |       ]
 282 |      },
 283 |      "execution_count": 171,
 284 |      "metadata": {},
 285 |      "output_type": "execute_result"
 286 |     }
 287 |    ],
 288 |    "source": [
 289 |     "df_snapshot.head()"
 290 |    ]
 291 |   },
 292 |   {
 293 |    "cell_type": "markdown",
 294 |    "metadata": {},
 295 |    "source": [
 296 |     "# Example Bid Price, Moving Up"
 297 |    ]
 298 |   },
 299 |   {
 300 |    "cell_type": "markdown",
 301 |    "metadata": {},
 302 |    "source": [
 303 |     "Neural network modelling p(Y = y1 |Y >= y1, X=x)\n",
 304 |     "\n",
 305 |     "There can be zero levels both in the spread as well as in the bid and ask sides. Furthermore, many changes are happening within our 100ms snapshot window. I thus see two ways by which to calculate the price level change.\n",
 306 |     "\n",
 307 |     "#### 1.\n",
 308 |     "As the price is to 0.01 accuracy one can simply take the difference in the best bid price and multiply by 100 to get it in the terms of how many $0.01$ levels it has increased by.\n",
 309 |     "\n",
 310 |     "#### 2.\n",
 311 |     "Or one can approximate the price level change by taking that which is closest to the current non-zero levels present.\n",
 312 |     " e.g. if the bid price moves up in the next step to 5.20 and at the current time step there are ask price levels at 5.15, 5.18, 5.22, ... then there will be a price level increase of 2 levels. This seems somewhat more arbitrary and heavily dependent on the current state of the orderbook."
 313 |    ]
 314 |   },
 315 |   {
 316 |    "cell_type": "markdown",
 317 |    "metadata": {},
 318 |    "source": [
 319 |     "Each snapshot file contains one hour of data at 100ms time interval. We will combine all this data into four separate dataframes (one to demonstrate the principles). One for where the bid price has increased, one where it has decreased and the same for the ask price. Thus for the example case of bid price increasing:"
 320 |    ]
 321 |   },
 322 |   {
 323 |    "cell_type": "code",
 324 |    "execution_count": 203,
 325 |    "metadata": {},
 326 |    "outputs": [
 327 |     {
 328 |      "name": "stdout",
 329 |      "output_type": "stream",
 330 |      "text": [
 331 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_0000_0100.parquet\n",
 332 |       "Non na entries: 28547\n",
 333 |       "Entries left: 3589\n",
 334 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_0100_0200.parquet\n",
 335 |       "Non na entries: 29365\n",
 336 |       "Entries left: 4635\n",
 337 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_0200_0300.parquet\n",
 338 |       "Non na entries: 28656\n",
 339 |       "Entries left: 3228\n",
 340 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_0300_0400.parquet\n",
 341 |       "Non na entries: 30871\n",
 342 |       "Entries left: 4663\n",
 343 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_0400_0500.parquet\n",
 344 |       "Non na entries: 29187\n",
 345 |       "Entries left: 3622\n",
 346 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_0500_0600.parquet\n",
 347 |       "Non na entries: 31956\n",
 348 |       "Entries left: 5930\n",
 349 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_0600_0700.parquet\n",
 350 |       "Non na entries: 23169\n",
 351 |       "Entries left: 3822\n",
 352 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_0700_0800.parquet\n",
 353 |       "Non na entries: 22659\n",
 354 |       "Entries left: 1894\n",
 355 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_0800_0900.parquet\n",
 356 |       "Non na entries: 19565\n",
 357 |       "Entries left: 1055\n",
 358 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_0900_1000.parquet\n",
 359 |       "Non na entries: 22605\n",
 360 |       "Entries left: 1782\n",
 361 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_1000_1100.parquet\n",
 362 |       "Non na entries: 17245\n",
 363 |       "Entries left: 1046\n",
 364 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_1100_1200.parquet\n",
 365 |       "Non na entries: 20052\n",
 366 |       "Entries left: 2119\n",
 367 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_1200_1300.parquet\n",
 368 |       "Non na entries: 31358\n",
 369 |       "Entries left: 4601\n",
 370 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_1300_1400.parquet\n",
 371 |       "Non na entries: 26075\n",
 372 |       "Entries left: 2978\n",
 373 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_1400_1500.parquet\n",
 374 |       "Non na entries: 22544\n",
 375 |       "Entries left: 1559\n",
 376 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_1500_1600.parquet\n",
 377 |       "Non na entries: 28214\n",
 378 |       "Entries left: 4909\n",
 379 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_1600_1700.parquet\n",
 380 |       "Non na entries: 25146\n",
 381 |       "Entries left: 2304\n",
 382 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_1700_1800.parquet\n",
 383 |       "Non na entries: 20224\n",
 384 |       "Entries left: 1736\n",
 385 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_1800_1900.parquet\n",
 386 |       "Non na entries: 24429\n",
 387 |       "Entries left: 1598\n",
 388 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_1900_2000.parquet\n",
 389 |       "Non na entries: 27789\n",
 390 |       "Entries left: 2517\n",
 391 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_2000_2100.parquet\n",
 392 |       "Non na entries: 27963\n",
 393 |       "Entries left: 3174\n",
 394 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_2100_2200.parquet\n",
 395 |       "Non na entries: 27390\n",
 396 |       "Entries left: 3614\n",
 397 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_2200_2300.parquet\n",
 398 |       "Non na entries: 26435\n",
 399 |       "Entries left: 2435\n",
 400 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_12_2300_0000.parquet\n",
 401 |       "Non na entries: 27631\n",
 402 |       "Entries left: 3604\n",
 403 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_0000_0100.parquet\n",
 404 |       "Non na entries: 28699\n",
 405 |       "Entries left: 5762\n",
 406 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_0100_0200.parquet\n",
 407 |       "Non na entries: 27539\n",
 408 |       "Entries left: 1893\n",
 409 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_0200_0300.parquet\n",
 410 |       "Non na entries: 22991\n",
 411 |       "Entries left: 2141\n",
 412 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_0300_0400.parquet\n",
 413 |       "Non na entries: 26243\n",
 414 |       "Entries left: 1950\n",
 415 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_0400_0500.parquet\n",
 416 |       "Non na entries: 30195\n",
 417 |       "Entries left: 1472\n",
 418 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_0500_0600.parquet\n",
 419 |       "Non na entries: 23088\n",
 420 |       "Entries left: 3635\n",
 421 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_0600_0700.parquet\n",
 422 |       "Non na entries: 25195\n",
 423 |       "Entries left: 2649\n",
 424 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_0700_0800.parquet\n",
 425 |       "Non na entries: 28168\n",
 426 |       "Entries left: 872\n",
 427 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_0800_0900.parquet\n",
 428 |       "Non na entries: 28398\n",
 429 |       "Entries left: 451\n",
 430 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_0900_1000.parquet\n",
 431 |       "Non na entries: 22992\n",
 432 |       "Entries left: 464\n",
 433 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_1000_1100.parquet\n",
 434 |       "Non na entries: 26048\n",
 435 |       "Entries left: 855\n",
 436 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_1100_1200.parquet\n",
 437 |       "Non na entries: 27290\n",
 438 |       "Entries left: 1647\n",
 439 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_1200_1300.parquet\n",
 440 |       "Non na entries: 25123\n",
 441 |       "Entries left: 1035\n",
 442 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_1300_1400.parquet\n",
 443 |       "Non na entries: 22972\n",
 444 |       "Entries left: 1174\n",
 445 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_1400_1500.parquet\n",
 446 |       "Non na entries: 26767\n",
 447 |       "Entries left: 872\n",
 448 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_1500_1600.parquet\n",
 449 |       "Non na entries: 26669\n",
 450 |       "Entries left: 527\n",
 451 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_1600_1700.parquet\n",
 452 |       "Non na entries: 23612\n",
 453 |       "Entries left: 492\n",
 454 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_1700_1800.parquet\n",
 455 |       "Non na entries: 27038\n",
 456 |       "Entries left: 838\n",
 457 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_1800_1900.parquet\n",
 458 |       "Non na entries: 22585\n",
 459 |       "Entries left: 1282\n",
 460 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_1900_2000.parquet\n",
 461 |       "Non na entries: 25196\n",
 462 |       "Entries left: 848\n",
 463 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_2000_2100.parquet\n",
 464 |       "Non na entries: 20146\n",
 465 |       "Entries left: 740\n",
 466 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_2100_2200.parquet\n",
 467 |       "Non na entries: 16088\n",
 468 |       "Entries left: 608\n",
 469 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_2200_2300.parquet\n",
 470 |       "Non na entries: 17470\n",
 471 |       "Entries left: 1683\n",
 472 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_13_2300_0000.parquet\n",
 473 |       "Non na entries: 21760\n",
 474 |       "Entries left: 2280\n",
 475 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_0000_0100.parquet\n",
 476 |       "Non na entries: 26976\n",
 477 |       "Entries left: 1131\n",
 478 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_0100_0200.parquet\n",
 479 |       "Non na entries: 24938\n",
 480 |       "Entries left: 1166\n",
 481 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_0200_0300.parquet\n",
 482 |       "Non na entries: 27933\n",
 483 |       "Entries left: 1484\n",
 484 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_0300_0400.parquet\n",
 485 |       "Non na entries: 25346\n",
 486 |       "Entries left: 1344\n",
 487 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_0400_0500.parquet\n",
 488 |       "Non na entries: 32060\n",
 489 |       "Entries left: 2933\n",
 490 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_0500_0600.parquet\n",
 491 |       "Non na entries: 30437\n",
 492 |       "Entries left: 5557\n",
 493 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_0600_0700.parquet\n",
 494 |       "Non na entries: 26509\n",
 495 |       "Entries left: 2764\n",
 496 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_0700_0800.parquet\n",
 497 |       "Non na entries: 30204\n",
 498 |       "Entries left: 2073\n",
 499 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_0800_0900.parquet\n",
 500 |       "Non na entries: 30901\n",
 501 |       "Entries left: 4342\n",
 502 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_0900_1000.parquet\n",
 503 |       "Non na entries: 22159\n",
 504 |       "Entries left: 1763\n",
 505 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_1000_1100.parquet\n",
 506 |       "Non na entries: 21288\n",
 507 |       "Entries left: 1621\n",
 508 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_1100_1200.parquet\n",
 509 |       "Non na entries: 21730\n",
 510 |       "Entries left: 2560\n",
 511 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_1200_1300.parquet\n",
 512 |       "Non na entries: 23288\n",
 513 |       "Entries left: 2786\n",
 514 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_1300_1400.parquet\n",
 515 |       "Non na entries: 27118\n",
 516 |       "Entries left: 4641\n",
 517 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_1400_1500.parquet\n",
 518 |       "Non na entries: 23464\n",
 519 |       "Entries left: 3516\n",
 520 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_1500_1600.parquet\n",
 521 |       "Non na entries: 23731\n",
 522 |       "Entries left: 3475\n",
 523 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_1600_1700.parquet\n",
 524 |       "Non na entries: 24287\n",
 525 |       "Entries left: 4047\n",
 526 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_1700_1800.parquet\n",
 527 |       "Non na entries: 17221\n",
 528 |       "Entries left: 1142\n",
 529 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_1800_1900.parquet\n",
 530 |       "Non na entries: 19754\n",
 531 |       "Entries left: 1411\n",
 532 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_1900_2000.parquet\n",
 533 |       "Non na entries: 17295\n",
 534 |       "Entries left: 1696\n",
 535 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_2000_2100.parquet\n",
 536 |       "Non na entries: 23238\n",
 537 |       "Entries left: 1227\n",
 538 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_2100_2200.parquet\n",
 539 |       "Non na entries: 27619\n",
 540 |       "Entries left: 3305\n",
 541 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_2200_2300.parquet\n",
 542 |       "Non na entries: 27030\n",
 543 |       "Entries left: 1587\n",
 544 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_14_2300_0000.parquet\n",
 545 |       "Non na entries: 26169\n",
 546 |       "Entries left: 1979\n",
 547 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_0000_0100.parquet\n"
 548 |      ]
 549 |     },
 550 |     {
 551 |      "name": "stdout",
 552 |      "output_type": "stream",
 553 |      "text": [
 554 |       "Non na entries: 27521\n",
 555 |       "Entries left: 1087\n",
 556 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_0100_0200.parquet\n",
 557 |       "Non na entries: 24930\n",
 558 |       "Entries left: 2105\n",
 559 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_0200_0300.parquet\n",
 560 |       "Non na entries: 23995\n",
 561 |       "Entries left: 3622\n",
 562 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_0300_0400.parquet\n",
 563 |       "Non na entries: 23348\n",
 564 |       "Entries left: 3409\n",
 565 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_0400_0500.parquet\n",
 566 |       "Non na entries: 23337\n",
 567 |       "Entries left: 3592\n",
 568 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_0500_0600.parquet\n",
 569 |       "Non na entries: 20327\n",
 570 |       "Entries left: 1931\n",
 571 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_0600_0700.parquet\n",
 572 |       "Non na entries: 19324\n",
 573 |       "Entries left: 2648\n",
 574 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_0700_0800.parquet\n",
 575 |       "Non na entries: 19165\n",
 576 |       "Entries left: 1062\n",
 577 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_0800_0900.parquet\n",
 578 |       "Non na entries: 19953\n",
 579 |       "Entries left: 2897\n",
 580 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_0900_1000.parquet\n",
 581 |       "Non na entries: 28001\n",
 582 |       "Entries left: 4639\n",
 583 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_1000_1100.parquet\n",
 584 |       "Non na entries: 22787\n",
 585 |       "Entries left: 2844\n",
 586 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_1100_1200.parquet\n",
 587 |       "Non na entries: 23247\n",
 588 |       "Entries left: 4012\n",
 589 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_1200_1300.parquet\n",
 590 |       "Non na entries: 25178\n",
 591 |       "Entries left: 4377\n",
 592 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_1300_1400.parquet\n",
 593 |       "Non na entries: 23804\n",
 594 |       "Entries left: 5390\n",
 595 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_1400_1500.parquet\n",
 596 |       "Non na entries: 31365\n",
 597 |       "Entries left: 4904\n",
 598 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_1500_1600.parquet\n",
 599 |       "Non na entries: 28866\n",
 600 |       "Entries left: 4270\n",
 601 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_1600_1700.parquet\n",
 602 |       "Non na entries: 30040\n",
 603 |       "Entries left: 6297\n",
 604 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_1700_1800.parquet\n",
 605 |       "Non na entries: 22322\n",
 606 |       "Entries left: 2077\n",
 607 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_1800_1900.parquet\n",
 608 |       "Non na entries: 24886\n",
 609 |       "Entries left: 4068\n",
 610 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_1900_2000.parquet\n",
 611 |       "Non na entries: 24876\n",
 612 |       "Entries left: 3511\n",
 613 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_2000_2100.parquet\n",
 614 |       "Non na entries: 23560\n",
 615 |       "Entries left: 4554\n",
 616 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_2100_2200.parquet\n",
 617 |       "Non na entries: 22572\n",
 618 |       "Entries left: 3836\n",
 619 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_2200_2300.parquet\n",
 620 |       "Non na entries: 20768\n",
 621 |       "Entries left: 4600\n",
 622 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_15_2300_0000.parquet\n",
 623 |       "Non na entries: 26129\n",
 624 |       "Entries left: 4102\n",
 625 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_0000_0100.parquet\n",
 626 |       "Non na entries: 23843\n",
 627 |       "Entries left: 4284\n",
 628 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_0100_0200.parquet\n",
 629 |       "Non na entries: 22470\n",
 630 |       "Entries left: 4241\n",
 631 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_0200_0300.parquet\n",
 632 |       "Non na entries: 25021\n",
 633 |       "Entries left: 6903\n",
 634 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_0300_0400.parquet\n",
 635 |       "Non na entries: 20527\n",
 636 |       "Entries left: 5106\n",
 637 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_0400_0500.parquet\n",
 638 |       "Non na entries: 21365\n",
 639 |       "Entries left: 4574\n",
 640 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_0500_0600.parquet\n",
 641 |       "Non na entries: 21585\n",
 642 |       "Entries left: 5946\n",
 643 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_0600_0700.parquet\n",
 644 |       "Non na entries: 20304\n",
 645 |       "Entries left: 5802\n",
 646 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_0700_0800.parquet\n",
 647 |       "Non na entries: 22604\n",
 648 |       "Entries left: 4469\n",
 649 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_0800_0900.parquet\n",
 650 |       "Non na entries: 19245\n",
 651 |       "Entries left: 3182\n",
 652 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_0900_1000.parquet\n",
 653 |       "Non na entries: 22239\n",
 654 |       "Entries left: 4182\n",
 655 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_1000_1100.parquet\n",
 656 |       "Non na entries: 26071\n",
 657 |       "Entries left: 3290\n",
 658 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_1100_1200.parquet\n",
 659 |       "Non na entries: 21796\n",
 660 |       "Entries left: 5027\n",
 661 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_1200_1300.parquet\n",
 662 |       "Non na entries: 16273\n",
 663 |       "Entries left: 589\n",
 664 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_1300_1400.parquet\n",
 665 |       "Non na entries: 20885\n",
 666 |       "Entries left: 2690\n",
 667 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_1400_1500.parquet\n",
 668 |       "Non na entries: 24523\n",
 669 |       "Entries left: 4340\n",
 670 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_1500_1600.parquet\n",
 671 |       "Non na entries: 18443\n",
 672 |       "Entries left: 1550\n",
 673 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_1600_1700.parquet\n",
 674 |       "Non na entries: 18488\n",
 675 |       "Entries left: 1762\n",
 676 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_1700_1800.parquet\n",
 677 |       "Non na entries: 13384\n",
 678 |       "Entries left: 329\n",
 679 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_1800_1900.parquet\n",
 680 |       "Non na entries: 12533\n",
 681 |       "Entries left: 307\n",
 682 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_1900_2000.parquet\n",
 683 |       "Non na entries: 13175\n",
 684 |       "Entries left: 1285\n",
 685 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_2000_2100.parquet\n",
 686 |       "Non na entries: 12884\n",
 687 |       "Entries left: 729\n",
 688 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_2100_2200.parquet\n",
 689 |       "Non na entries: 12538\n",
 690 |       "Entries left: 600\n",
 691 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_2200_2300.parquet\n",
 692 |       "Non na entries: 15634\n",
 693 |       "Entries left: 1547\n",
 694 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_16_2300_0000.parquet\n",
 695 |       "Non na entries: 10401\n",
 696 |       "Entries left: 373\n",
 697 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_0000_0100.parquet\n",
 698 |       "Non na entries: 17868\n",
 699 |       "Entries left: 2419\n",
 700 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_0100_0200.parquet\n",
 701 |       "Non na entries: 15572\n",
 702 |       "Entries left: 1242\n",
 703 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_0200_0300.parquet\n",
 704 |       "Non na entries: 21304\n",
 705 |       "Entries left: 7847\n",
 706 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_0300_0400.parquet\n",
 707 |       "Non na entries: 22867\n",
 708 |       "Entries left: 4924\n",
 709 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_0400_0500.parquet\n",
 710 |       "Non na entries: 18906\n",
 711 |       "Entries left: 4928\n",
 712 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_0500_0600.parquet\n",
 713 |       "Non na entries: 20459\n",
 714 |       "Entries left: 5253\n",
 715 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_0600_0700.parquet\n",
 716 |       "Non na entries: 28776\n",
 717 |       "Entries left: 1699\n",
 718 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_0700_0800.parquet\n",
 719 |       "Non na entries: 27138\n",
 720 |       "Entries left: 2122\n",
 721 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_0800_0900.parquet\n",
 722 |       "Non na entries: 26465\n",
 723 |       "Entries left: 5014\n",
 724 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_0900_1000.parquet\n",
 725 |       "Non na entries: 24802\n",
 726 |       "Entries left: 6052\n",
 727 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_1000_1100.parquet\n",
 728 |       "Non na entries: 15558\n",
 729 |       "Entries left: 975\n",
 730 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_1100_1200.parquet\n",
 731 |       "Non na entries: 16872\n",
 732 |       "Entries left: 1729\n",
 733 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_1200_1300.parquet\n",
 734 |       "Non na entries: 20356\n",
 735 |       "Entries left: 2549\n",
 736 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_1300_1400.parquet\n",
 737 |       "Non na entries: 28232\n",
 738 |       "Entries left: 3189\n",
 739 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_1400_1500.parquet\n",
 740 |       "Non na entries: 19224\n",
 741 |       "Entries left: 3984\n",
 742 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_1500_1600.parquet\n",
 743 |       "Non na entries: 14991\n",
 744 |       "Entries left: 1114\n",
 745 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_1600_1700.parquet\n",
 746 |       "Non na entries: 19174\n",
 747 |       "Entries left: 1791\n",
 748 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_1700_1800.parquet\n",
 749 |       "Non na entries: 16804\n",
 750 |       "Entries left: 432\n",
 751 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_1800_1900.parquet\n",
 752 |       "Non na entries: 17178\n",
 753 |       "Entries left: 1167\n",
 754 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_1900_2000.parquet\n",
 755 |       "Non na entries: 17797\n",
 756 |       "Entries left: 638\n",
 757 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_2000_2100.parquet\n",
 758 |       "Non na entries: 22174\n",
 759 |       "Entries left: 1353\n",
 760 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_2100_2200.parquet\n",
 761 |       "Non na entries: 19309\n",
 762 |       "Entries left: 1007\n",
 763 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_2200_2300.parquet\n",
 764 |       "Non na entries: 16829\n",
 765 |       "Entries left: 1665\n",
 766 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_17_2300_0000.parquet\n",
 767 |       "Non na entries: 22010\n",
 768 |       "Entries left: 2108\n",
 769 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_0000_0100.parquet\n",
 770 |       "Non na entries: 23182\n",
 771 |       "Entries left: 1040\n",
 772 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_0100_0200.parquet\n"
 773 |      ]
 774 |     },
 775 |     {
 776 |      "name": "stdout",
 777 |      "output_type": "stream",
 778 |      "text": [
 779 |       "Non na entries: 18296\n",
 780 |       "Entries left: 485\n",
 781 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_0200_0300.parquet\n",
 782 |       "Non na entries: 22012\n",
 783 |       "Entries left: 770\n",
 784 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_0300_0400.parquet\n",
 785 |       "Non na entries: 22133\n",
 786 |       "Entries left: 1118\n",
 787 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_0400_0500.parquet\n",
 788 |       "Non na entries: 21448\n",
 789 |       "Entries left: 330\n",
 790 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_0500_0600.parquet\n",
 791 |       "Non na entries: 19851\n",
 792 |       "Entries left: 579\n",
 793 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_0600_0700.parquet\n",
 794 |       "Non na entries: 19670\n",
 795 |       "Entries left: 521\n",
 796 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_0700_0800.parquet\n",
 797 |       "Non na entries: 18773\n",
 798 |       "Entries left: 302\n",
 799 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_0800_0900.parquet\n",
 800 |       "Non na entries: 19211\n",
 801 |       "Entries left: 741\n",
 802 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_0900_1000.parquet\n",
 803 |       "Non na entries: 15352\n",
 804 |       "Entries left: 137\n",
 805 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_1000_1100.parquet\n",
 806 |       "Non na entries: 19082\n",
 807 |       "Entries left: 184\n",
 808 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_1100_1200.parquet\n",
 809 |       "Non na entries: 20401\n",
 810 |       "Entries left: 298\n",
 811 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_1200_1300.parquet\n",
 812 |       "Non na entries: 24121\n",
 813 |       "Entries left: 925\n",
 814 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_1300_1400.parquet\n",
 815 |       "Non na entries: 21732\n",
 816 |       "Entries left: 1097\n",
 817 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_1400_1500.parquet\n",
 818 |       "Non na entries: 16748\n",
 819 |       "Entries left: 406\n",
 820 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_1500_1600.parquet\n",
 821 |       "Non na entries: 19776\n",
 822 |       "Entries left: 516\n",
 823 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_1600_1700.parquet\n",
 824 |       "Non na entries: 16940\n",
 825 |       "Entries left: 225\n",
 826 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_1700_1800.parquet\n",
 827 |       "Non na entries: 26566\n",
 828 |       "Entries left: 2181\n",
 829 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_1800_1900.parquet\n",
 830 |       "Non na entries: 31398\n",
 831 |       "Entries left: 5146\n",
 832 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_1900_2000.parquet\n",
 833 |       "Non na entries: 25755\n",
 834 |       "Entries left: 3526\n",
 835 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_2000_2100.parquet\n",
 836 |       "Non na entries: 21328\n",
 837 |       "Entries left: 794\n",
 838 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_2100_2200.parquet\n",
 839 |       "Non na entries: 22571\n",
 840 |       "Entries left: 866\n",
 841 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_2200_2300.parquet\n",
 842 |       "Non na entries: 21180\n",
 843 |       "Entries left: 1593\n",
 844 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_18_2300_0000.parquet\n",
 845 |       "Non na entries: 18567\n",
 846 |       "Entries left: 445\n",
 847 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_0000_0100.parquet\n",
 848 |       "Non na entries: 21472\n",
 849 |       "Entries left: 464\n",
 850 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_0100_0200.parquet\n",
 851 |       "Non na entries: 22268\n",
 852 |       "Entries left: 601\n",
 853 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_0200_0300.parquet\n",
 854 |       "Non na entries: 18284\n",
 855 |       "Entries left: 713\n",
 856 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_0300_0400.parquet\n",
 857 |       "Non na entries: 21178\n",
 858 |       "Entries left: 703\n",
 859 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_0400_0500.parquet\n",
 860 |       "Non na entries: 24032\n",
 861 |       "Entries left: 407\n",
 862 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_0500_0600.parquet\n",
 863 |       "Non na entries: 21356\n",
 864 |       "Entries left: 713\n",
 865 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_0600_0700.parquet\n",
 866 |       "Non na entries: 21864\n",
 867 |       "Entries left: 281\n",
 868 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_0700_0800.parquet\n",
 869 |       "Non na entries: 22011\n",
 870 |       "Entries left: 271\n",
 871 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_0800_0900.parquet\n",
 872 |       "Non na entries: 23630\n",
 873 |       "Entries left: 348\n",
 874 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_0900_1000.parquet\n",
 875 |       "Non na entries: 19436\n",
 876 |       "Entries left: 155\n",
 877 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_1000_1100.parquet\n",
 878 |       "Non na entries: 18938\n",
 879 |       "Entries left: 348\n",
 880 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_1100_1200.parquet\n",
 881 |       "Non na entries: 28521\n",
 882 |       "Entries left: 1142\n",
 883 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_1200_1300.parquet\n",
 884 |       "Non na entries: 26141\n",
 885 |       "Entries left: 975\n",
 886 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_1300_1400.parquet\n",
 887 |       "Non na entries: 26839\n",
 888 |       "Entries left: 1325\n",
 889 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_1400_1500.parquet\n",
 890 |       "Non na entries: 25620\n",
 891 |       "Entries left: 1884\n",
 892 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_1500_1600.parquet\n",
 893 |       "Non na entries: 21827\n",
 894 |       "Entries left: 1053\n",
 895 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_1600_1700.parquet\n",
 896 |       "Non na entries: 23805\n",
 897 |       "Entries left: 1587\n",
 898 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_1700_1800.parquet\n",
 899 |       "Non na entries: 22752\n",
 900 |       "Entries left: 1485\n",
 901 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_1800_1900.parquet\n",
 902 |       "Non na entries: 21289\n",
 903 |       "Entries left: 851\n",
 904 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_1900_2000.parquet\n",
 905 |       "Non na entries: 20266\n",
 906 |       "Entries left: 1935\n",
 907 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_2000_2100.parquet\n",
 908 |       "Non na entries: 22559\n",
 909 |       "Entries left: 3647\n",
 910 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_2100_2200.parquet\n",
 911 |       "Non na entries: 17410\n",
 912 |       "Entries left: 2039\n",
 913 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_2200_2300.parquet\n",
 914 |       "Non na entries: 20513\n",
 915 |       "Entries left: 2285\n",
 916 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_19_2300_0000.parquet\n",
 917 |       "Non na entries: 16284\n",
 918 |       "Entries left: 386\n",
 919 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_0000_0100.parquet\n",
 920 |       "Non na entries: 20526\n",
 921 |       "Entries left: 725\n",
 922 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_0100_0200.parquet\n",
 923 |       "Non na entries: 19294\n",
 924 |       "Entries left: 547\n",
 925 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_0200_0300.parquet\n",
 926 |       "Non na entries: 16827\n",
 927 |       "Entries left: 597\n",
 928 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_0300_0400.parquet\n",
 929 |       "Non na entries: 16628\n",
 930 |       "Entries left: 325\n",
 931 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_0400_0500.parquet\n",
 932 |       "Non na entries: 17306\n",
 933 |       "Entries left: 350\n",
 934 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_0500_0600.parquet\n",
 935 |       "Non na entries: 18860\n",
 936 |       "Entries left: 399\n",
 937 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_0600_0700.parquet\n",
 938 |       "Non na entries: 16651\n",
 939 |       "Entries left: 376\n",
 940 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_0700_0800.parquet\n",
 941 |       "Non na entries: 15018\n",
 942 |       "Entries left: 110\n",
 943 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_0800_0900.parquet\n",
 944 |       "Non na entries: 19297\n",
 945 |       "Entries left: 143\n",
 946 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_0900_1000.parquet\n",
 947 |       "Non na entries: 18229\n",
 948 |       "Entries left: 173\n",
 949 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_1000_1100.parquet\n",
 950 |       "Non na entries: 21159\n",
 951 |       "Entries left: 65\n",
 952 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_1100_1200.parquet\n",
 953 |       "Non na entries: 14720\n",
 954 |       "Entries left: 102\n",
 955 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_1200_1300.parquet\n",
 956 |       "Non na entries: 20426\n",
 957 |       "Entries left: 281\n",
 958 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_1300_1400.parquet\n",
 959 |       "Non na entries: 16789\n",
 960 |       "Entries left: 182\n",
 961 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_1400_1500.parquet\n",
 962 |       "Non na entries: 19182\n",
 963 |       "Entries left: 364\n",
 964 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_1500_1600.parquet\n",
 965 |       "Non na entries: 26219\n",
 966 |       "Entries left: 1640\n",
 967 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_1600_1700.parquet\n",
 968 |       "Non na entries: 18598\n",
 969 |       "Entries left: 304\n",
 970 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_1700_1800.parquet\n",
 971 |       "Non na entries: 16120\n",
 972 |       "Entries left: 305\n",
 973 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_1800_1900.parquet\n",
 974 |       "Non na entries: 29157\n",
 975 |       "Entries left: 1472\n",
 976 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_1900_2000.parquet\n",
 977 |       "Non na entries: 21949\n",
 978 |       "Entries left: 494\n",
 979 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_2000_2100.parquet\n",
 980 |       "Non na entries: 21123\n",
 981 |       "Entries left: 558\n",
 982 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_2100_2200.parquet\n",
 983 |       "Non na entries: 17445\n",
 984 |       "Entries left: 365\n",
 985 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_2200_2300.parquet\n",
 986 |       "Non na entries: 22824\n",
 987 |       "Entries left: 777\n",
 988 |       "coinbase_btc_usd_l2_book_snapshots_depth50_2019_11_20_2300_0000.parquet\n",
 989 |       "Non na entries: 21173\n",
 990 |       "Entries left: 1124\n"
 991 |      ]
 992 |     }
 993 |    ],
 994 |    "source": [
 995 |     "i = 0\n",
 996 |     "for filename in os.listdir(r\"Data\\coinbase_btc_usd\\coinbase\\btc_usd\\l2_snapshots\\100ms\"):\n",
 997 |     "    print(filename)\n",
 998 |     "    \n",
 999 |     "    df_snapshot = pd.read_parquet(os.path.join(r\"Data\\coinbase_btc_usd\\coinbase\\btc_usd\\l2_snapshots\\100ms\", filename))\n",
1000 |     "    \n",
1001 |     "    df_snapshot.dropna(inplace=True) #Drop na values\n",
1002 |     "    print('Non na entries:', np.shape(df_snapshot)[0])\n",
1003 |     "\n",
1004 |     "    # Check no price levels with zero quantities:\n",
1005 |     "    assert np.count_nonzero(df_snapshot.values) == np.shape(df_snapshot)[0] * np.shape(df_snapshot)[1]\n",
1006 |     "\n",
1007 |     "    df_snapshot['future_bid'] = df_snapshot['b1'].shift(-1)\n",
1008 |     "\n",
1009 |     "    # Calculate how many levels the bid/ask price changed by\n",
1010 |     "    df_snapshot['bid_change_n'] = df_snapshot.apply(lambda x: 100*(x['future_bid'] - x['b1']), axis=1)\n",
1011 |     "\n",
1012 |     "    #Delete last one as don't know how much it has changed by\n",
1013 |     "    df_snapshot.drop(df_snapshot.index[-1], axis=0, inplace=True)\n",
1014 |     "\n",
1015 |     "    # Ignore the prices columns\n",
1016 |     "    df_snapshot_cut = df_snapshot[df_snapshot.columns[100:]]\n",
1017 |     "\n",
1018 |     "    # Take only the values where the bid price increases\n",
1019 |     "    df_bid_up = df_snapshot_cut[df_snapshot_cut['bid_change_n'] > 0].drop(['future_bid'], axis=1)\n",
1020 |     "\n",
1021 |     "    # How many entries in this hour left?\n",
1022 |     "    print('Entries left:', np.shape(df_bid_up)[0])\n",
1023 |     "    \n",
1024 |     "    if i==0:\n",
1025 |     "        df = df_bid_up\n",
1026 |     "        i += 1\n",
1027 |     "    else:\n",
1028 |     "        df = df.append(df_bid_up)\n"
1029 |    ]
1030 |   },
1031 |   {
1032 |    "cell_type": "code",
1033 |    "execution_count": 205,
1034 |    "metadata": {},
1035 |    "outputs": [],
1036 |    "source": [
1037 |     "df.to_csv(\"df_snapshot_100ms_bid_up.csv\")"
1038 |    ]
1039 |   },
1040 |   {
1041 |    "cell_type": "code",
1042 |    "execution_count": 21,
1043 |    "metadata": {},
1044 |    "outputs": [
1045 |     {
1046 |      "data": {
1047 |       "text/plain": [
1048 |        "(array([   1,    2,    3,    4,    5,    6,    7,    8,    9,   10,   11,\n",
1049 |        "          12,   13,   14,   15,   16,   17,   18,   19,   20,   21,   22,\n",
1050 |        "          23,   24,   25,   26,   27,   28,   29,   30,   31,   32,   33,\n",
1051 |        "          34,   35,   36,   37,   38,   39,   40,   41,   42,   43,   44,\n",
1052 |        "          45,   46,   47,   48,   49,   50,   51,   52,   53,   54,   55,\n",
1053 |        "          56,   57,   58,   59,   60,   61,   62,   63,   64,   65,   66,\n",
1054 |        "          67,   68,   69,   70,   71,   72,   73,   74,   75,   76,   77,\n",
1055 |        "          78,   79,   80,   81,   82,   83,   84,   85,   86,   87,   88,\n",
1056 |        "          89,   90,   91,   92,   93,   94,   95,   96,   97,   98,   99,\n",
1057 |        "         100,  101,  102,  103,  104,  105,  106,  107,  108,  109,  110,\n",
1058 |        "         111,  112,  113,  114,  115,  116,  117,  118,  119,  120,  121,\n",
1059 |        "         122,  123,  124,  125,  126,  127,  128,  129,  130,  131,  132,\n",
1060 |        "         133,  134,  135,  136,  137,  138,  139,  140,  141,  142,  143,\n",
1061 |        "         144,  145,  146,  147,  148,  149,  150,  151,  152,  153,  154,\n",
1062 |        "         155,  156,  157,  158,  159,  160,  161,  162,  163,  164,  165,\n",
1063 |        "         166,  167,  168,  169,  170,  171,  172,  173,  174,  175,  176,\n",
1064 |        "         177,  178,  179,  180,  181,  182,  183,  184,  185,  186,  187,\n",
1065 |        "         188,  189,  190,  191,  192,  193,  194,  195,  196,  197,  198,\n",
1066 |        "         199,  200,  201,  202,  203,  204,  205,  206,  207,  208,  209,\n",
1067 |        "         210,  211,  212,  213,  214,  215,  216,  217,  218,  219,  220,\n",
1068 |        "         221,  222,  223,  224,  225,  226,  227,  228,  229,  230,  231,\n",
1069 |        "         232,  233,  234,  235,  236,  237,  238,  239,  240,  241,  242,\n",
1070 |        "         243,  244,  245,  246,  247,  248,  249,  250,  251,  252,  253,\n",
1071 |        "         254,  255,  256,  257,  258,  259,  260,  261,  262,  263,  264,\n",
1072 |        "         265,  266,  267,  268,  269,  270,  271,  272,  273,  274,  275,\n",
1073 |        "         276,  277,  278,  279,  280,  281,  282,  283,  284,  285,  286,\n",
1074 |        "         287,  288,  289,  290,  291,  292,  293,  294,  295,  296,  297,\n",
1075 |        "         298,  299,  300,  301,  302,  303,  304,  305,  306,  307,  308,\n",
1076 |        "         309,  310,  311,  312,  313,  314,  315,  316,  317,  318,  319,\n",
1077 |        "         320,  321,  322,  323,  324,  325,  326,  327,  328,  329,  330,\n",
1078 |        "         331,  332,  333,  334,  335,  336,  337,  338,  339,  340,  341,\n",
1079 |        "         342,  343,  344,  345,  346,  347,  348,  349,  350,  351,  352,\n",
1080 |        "         353,  354,  355,  356,  357,  358,  359,  360,  361,  362,  363,\n",
1081 |        "         364,  365,  366,  367,  368,  369,  370,  371,  372,  373,  374,\n",
1082 |        "         375,  376,  377,  378,  379,  380,  381,  382,  383,  384,  385,\n",
1083 |        "         386,  387,  388,  389,  390,  391,  392,  393,  394,  395,  396,\n",
1084 |        "         397,  398,  399,  400,  401,  402,  403,  404,  405,  406,  407,\n",
1085 |        "         408,  409,  410,  411,  412,  413,  414,  415,  416,  417,  418,\n",
1086 |        "         419,  420,  421,  422,  423,  424,  425,  426,  427,  428,  429,\n",
1087 |        "         430,  431,  432,  433,  434,  435,  436,  437,  438,  439,  440,\n",
1088 |        "         441,  442,  443,  444,  445,  446,  447,  448,  449,  450,  451,\n",
1089 |        "         452,  453,  454,  455,  456,  457,  458,  459,  460,  461,  462,\n",
1090 |        "         463,  464,  465,  466,  467,  468,  469,  472,  473,  474,  475,\n",
1091 |        "         476,  477,  478,  479,  480,  481,  482,  483,  485,  486,  487,\n",
1092 |        "         488,  489,  490,  491,  492,  493,  494,  495,  496,  497,  498,\n",
1093 |        "         499,  500,  501,  502,  503,  504,  505,  506,  507,  508,  509,\n",
1094 |        "         510,  511,  512,  513,  514,  515,  516,  517,  518,  519,  520,\n",
1095 |        "         521,  522,  523,  525,  528,  530,  531,  532,  536,  537,  538,\n",
1096 |        "         539,  540,  542,  543,  544,  546,  547,  549,  550,  551,  552,\n",
1097 |        "         553,  554,  556,  557,  558,  559,  560,  566,  568,  569,  570,\n",
1098 |        "         571,  572,  573,  574,  575,  576,  577,  578,  579,  582,  584,\n",
1099 |        "         586,  587,  588,  589,  591,  594,  597,  598,  600,  601,  603,\n",
1100 |        "         604,  607,  610,  611,  613,  615,  616,  620,  621,  622,  623,\n",
1101 |        "         625,  630,  632,  634,  635,  642,  643,  645,  650,  651,  652,\n",
1102 |        "         653,  655,  656,  663,  664,  667,  671,  673,  674,  676,  677,\n",
1103 |        "         678,  684,  688,  691,  693,  696,  699,  700,  708,  714,  715,\n",
1104 |        "         716,  718,  720,  722,  723,  727,  728,  733,  735,  736,  742,\n",
1105 |        "         748,  749,  750,  751,  752,  762,  765,  768,  772,  774,  775,\n",
1106 |        "         776,  777,  786,  787,  788,  790,  797,  799,  800,  801,  817,\n",
1107 |        "         824,  828,  832,  837,  854,  857,  858,  860,  867,  874,  895,\n",
1108 |        "         896,  901,  905,  913,  918,  929,  950,  953,  960,  979, 1002,\n",
1109 |        "        1023, 1052, 1088, 1103, 1148, 1162, 1185, 1197, 1218, 1227, 1232,\n",
1110 |        "        1234, 1264, 1330, 1364, 1373, 1495, 1499, 1586, 2564, 2733]),\n",
1111 |        " array([105912, 106103,  92412,  69731,  13148,   8756,   3700,   1877,\n",
1112 |        "          1493,   1745,   1480,   1163,   1050,    814,    731,    636,\n",
1113 |        "           575,    542,    555,    603,    523,    479,    438,    449,\n",
1114 |        "           474,    444,    450,    413,    420,    380,    375,    397,\n",
1115 |        "           376,    351,    375,    360,    358,    362,    372,    382,\n",
1116 |        "           349,    334,    363,    341,    360,    355,    376,    372,\n",
1117 |        "           419,    572,    475,    373,    295,    303,    278,    298,\n",
1118 |        "           306,    309,    289,    306,    333,    256,    266,    278,\n",
1119 |        "           284,    239,    286,    250,    264,    250,    266,    280,\n",
1120 |        "           250,    264,    243,    266,    377,    432,    335,    292,\n",
1121 |        "           234,    244,    253,    237,    248,    250,    263,    269,\n",
1122 |        "           293,    295,    288,    306,    357,    330,    294,    371,\n",
1123 |        "           429,    490,    651,   1268,    237,    189,    140,    142,\n",
1124 |        "           102,    116,    173,    118,    118,    108,    129,    101,\n",
1125 |        "           101,    101,    106,    111,    109,     89,     85,    105,\n",
1126 |        "           109,     99,     92,     90,     87,     78,     76,     82,\n",
1127 |        "            87,     88,     88,    102,     86,     74,     90,     83,\n",
1128 |        "            75,     86,     86,     76,     68,     73,     81,     75,\n",
1129 |        "            66,     72,     76,     73,     79,     89,     91,     77,\n",
1130 |        "            88,     63,     71,     77,     68,     83,     68,     53,\n",
1131 |        "            57,     54,     59,     68,     62,     56,     79,     59,\n",
1132 |        "            82,     81,     67,     54,     49,     63,     72,     71,\n",
1133 |        "            52,     53,     64,     61,     64,     50,     46,     49,\n",
1134 |        "            51,     54,     63,     52,     63,     47,     42,     48,\n",
1135 |        "            43,     56,     51,     53,     62,     63,     77,     84,\n",
1136 |        "            67,     53,     44,     39,     35,     50,     38,     42,\n",
1137 |        "            30,     44,     35,     34,     48,     44,     43,     36,\n",
1138 |        "            38,     33,     30,     39,     37,     33,     29,     24,\n",
1139 |        "            31,     44,     49,     38,     40,     33,     34,     37,\n",
1140 |        "            30,     26,     48,     28,     25,     31,     25,     27,\n",
1141 |        "            17,     18,     27,     23,     26,     32,     21,     27,\n",
1142 |        "            23,     29,     15,     34,     19,     28,     31,     18,\n",
1143 |        "            25,     35,     21,     21,     26,     20,     21,     22,\n",
1144 |        "            27,     30,     21,     22,     31,     17,     28,     17,\n",
1145 |        "            25,     19,     23,     20,     10,     19,     11,     14,\n",
1146 |        "            20,     21,     18,     29,     10,     21,     13,     16,\n",
1147 |        "            20,     22,     16,     20,     18,     12,     21,     18,\n",
1148 |        "            17,     28,      9,     16,     22,     15,     20,     16,\n",
1149 |        "            15,     11,     18,     18,      6,     18,     13,     15,\n",
1150 |        "            22,      7,     16,      8,     14,     11,     10,     27,\n",
1151 |        "            12,     21,     23,      8,     13,     20,     14,     13,\n",
1152 |        "             9,     11,     19,      8,     14,     12,     17,     31,\n",
1153 |        "            23,     13,     21,     14,     14,     14,     19,     11,\n",
1154 |        "            11,     12,     11,     13,     10,      4,      6,     11,\n",
1155 |        "             6,     10,     10,      3,      8,     10,      6,      8,\n",
1156 |        "             9,      9,      4,      7,      7,      4,      7,      6,\n",
1157 |        "             4,      6,      1,      5,      6,      3,      7,      5,\n",
1158 |        "            14,     10,      7,      4,      6,      4,      7,      6,\n",
1159 |        "             7,      8,      6,      9,      5,      3,      4,      2,\n",
1160 |        "             5,     12,      8,      6,     14,      7,      1,      9,\n",
1161 |        "             1,      4,      4,      4,      2,      3,      3,      6,\n",
1162 |        "             3,      7,      4,      6,      7,      6,      2,      6,\n",
1163 |        "             6,      6,      5,      3,      2,      4,      4,      5,\n",
1164 |        "             4,      6,      3,      6,      4,      1,      2,      7,\n",
1165 |        "             8,      5,      5,      6,      6,      2,      4,      1,\n",
1166 |        "             5,      4,      4,      1,      3,      2,      2,      3,\n",
1167 |        "             1,      2,      4,      7,      2,      2,      5,      3,\n",
1168 |        "             2,      3,      4,      2,      5,      3,      2,      3,\n",
1169 |        "             3,      5,      2,      1,      1,      4,      1,      2,\n",
1170 |        "             2,      3,      1,      3,      2,      1,      5,      3,\n",
1171 |        "             2,      3,      1,      2,      3,      2,      2,      2,\n",
1172 |        "             3,      2,      1,      1,      2,      1,      4,      2,\n",
1173 |        "             4,      1,      5,      3,      1,      1,      2,      3,\n",
1174 |        "             1,      4,      3,      1,      1,      3,      4,      1,\n",
1175 |        "             1,      1,      2,      1,      2,      1,      1,      1,\n",
1176 |        "             1,      3,      2,      1,      3,      2,      1,      2,\n",
1177 |        "             2,      2,      1,      1,      2,      2,      1,      2,\n",
1178 |        "             1,      1,      1,      1,      3,      4,      1,      2,\n",
1179 |        "             1,      1,      3,      2,      3,      1,      2,      1,\n",
1180 |        "             1,      1,      3,      1,      2,      2,      1,      1,\n",
1181 |        "             1,      1,      2,      1,      1,      3,      1,      1,\n",
1182 |        "             2,      3,      3,      1,      1,      1,      3,      1,\n",
1183 |        "             1,      1,      1,      2,      1,      1,      2,      1,\n",
1184 |        "             1,      1,      1,      2,      3,      1,      1,      3,\n",
1185 |        "             2,      1,      1,      1,      2,      1,      1,      1,\n",
1186 |        "             1,      1,      1,      1,      1,      2,      2,      1,\n",
1187 |        "             2,      2,      1,      1,      1,      1,      1,      2,\n",
1188 |        "             1,      1,      1,      1,      1,      1,      2,      1,\n",
1189 |        "             2,      1,      1,      1,      1,      1,      1,      1,\n",
1190 |        "             1,      1,      1,      1,      1,      1,      1,      1,\n",
1191 |        "             1,      1,      1,      1,      1,      1,      1,      1,\n",
1192 |        "             1,      1,      1,      1,      1,      1,      1,      2,\n",
1193 |        "             1,      1,      1,      1,      1,      1,      1,      1,\n",
1194 |        "             2,      1,      1,      1,      1,      1,      1,      1,\n",
1195 |        "             1,      1,      1,      1,      1,      1,      1,      1,\n",
1196 |        "             1,      1,      1,      1,      1,      1,      1,      1,\n",
1197 |        "             1,      1,      1,      1], dtype=int64))"
1198 |       ]
1199 |      },
1200 |      "execution_count": 21,
1201 |      "metadata": {},
1202 |      "output_type": "execute_result"
1203 |     }
1204 |    ],
1205 |    "source": [
1206 |     "value_counts = np.unique(np.round(df_bid_up['bid_change_n'].values).astype(int), return_counts=True)\n",
1207 |     "value_counts"
1208 |    ]
1209 |   },
1210 |   {
1211 |    "cell_type": "code",
1212 |    "execution_count": null,
1213 |    "metadata": {},
1214 |    "outputs": [],
1215 |    "source": []
1216 |   }
1217 |  ],
1218 |  "metadata": {
1219 |   "kernelspec": {
1220 |    "display_name": "Python 3",
1221 |    "language": "python",
1222 |    "name": "python3"
1223 |   },
1224 |   "language_info": {
1225 |    "codemirror_mode": {
1226 |     "name": "ipython",
1227 |     "version": 3
1228 |    },
1229 |    "file_extension": ".py",
1230 |    "mimetype": "text/x-python",
1231 |    "name": "python",
1232 |    "nbconvert_exporter": "python",
1233 |    "pygments_lexer": "ipython3",
1234 |    "version": "3.7.5"
1235 |   }
1236 |  },
1237 |  "nbformat": 4,
1238 |  "nbformat_minor": 2
1239 | }
1240 | 


--------------------------------------------------------------------------------
/LICENSE:
--------------------------------------------------------------------------------
 1 | MIT License
 2 | 
 3 | Copyright (c) 2019 Globe Research
 4 | 
 5 | Permission is hereby granted, free of charge, to any person obtaining a copy
 6 | of this software and associated documentation files (the "Software"), to deal
 7 | in the Software without restriction, including without limitation the rights
 8 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
 9 | copies of the Software, and to permit persons to whom the Software is
10 | furnished to do so, subject to the following conditions:
11 | 
12 | The above copyright notice and this permission notice shall be included in all
13 | copies or substantial portions of the Software.
14 | 
15 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
16 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
17 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
18 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
19 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
20 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
21 | SOFTWARE.
22 | 


--------------------------------------------------------------------------------
/README.md:
--------------------------------------------------------------------------------
1 | # Deep Learning for Digital Asset Limit Order Books
2 | 
3 | This paper shows that temporal CNNs accurately predict bitcoin spot price movements from limit order book data. On a 2 second prediction time horizon we achieve 76% walk-forward accuracy on the popular cryptocurrency exchange coinbase. Our model can be trained in less than a day on commodity GPUs which could be installed into colocation centers allowing for model sync with existing faster orderbook prediction models.
4 | 
5 | See paper at https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3704098 
6 | 


--------------------------------------------------------------------------------
/WalkForwardV2.ipynb:
--------------------------------------------------------------------------------
  1 | {
  2 |  "cells": [
  3 |   {
  4 |    "cell_type": "code",
  5 |    "execution_count": 1,
  6 |    "metadata": {},
  7 |    "outputs": [],
  8 |    "source": [
  9 |     "import gc\n",
 10 |     "import numpy as np\n",
 11 |     "import pandas as pd\n",
 12 |     "from tqdm import tqdm\n",
 13 |     "from google.cloud import storage\n",
 14 |     "from io import StringIO\n",
 15 |     "%matplotlib inline\n",
 16 |     "import seaborn as sns\n",
 17 |     "import matplotlib.pyplot as plt"
 18 |    ]
 19 |   },
 20 |   {
 21 |    "cell_type": "code",
 22 |    "execution_count": 2,
 23 |    "metadata": {},
 24 |    "outputs": [],
 25 |    "source": [
 26 |     "client = storage.Client()\n",
 27 |     "bucket = client.get_bucket('bucket_l2_snapshot')"
 28 |    ]
 29 |   },
 30 |   {
 31 |    "cell_type": "code",
 32 |    "execution_count": 3,
 33 |    "metadata": {},
 34 |    "outputs": [],
 35 |    "source": [
 36 |     "blob = bucket.blob('l2_snapshot_100ms.csv')\n",
 37 |     "bt = blob.download_as_string()\n",
 38 |     "s = str(bt, 'utf-8')\n",
 39 |     "s = StringIO(s)"
 40 |    ]
 41 |   },
 42 |   {
 43 |    "cell_type": "code",
 44 |    "execution_count": 4,
 45 |    "metadata": {},
 46 |    "outputs": [],
 47 |    "source": [
 48 |     "df = pd.read_csv(s, index_col=0, infer_datetime_format=True)"
 49 |    ]
 50 |   },
 51 |   {
 52 |    "cell_type": "code",
 53 |    "execution_count": 5,
 54 |    "metadata": {},
 55 |    "outputs": [
 56 |     {
 57 |      "data": {
 58 |       "text/html": [
 59 |        "<div>\n",
 60 |        "<style scoped>\n",
 61 |        "    .dataframe tbody tr th:only-of-type {\n",
 62 |        "        vertical-align: middle;\n",
 63 |        "    }\n",
 64 |        "\n",
 65 |        "    .dataframe tbody tr th {\n",
 66 |        "        vertical-align: top;\n",
 67 |        "    }\n",
 68 |        "\n",
 69 |        "    .dataframe thead th {\n",
 70 |        "        text-align: right;\n",
 71 |        "    }\n",
 72 |        "</style>\n",
 73 |        "<table border=\"1\" class=\"dataframe\">\n",
 74 |        "  <thead>\n",
 75 |        "    <tr style=\"text-align: right;\">\n",
 76 |        "      <th></th>\n",
 77 |        "      <th>b1</th>\n",
 78 |        "      <th>b2</th>\n",
 79 |        "      <th>b3</th>\n",
 80 |        "      <th>b4</th>\n",
 81 |        "      <th>b5</th>\n",
 82 |        "      <th>b6</th>\n",
 83 |        "      <th>b7</th>\n",
 84 |        "      <th>b8</th>\n",
 85 |        "      <th>b9</th>\n",
 86 |        "      <th>b10</th>\n",
 87 |        "      <th>...</th>\n",
 88 |        "      <th>aq1</th>\n",
 89 |        "      <th>aq2</th>\n",
 90 |        "      <th>aq3</th>\n",
 91 |        "      <th>aq4</th>\n",
 92 |        "      <th>aq5</th>\n",
 93 |        "      <th>aq6</th>\n",
 94 |        "      <th>aq7</th>\n",
 95 |        "      <th>aq8</th>\n",
 96 |        "      <th>aq9</th>\n",
 97 |        "      <th>aq10</th>\n",
 98 |        "    </tr>\n",
 99 |        "    <tr>\n",
100 |        "      <th>timestamp</th>\n",
101 |        "      <th></th>\n",
102 |        "      <th></th>\n",
103 |        "      <th></th>\n",
104 |        "      <th></th>\n",
105 |        "      <th></th>\n",
106 |        "      <th></th>\n",
107 |        "      <th></th>\n",
108 |        "      <th></th>\n",
109 |        "      <th></th>\n",
110 |        "      <th></th>\n",
111 |        "      <th></th>\n",
112 |        "      <th></th>\n",
113 |        "      <th></th>\n",
114 |        "      <th></th>\n",
115 |        "      <th></th>\n",
116 |        "      <th></th>\n",
117 |        "      <th></th>\n",
118 |        "      <th></th>\n",
119 |        "      <th></th>\n",
120 |        "      <th></th>\n",
121 |        "      <th></th>\n",
122 |        "    </tr>\n",
123 |        "  </thead>\n",
124 |        "  <tbody>\n",
125 |        "    <tr>\n",
126 |        "      <th>2019-11-12 00:00:00.000</th>\n",
127 |        "      <td>8721.53</td>\n",
128 |        "      <td>8720.59</td>\n",
129 |        "      <td>8719.55</td>\n",
130 |        "      <td>8719.50</td>\n",
131 |        "      <td>8719.0</td>\n",
132 |        "      <td>8718.02</td>\n",
133 |        "      <td>8717.87</td>\n",
134 |        "      <td>8717.85</td>\n",
135 |        "      <td>8717.56</td>\n",
136 |        "      <td>8716.06</td>\n",
137 |        "      <td>...</td>\n",
138 |        "      <td>5.802204</td>\n",
139 |        "      <td>2.859</td>\n",
140 |        "      <td>0.12189</td>\n",
141 |        "      <td>0.142575</td>\n",
142 |        "      <td>0.002692</td>\n",
143 |        "      <td>0.46</td>\n",
144 |        "      <td>0.038468</td>\n",
145 |        "      <td>0.200000</td>\n",
146 |        "      <td>0.001890</td>\n",
147 |        "      <td>1.00000</td>\n",
148 |        "    </tr>\n",
149 |        "    <tr>\n",
150 |        "      <th>2019-11-12 00:00:00.100</th>\n",
151 |        "      <td>8721.53</td>\n",
152 |        "      <td>8720.59</td>\n",
153 |        "      <td>8719.56</td>\n",
154 |        "      <td>8719.55</td>\n",
155 |        "      <td>8719.0</td>\n",
156 |        "      <td>8718.02</td>\n",
157 |        "      <td>8717.87</td>\n",
158 |        "      <td>8717.85</td>\n",
159 |        "      <td>8717.56</td>\n",
160 |        "      <td>8716.06</td>\n",
161 |        "      <td>...</td>\n",
162 |        "      <td>5.802204</td>\n",
163 |        "      <td>2.859</td>\n",
164 |        "      <td>0.12189</td>\n",
165 |        "      <td>0.142575</td>\n",
166 |        "      <td>0.002692</td>\n",
167 |        "      <td>0.46</td>\n",
168 |        "      <td>0.038468</td>\n",
169 |        "      <td>0.200000</td>\n",
170 |        "      <td>0.001890</td>\n",
171 |        "      <td>1.00000</td>\n",
172 |        "    </tr>\n",
173 |        "    <tr>\n",
174 |        "      <th>2019-11-12 00:00:00.200</th>\n",
175 |        "      <td>8721.53</td>\n",
176 |        "      <td>8720.59</td>\n",
177 |        "      <td>8719.61</td>\n",
178 |        "      <td>8719.56</td>\n",
179 |        "      <td>8719.0</td>\n",
180 |        "      <td>8718.02</td>\n",
181 |        "      <td>8717.87</td>\n",
182 |        "      <td>8717.85</td>\n",
183 |        "      <td>8716.06</td>\n",
184 |        "      <td>8716.00</td>\n",
185 |        "      <td>...</td>\n",
186 |        "      <td>5.802204</td>\n",
187 |        "      <td>2.859</td>\n",
188 |        "      <td>0.12189</td>\n",
189 |        "      <td>0.142575</td>\n",
190 |        "      <td>0.002692</td>\n",
191 |        "      <td>0.46</td>\n",
192 |        "      <td>0.038468</td>\n",
193 |        "      <td>0.200000</td>\n",
194 |        "      <td>1.086016</td>\n",
195 |        "      <td>0.00189</td>\n",
196 |        "    </tr>\n",
197 |        "    <tr>\n",
198 |        "      <th>2019-11-12 00:00:00.300</th>\n",
199 |        "      <td>8721.53</td>\n",
200 |        "      <td>8720.59</td>\n",
201 |        "      <td>8719.61</td>\n",
202 |        "      <td>8719.56</td>\n",
203 |        "      <td>8719.0</td>\n",
204 |        "      <td>8718.02</td>\n",
205 |        "      <td>8718.00</td>\n",
206 |        "      <td>8717.87</td>\n",
207 |        "      <td>8717.85</td>\n",
208 |        "      <td>8716.60</td>\n",
209 |        "      <td>...</td>\n",
210 |        "      <td>7.802204</td>\n",
211 |        "      <td>2.859</td>\n",
212 |        "      <td>0.12189</td>\n",
213 |        "      <td>0.142575</td>\n",
214 |        "      <td>0.002692</td>\n",
215 |        "      <td>0.46</td>\n",
216 |        "      <td>0.038468</td>\n",
217 |        "      <td>0.200000</td>\n",
218 |        "      <td>1.086016</td>\n",
219 |        "      <td>0.00189</td>\n",
220 |        "    </tr>\n",
221 |        "    <tr>\n",
222 |        "      <th>2019-11-12 00:00:00.400</th>\n",
223 |        "      <td>8721.53</td>\n",
224 |        "      <td>8720.59</td>\n",
225 |        "      <td>8719.61</td>\n",
226 |        "      <td>8719.56</td>\n",
227 |        "      <td>8719.0</td>\n",
228 |        "      <td>8718.99</td>\n",
229 |        "      <td>8718.02</td>\n",
230 |        "      <td>8718.00</td>\n",
231 |        "      <td>8717.87</td>\n",
232 |        "      <td>8717.85</td>\n",
233 |        "      <td>...</td>\n",
234 |        "      <td>7.802204</td>\n",
235 |        "      <td>2.859</td>\n",
236 |        "      <td>0.12189</td>\n",
237 |        "      <td>0.142575</td>\n",
238 |        "      <td>0.002692</td>\n",
239 |        "      <td>0.46</td>\n",
240 |        "      <td>0.200000</td>\n",
241 |        "      <td>0.038468</td>\n",
242 |        "      <td>1.086016</td>\n",
243 |        "      <td>0.00189</td>\n",
244 |        "    </tr>\n",
245 |        "  </tbody>\n",
246 |        "</table>\n",
247 |        "<p>5 rows × 40 columns</p>\n",
248 |        "</div>"
249 |       ],
250 |       "text/plain": [
251 |        "                              b1       b2       b3       b4      b5       b6  \\\n",
252 |        "timestamp                                                                      \n",
253 |        "2019-11-12 00:00:00.000  8721.53  8720.59  8719.55  8719.50  8719.0  8718.02   \n",
254 |        "2019-11-12 00:00:00.100  8721.53  8720.59  8719.56  8719.55  8719.0  8718.02   \n",
255 |        "2019-11-12 00:00:00.200  8721.53  8720.59  8719.61  8719.56  8719.0  8718.02   \n",
256 |        "2019-11-12 00:00:00.300  8721.53  8720.59  8719.61  8719.56  8719.0  8718.02   \n",
257 |        "2019-11-12 00:00:00.400  8721.53  8720.59  8719.61  8719.56  8719.0  8718.99   \n",
258 |        "\n",
259 |        "                              b7       b8       b9      b10  ...       aq1  \\\n",
260 |        "timestamp                                                    ...             \n",
261 |        "2019-11-12 00:00:00.000  8717.87  8717.85  8717.56  8716.06  ...  5.802204   \n",
262 |        "2019-11-12 00:00:00.100  8717.87  8717.85  8717.56  8716.06  ...  5.802204   \n",
263 |        "2019-11-12 00:00:00.200  8717.87  8717.85  8716.06  8716.00  ...  5.802204   \n",
264 |        "2019-11-12 00:00:00.300  8718.00  8717.87  8717.85  8716.60  ...  7.802204   \n",
265 |        "2019-11-12 00:00:00.400  8718.02  8718.00  8717.87  8717.85  ...  7.802204   \n",
266 |        "\n",
267 |        "                           aq2      aq3       aq4       aq5   aq6       aq7  \\\n",
268 |        "timestamp                                                                     \n",
269 |        "2019-11-12 00:00:00.000  2.859  0.12189  0.142575  0.002692  0.46  0.038468   \n",
270 |        "2019-11-12 00:00:00.100  2.859  0.12189  0.142575  0.002692  0.46  0.038468   \n",
271 |        "2019-11-12 00:00:00.200  2.859  0.12189  0.142575  0.002692  0.46  0.038468   \n",
272 |        "2019-11-12 00:00:00.300  2.859  0.12189  0.142575  0.002692  0.46  0.038468   \n",
273 |        "2019-11-12 00:00:00.400  2.859  0.12189  0.142575  0.002692  0.46  0.200000   \n",
274 |        "\n",
275 |        "                              aq8       aq9     aq10  \n",
276 |        "timestamp                                             \n",
277 |        "2019-11-12 00:00:00.000  0.200000  0.001890  1.00000  \n",
278 |        "2019-11-12 00:00:00.100  0.200000  0.001890  1.00000  \n",
279 |        "2019-11-12 00:00:00.200  0.200000  1.086016  0.00189  \n",
280 |        "2019-11-12 00:00:00.300  0.200000  1.086016  0.00189  \n",
281 |        "2019-11-12 00:00:00.400  0.038468  1.086016  0.00189  \n",
282 |        "\n",
283 |        "[5 rows x 40 columns]"
284 |       ]
285 |      },
286 |      "execution_count": 5,
287 |      "metadata": {},
288 |      "output_type": "execute_result"
289 |     }
290 |    ],
291 |    "source": [
292 |     "df.head()"
293 |    ]
294 |   },
295 |   {
296 |    "cell_type": "code",
297 |    "execution_count": 6,
298 |    "metadata": {},
299 |    "outputs": [],
300 |    "source": [
301 |     "df.index = pd.to_datetime(df.index)"
302 |    ]
303 |   },
304 |   {
305 |    "cell_type": "code",
306 |    "execution_count": 7,
307 |    "metadata": {},
308 |    "outputs": [],
309 |    "source": [
310 |     "def concat_data(base_path, features):\n",
311 |     "    \"\"\"Concatenate all the files in basepath keeping only the\n",
312 |     "    columns specified by features.\n",
313 |     "    \"\"\"\n",
314 |     "    l2_snapshot = pd.DataFrame()\n",
315 |     "    for i, x in enumerate(tqdm(sorted(os.listdir(base_path)))):\n",
316 |     "        if base_path is None:\n",
317 |     "            path = x\n",
318 |     "        else:\n",
319 |     "            path = base_path + x\n",
320 |     "        df_hour = pd.read_parquet(path, columns=features)\n",
321 |     "        l2_snapshot = pd.concat([l2_snapshot, df_hour.dropna()])\n",
322 |     "        \n",
323 |     "    return l2_snapshot\n",
324 |     "\n",
325 |     "def extend_matrix(A, n):\n",
326 |     "    \"\"\"Extend a matrix A by duplicating rows as specified by the list n.\"\"\"\n",
327 |     "    n = n[1:]  # Do not duplicate rows for the first day\n",
328 |     "    A = A[:-1]  # Do not duplicate the last day's row\n",
329 |     "    A = np.repeat(A, repeats=n, axis=0)\n",
330 |     "    return A\n",
331 |     "\n",
332 |     "def normalise_data_per_day(df):\n",
333 |     "    df_mean = df.resample('D').mean()\n",
334 |     "    df_var = df.resample('D').var()\n",
335 |     "    \n",
336 |     "    timestamps_per_day = np.unique(df.index.date, return_counts=True)[1]\n",
337 |     "    mean_array = extend_matrix(df_mean.to_numpy(), timestamps_per_day)\n",
338 |     "    var_array = extend_matrix(df_var.to_numpy(), timestamps_per_day)\n",
339 |     "    \n",
340 |     "    # Drop the rows of the first day\n",
341 |     "    df = df[df.index.date != df.index[0].date()]\n",
342 |     "    \n",
343 |     "    df = (df - mean_array) / np.sqrt(var_array)\n",
344 |     "\n",
345 |     "    return df\n",
346 |     "\n",
347 |     "def balance_classes(y):\n",
348 |     "    unique = np.unique(y, return_counts=True)\n",
349 |     "\n",
350 |     "    # Take smallest number as class size\n",
351 |     "    class_size = np.min(unique[1])\n",
352 |     "    class_size_index = np.argmin(unique[1])\n",
353 |     "    timestamps = np.array([], dtype=int)\n",
354 |     "    for i, category in enumerate(unique[0]):\n",
355 |     "        if i == class_size_index:\n",
356 |     "            continue\n",
357 |     "        index = np.argwhere(y==category)\n",
358 |     "        index = index.reshape(len(index))\n",
359 |     "        random_timestamps = np.random.choice(index, (unique[1][i] - class_size), replace=False)\n",
360 |     "        timestamps = np.concatenate((timestamps, random_timestamps), axis=None)\n",
361 |     "        \n",
362 |     "    return timestamps\n",
363 |     "\n",
364 |     "def generate_y(df_snapshot, T=100, D=40, best_ask='a1', best_bid='b1', k=20, alpha=10e-5):\n",
365 |     "    \"\"\"Return X, y from the snapshot dataframe and the best ask/bid columns.\"\"\"\n",
366 |     "    df = pd.DataFrame()\n",
367 |     "    df['mid_price'] = (df_snapshot[best_ask].to_numpy()+df_snapshot[best_bid].to_numpy())/2\n",
368 |     "\n",
369 |     "    # Create columns delayed by -k to k-1\n",
370 |     "    for i in range(-k, k):\n",
371 |     "        df[i] = df['mid_price'].shift(periods=i)\n",
372 |     "\n",
373 |     "    # Drop first k-1 rows and last k rows\n",
374 |     "    df.drop(range(0,20), axis=0, inplace=True)\n",
375 |     "    df.drop(range(len(df_snapshot)-20,len(df_snapshot)), axis=0, inplace=True)\n",
376 |     "    \n",
377 |     "    # Compute mean of previous k and next k\n",
378 |     "    df['m_b'] = df[range(0,20)].mean(axis=1)\n",
379 |     "    df['m_a'] = df[range(-20,0)].mean(axis=1)\n",
380 |     "    \n",
381 |     "    # Compute label of increasing or decreasing\n",
382 |     "    y_increase = np.where(df['m_b'] > df['m_a'] * (1+alpha), 1, 0)\n",
383 |     "    y_decrease = np.where(df['m_b'] < df['m_a'] * (1-alpha), -1, 0)\n",
384 |     "    y = y_increase + y_decrease\n",
385 |     "\n",
386 |     "    # 100 most recent limit orders used so ignore first 100 timesteps\n",
387 |     "    y = y[T:]\n",
388 |     "    y += 1\n",
389 |     "\n",
390 |     "    return y\n",
391 |     "\n",
392 |     "def generate_preX(df_snapshot):\n",
393 |     "    # First and last 20 can't create labels for as previous and next k=20 needed\n",
394 |     "    preX = df_snapshot.to_numpy()[20:-20]\n",
395 |     "\n",
396 |     "    return preX\n",
397 |     "\n",
398 |     "def generate_X(preX, T=100, D=40):\n",
399 |     "    # For each timestep create matrix of 100 most recent limit orders\n",
400 |     "    X = np.array([preX[t:t+T] for t in range(len(preX)-T)], dtype='float32')\n",
401 |     "        \n",
402 |     "    return X"
403 |    ]
404 |   },
405 |   {
406 |    "cell_type": "code",
407 |    "execution_count": 8,
408 |    "metadata": {},
409 |    "outputs": [],
410 |    "source": [
411 |     "l2_norm = normalise_data_per_day(df)"
412 |    ]
413 |   },
414 |   {
415 |    "cell_type": "code",
416 |    "execution_count": 9,
417 |    "metadata": {},
418 |    "outputs": [
419 |     {
420 |      "data": {
421 |       "text/html": [
422 |        "<div>\n",
423 |        "<style scoped>\n",
424 |        "    .dataframe tbody tr th:only-of-type {\n",
425 |        "        vertical-align: middle;\n",
426 |        "    }\n",
427 |        "\n",
428 |        "    .dataframe tbody tr th {\n",
429 |        "        vertical-align: top;\n",
430 |        "    }\n",
431 |        "\n",
432 |        "    .dataframe thead th {\n",
433 |        "        text-align: right;\n",
434 |        "    }\n",
435 |        "</style>\n",
436 |        "<table border=\"1\" class=\"dataframe\">\n",
437 |        "  <thead>\n",
438 |        "    <tr style=\"text-align: right;\">\n",
439 |        "      <th></th>\n",
440 |        "      <th>b1</th>\n",
441 |        "      <th>b2</th>\n",
442 |        "      <th>b3</th>\n",
443 |        "      <th>b4</th>\n",
444 |        "      <th>b5</th>\n",
445 |        "      <th>b6</th>\n",
446 |        "      <th>b7</th>\n",
447 |        "      <th>b8</th>\n",
448 |        "      <th>b9</th>\n",
449 |        "      <th>b10</th>\n",
450 |        "      <th>...</th>\n",
451 |        "      <th>aq1</th>\n",
452 |        "      <th>aq2</th>\n",
453 |        "      <th>aq3</th>\n",
454 |        "      <th>aq4</th>\n",
455 |        "      <th>aq5</th>\n",
456 |        "      <th>aq6</th>\n",
457 |        "      <th>aq7</th>\n",
458 |        "      <th>aq8</th>\n",
459 |        "      <th>aq9</th>\n",
460 |        "      <th>aq10</th>\n",
461 |        "    </tr>\n",
462 |        "    <tr>\n",
463 |        "      <th>timestamp</th>\n",
464 |        "      <th></th>\n",
465 |        "      <th></th>\n",
466 |        "      <th></th>\n",
467 |        "      <th></th>\n",
468 |        "      <th></th>\n",
469 |        "      <th></th>\n",
470 |        "      <th></th>\n",
471 |        "      <th></th>\n",
472 |        "      <th></th>\n",
473 |        "      <th></th>\n",
474 |        "      <th></th>\n",
475 |        "      <th></th>\n",
476 |        "      <th></th>\n",
477 |        "      <th></th>\n",
478 |        "      <th></th>\n",
479 |        "      <th></th>\n",
480 |        "      <th></th>\n",
481 |        "      <th></th>\n",
482 |        "      <th></th>\n",
483 |        "      <th></th>\n",
484 |        "      <th></th>\n",
485 |        "    </tr>\n",
486 |        "  </thead>\n",
487 |        "  <tbody>\n",
488 |        "    <tr>\n",
489 |        "      <th>2019-11-13 00:00:00.000</th>\n",
490 |        "      <td>2.069669</td>\n",
491 |        "      <td>2.066221</td>\n",
492 |        "      <td>2.049359</td>\n",
493 |        "      <td>2.032164</td>\n",
494 |        "      <td>2.045755</td>\n",
495 |        "      <td>2.056132</td>\n",
496 |        "      <td>2.062144</td>\n",
497 |        "      <td>2.072841</td>\n",
498 |        "      <td>2.064472</td>\n",
499 |        "      <td>2.076723</td>\n",
500 |        "      <td>...</td>\n",
501 |        "      <td>-0.630327</td>\n",
502 |        "      <td>-0.585376</td>\n",
503 |        "      <td>-0.603213</td>\n",
504 |        "      <td>-0.584939</td>\n",
505 |        "      <td>-0.568453</td>\n",
506 |        "      <td>-0.547841</td>\n",
507 |        "      <td>-0.503235</td>\n",
508 |        "      <td>-0.459792</td>\n",
509 |        "      <td>-0.545609</td>\n",
510 |        "      <td>-0.581027</td>\n",
511 |        "    </tr>\n",
512 |        "    <tr>\n",
513 |        "      <th>2019-11-13 00:00:00.100</th>\n",
514 |        "      <td>2.096553</td>\n",
515 |        "      <td>2.099017</td>\n",
516 |        "      <td>2.099095</td>\n",
517 |        "      <td>2.093216</td>\n",
518 |        "      <td>2.076695</td>\n",
519 |        "      <td>2.060169</td>\n",
520 |        "      <td>2.073992</td>\n",
521 |        "      <td>2.084426</td>\n",
522 |        "      <td>2.090621</td>\n",
523 |        "      <td>2.101545</td>\n",
524 |        "      <td>...</td>\n",
525 |        "      <td>-0.630484</td>\n",
526 |        "      <td>-0.585376</td>\n",
527 |        "      <td>-0.603213</td>\n",
528 |        "      <td>-0.584939</td>\n",
529 |        "      <td>-0.568453</td>\n",
530 |        "      <td>-0.547841</td>\n",
531 |        "      <td>-0.559809</td>\n",
532 |        "      <td>-0.556070</td>\n",
533 |        "      <td>-0.552547</td>\n",
534 |        "      <td>-0.576830</td>\n",
535 |        "    </tr>\n",
536 |        "    <tr>\n",
537 |        "      <th>2019-11-13 00:00:00.200</th>\n",
538 |        "      <td>2.096553</td>\n",
539 |        "      <td>2.099017</td>\n",
540 |        "      <td>2.099095</td>\n",
541 |        "      <td>2.093216</td>\n",
542 |        "      <td>2.076695</td>\n",
543 |        "      <td>2.060169</td>\n",
544 |        "      <td>2.073992</td>\n",
545 |        "      <td>2.084426</td>\n",
546 |        "      <td>2.090621</td>\n",
547 |        "      <td>2.101545</td>\n",
548 |        "      <td>...</td>\n",
549 |        "      <td>-0.630484</td>\n",
550 |        "      <td>-0.585376</td>\n",
551 |        "      <td>-0.603213</td>\n",
552 |        "      <td>-0.584939</td>\n",
553 |        "      <td>-0.568453</td>\n",
554 |        "      <td>-0.547841</td>\n",
555 |        "      <td>-0.559809</td>\n",
556 |        "      <td>-0.556070</td>\n",
557 |        "      <td>-0.552547</td>\n",
558 |        "      <td>-0.576830</td>\n",
559 |        "    </tr>\n",
560 |        "    <tr>\n",
561 |        "      <th>2019-11-13 00:00:00.300</th>\n",
562 |        "      <td>2.096553</td>\n",
563 |        "      <td>2.085576</td>\n",
564 |        "      <td>2.096676</td>\n",
565 |        "      <td>2.062824</td>\n",
566 |        "      <td>2.046024</td>\n",
567 |        "      <td>2.059900</td>\n",
568 |        "      <td>2.070222</td>\n",
569 |        "      <td>2.076344</td>\n",
570 |        "      <td>2.087117</td>\n",
571 |        "      <td>2.078882</td>\n",
572 |        "      <td>...</td>\n",
573 |        "      <td>-0.630484</td>\n",
574 |        "      <td>-0.585376</td>\n",
575 |        "      <td>-0.603213</td>\n",
576 |        "      <td>-0.584939</td>\n",
577 |        "      <td>-0.568453</td>\n",
578 |        "      <td>-0.547841</td>\n",
579 |        "      <td>-0.559809</td>\n",
580 |        "      <td>-0.556070</td>\n",
581 |        "      <td>-0.552547</td>\n",
582 |        "      <td>-0.576830</td>\n",
583 |        "    </tr>\n",
584 |        "    <tr>\n",
585 |        "      <th>2019-11-13 00:00:00.400</th>\n",
586 |        "      <td>2.096553</td>\n",
587 |        "      <td>2.085576</td>\n",
588 |        "      <td>2.096676</td>\n",
589 |        "      <td>2.062824</td>\n",
590 |        "      <td>2.046024</td>\n",
591 |        "      <td>2.059900</td>\n",
592 |        "      <td>2.070222</td>\n",
593 |        "      <td>2.076344</td>\n",
594 |        "      <td>2.064472</td>\n",
595 |        "      <td>2.076723</td>\n",
596 |        "      <td>...</td>\n",
597 |        "      <td>-0.630484</td>\n",
598 |        "      <td>-0.585376</td>\n",
599 |        "      <td>-0.603213</td>\n",
600 |        "      <td>-0.584939</td>\n",
601 |        "      <td>-0.568453</td>\n",
602 |        "      <td>-0.547841</td>\n",
603 |        "      <td>-0.559809</td>\n",
604 |        "      <td>-0.556070</td>\n",
605 |        "      <td>-0.552547</td>\n",
606 |        "      <td>-0.576830</td>\n",
607 |        "    </tr>\n",
608 |        "  </tbody>\n",
609 |        "</table>\n",
610 |        "<p>5 rows × 40 columns</p>\n",
611 |        "</div>"
612 |       ],
613 |       "text/plain": [
614 |        "                               b1        b2        b3        b4        b5  \\\n",
615 |        "timestamp                                                                   \n",
616 |        "2019-11-13 00:00:00.000  2.069669  2.066221  2.049359  2.032164  2.045755   \n",
617 |        "2019-11-13 00:00:00.100  2.096553  2.099017  2.099095  2.093216  2.076695   \n",
618 |        "2019-11-13 00:00:00.200  2.096553  2.099017  2.099095  2.093216  2.076695   \n",
619 |        "2019-11-13 00:00:00.300  2.096553  2.085576  2.096676  2.062824  2.046024   \n",
620 |        "2019-11-13 00:00:00.400  2.096553  2.085576  2.096676  2.062824  2.046024   \n",
621 |        "\n",
622 |        "                               b6        b7        b8        b9       b10  \\\n",
623 |        "timestamp                                                                   \n",
624 |        "2019-11-13 00:00:00.000  2.056132  2.062144  2.072841  2.064472  2.076723   \n",
625 |        "2019-11-13 00:00:00.100  2.060169  2.073992  2.084426  2.090621  2.101545   \n",
626 |        "2019-11-13 00:00:00.200  2.060169  2.073992  2.084426  2.090621  2.101545   \n",
627 |        "2019-11-13 00:00:00.300  2.059900  2.070222  2.076344  2.087117  2.078882   \n",
628 |        "2019-11-13 00:00:00.400  2.059900  2.070222  2.076344  2.064472  2.076723   \n",
629 |        "\n",
630 |        "                         ...       aq1       aq2       aq3       aq4  \\\n",
631 |        "timestamp                ...                                           \n",
632 |        "2019-11-13 00:00:00.000  ... -0.630327 -0.585376 -0.603213 -0.584939   \n",
633 |        "2019-11-13 00:00:00.100  ... -0.630484 -0.585376 -0.603213 -0.584939   \n",
634 |        "2019-11-13 00:00:00.200  ... -0.630484 -0.585376 -0.603213 -0.584939   \n",
635 |        "2019-11-13 00:00:00.300  ... -0.630484 -0.585376 -0.603213 -0.584939   \n",
636 |        "2019-11-13 00:00:00.400  ... -0.630484 -0.585376 -0.603213 -0.584939   \n",
637 |        "\n",
638 |        "                              aq5       aq6       aq7       aq8       aq9  \\\n",
639 |        "timestamp                                                                   \n",
640 |        "2019-11-13 00:00:00.000 -0.568453 -0.547841 -0.503235 -0.459792 -0.545609   \n",
641 |        "2019-11-13 00:00:00.100 -0.568453 -0.547841 -0.559809 -0.556070 -0.552547   \n",
642 |        "2019-11-13 00:00:00.200 -0.568453 -0.547841 -0.559809 -0.556070 -0.552547   \n",
643 |        "2019-11-13 00:00:00.300 -0.568453 -0.547841 -0.559809 -0.556070 -0.552547   \n",
644 |        "2019-11-13 00:00:00.400 -0.568453 -0.547841 -0.559809 -0.556070 -0.552547   \n",
645 |        "\n",
646 |        "                             aq10  \n",
647 |        "timestamp                          \n",
648 |        "2019-11-13 00:00:00.000 -0.581027  \n",
649 |        "2019-11-13 00:00:00.100 -0.576830  \n",
650 |        "2019-11-13 00:00:00.200 -0.576830  \n",
651 |        "2019-11-13 00:00:00.300 -0.576830  \n",
652 |        "2019-11-13 00:00:00.400 -0.576830  \n",
653 |        "\n",
654 |        "[5 rows x 40 columns]"
655 |       ]
656 |      },
657 |      "execution_count": 9,
658 |      "metadata": {},
659 |      "output_type": "execute_result"
660 |     }
661 |    ],
662 |    "source": [
663 |     "l2_norm.head()"
664 |    ]
665 |   },
666 |   {
667 |    "cell_type": "code",
668 |    "execution_count": 10,
669 |    "metadata": {},
670 |    "outputs": [
671 |     {
672 |      "name": "stdout",
673 |      "output_type": "stream",
674 |      "text": [
675 |       "preX Shape:  (4220032, 40)\n",
676 |       "X shape, y shape:  (4219932, 100, 40) (4219932,)\n",
677 |       "Unique y's:  [0 1 2]\n"
678 |      ]
679 |     }
680 |    ],
681 |    "source": [
682 |     "T = 100\n",
683 |     "D = 40\n",
684 |     "y = generate_y(l2_norm, T=T, D=D, best_ask='a1', best_bid = 'b1', alpha=0.002)\n",
685 |     "preX = generate_preX(l2_norm)\n",
686 |     "print(\"preX Shape: \", preX.shape)\n",
687 |     "X = generate_X(preX)\n",
688 |     "print(\"X shape, y shape: \", X.shape, y.shape)\n",
689 |     "print(\"Unique y's: \", np.unique(y))\n",
690 |     "\n",
691 |     "del preX\n",
692 |     "gc.collect()\n",
693 |     "\n",
694 |     "# First and last 20 are removed to create labels and then last 100 as previous 100 is required for input matrix\n",
695 |     "X_index = l2_norm[20:-120].index\n",
696 |     "\n",
697 |     "# Number of data points for the last day\n",
698 |     "num_test = np.unique(X_index.day, return_counts=True)[1][-1]\n",
699 |     "\n",
700 |     "# Split the data into the first seven days and the last day\n",
701 |     "X_train_val = X[:-num_test]\n",
702 |     "y_train_val = y[:-num_test]\n",
703 |     "X_test = X[-num_test:]\n",
704 |     "y_test = y[-num_test:]"
705 |    ]
706 |   },
707 |   {
708 |    "cell_type": "code",
709 |    "execution_count": 11,
710 |    "metadata": {},
711 |    "outputs": [],
712 |    "source": [
713 |     "import tensorflow as tf\n",
714 |     "from tensorflow.keras.layers import Dense\n",
715 |     "from tensorflow.keras import Input, Model\n",
716 |     "from tcn import TCN, tcn_full_summary"
717 |    ]
718 |   },
719 |   {
720 |    "cell_type": "code",
721 |    "execution_count": 12,
722 |    "metadata": {},
723 |    "outputs": [],
724 |    "source": [
725 |     "from tensorflow import keras"
726 |    ]
727 |   },
728 |   {
729 |    "cell_type": "code",
730 |    "execution_count": 13,
731 |    "metadata": {},
732 |    "outputs": [
733 |     {
734 |      "name": "stdout",
735 |      "output_type": "stream",
736 |      "text": [
737 |       "INFO:tensorflow:Using MirroredStrategy with devices ('/job:localhost/replica:0/task:0/device:GPU:0', '/job:localhost/replica:0/task:0/device:GPU:1', '/job:localhost/replica:0/task:0/device:GPU:2', '/job:localhost/replica:0/task:0/device:GPU:3')\n",
738 |       "INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).\n",
739 |       "INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).\n",
740 |       "INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).\n",
741 |       "INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).\n",
742 |       "INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).\n",
743 |       "INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).\n",
744 |       "INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).\n",
745 |       "INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).\n",
746 |       "INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).\n",
747 |       "INFO:tensorflow:Reduce to /job:localhost/replica:0/task:0/device:CPU:0 then broadcast to ('/job:localhost/replica:0/task:0/device:CPU:0',).\n"
748 |      ]
749 |     }
750 |    ],
751 |    "source": [
752 |     "strategy = tf.distribute.MirroredStrategy()\n",
753 |     "with strategy.scope():\n",
754 |     "    i = Input(batch_shape=(None, 100, 40))\n",
755 |     "    o = TCN(return_sequences=False, use_skip_connections=True, dropout_rate=0.4, dilations=[1, 2, 4, 8, 16, 32, 64], use_batch_norm=True)(i)\n",
756 |     "    o = Dense(3, activation='softmax')(o)\n",
757 |     "    m = Model(inputs=[i], outputs=[o])\n",
758 |     "    opt = keras.optimizers.Adam(learning_rate=0.01)\n",
759 |     "    m.compile(optimizer=opt, loss='sparse_categorical_crossentropy', metrics=['accuracy'])"
760 |    ]
761 |   },
762 |   {
763 |    "cell_type": "code",
764 |    "execution_count": 14,
765 |    "metadata": {},
766 |    "outputs": [],
767 |    "source": [
768 |     "from sklearn.utils.class_weight import compute_class_weight\n",
769 |     "class_weights = compute_class_weight('balanced', np.unique(y_train_val), y_train_val)\n",
770 |     "d_class_weights = dict(enumerate(class_weights))"
771 |    ]
772 |   },
773 |   {
774 |    "cell_type": "code",
775 |    "execution_count": 15,
776 |    "metadata": {},
777 |    "outputs": [
778 |     {
779 |      "name": "stdout",
780 |      "output_type": "stream",
781 |      "text": [
782 |       "WARNING:tensorflow:sample_weight modes were coerced from\n",
783 |       "  ...\n",
784 |       "    to  \n",
785 |       "  ['...']\n",
786 |       "WARNING:tensorflow:sample_weight modes were coerced from\n",
787 |       "  ...\n",
788 |       "    to  \n",
789 |       "  ['...']\n",
790 |       "Train on 3754532 samples, validate on 465400 samples\n",
791 |       "Epoch 1/1000\n",
792 |       "INFO:tensorflow:batch_all_reduce: 60 all-reduces with algorithm = nccl, num_packs = 1, agg_small_grads_max_bytes = 0 and agg_small_grads_max_group = 10\n",
793 |       "INFO:tensorflow:batch_all_reduce: 60 all-reduces with algorithm = nccl, num_packs = 1, agg_small_grads_max_bytes = 0 and agg_small_grads_max_group = 10\n",
794 |       "3754496/3754532 [============================>.] - ETA: 0s - loss: 0.7631 - accuracy: 0.6850\n",
795 |       "Epoch 00001: val_loss improved from inf to 0.66100, saving model to model_resnet_walkforward_tcn_actual.h5\n",
796 |       "3754532/3754532 [==============================] - 934s 249us/sample - loss: 0.7631 - accuracy: 0.6850 - val_loss: 0.6610 - val_accuracy: 0.6530\n",
797 |       "Epoch 2/1000\n",
798 |       "3754368/3754532 [============================>.] - ETA: 0s - loss: 0.7186 - accuracy: 0.7191\n",
799 |       "Epoch 00002: val_loss improved from 0.66100 to 0.65249, saving model to model_resnet_walkforward_tcn_actual.h5\n",
800 |       "3754532/3754532 [==============================] - 831s 221us/sample - loss: 0.7186 - accuracy: 0.7191 - val_loss: 0.6525 - val_accuracy: 0.6863\n",
801 |       "Epoch 3/1000\n",
802 |       "3754368/3754532 [============================>.] - ETA: 0s - loss: 0.7061 - accuracy: 0.7253\n",
803 |       "Epoch 00003: val_loss improved from 0.65249 to 0.64854, saving model to model_resnet_walkforward_tcn_actual.h5\n",
804 |       "3754532/3754532 [==============================] - 831s 221us/sample - loss: 0.7061 - accuracy: 0.7253 - val_loss: 0.6485 - val_accuracy: 0.7433\n",
805 |       "Epoch 4/1000\n",
806 |       "3754496/3754532 [============================>.] - ETA: 0s - loss: 0.6982 - accuracy: 0.7281\n",
807 |       "Epoch 00004: val_loss did not improve from 0.64854\n",
808 |       "3754532/3754532 [==============================] - 829s 221us/sample - loss: 0.6982 - accuracy: 0.7281 - val_loss: 0.6519 - val_accuracy: 0.6770\n",
809 |       "Epoch 5/1000\n",
810 |       "3754368/3754532 [============================>.] - ETA: 0s - loss: 0.6922 - accuracy: 0.7299\n",
811 |       "Epoch 00005: val_loss improved from 0.64854 to 0.64811, saving model to model_resnet_walkforward_tcn_actual.h5\n",
812 |       "3754532/3754532 [==============================] - 829s 221us/sample - loss: 0.6922 - accuracy: 0.7299 - val_loss: 0.6481 - val_accuracy: 0.6421\n",
813 |       "Epoch 6/1000\n",
814 |       "3754368/3754532 [============================>.] - ETA: 0s - loss: 0.6876 - accuracy: 0.7312\n",
815 |       "Epoch 00006: val_loss improved from 0.64811 to 0.64290, saving model to model_resnet_walkforward_tcn_actual.h5\n",
816 |       "3754532/3754532 [==============================] - 830s 221us/sample - loss: 0.6876 - accuracy: 0.7312 - val_loss: 0.6429 - val_accuracy: 0.6118\n",
817 |       "Epoch 7/1000\n",
818 |       "3754496/3754532 [============================>.] - ETA: 0s - loss: 0.6840 - accuracy: 0.7317\n",
819 |       "Epoch 00007: val_loss did not improve from 0.64290\n",
820 |       "3754532/3754532 [==============================] - 831s 221us/sample - loss: 0.6840 - accuracy: 0.7317 - val_loss: 0.6484 - val_accuracy: 0.7022\n",
821 |       "Epoch 8/1000\n",
822 |       "3754496/3754532 [============================>.] - ETA: 0s - loss: 0.6812 - accuracy: 0.7331\n",
823 |       "Epoch 00008: val_loss improved from 0.64290 to 0.63999, saving model to model_resnet_walkforward_tcn_actual.h5\n",
824 |       "3754532/3754532 [==============================] - 831s 221us/sample - loss: 0.6812 - accuracy: 0.7331 - val_loss: 0.6400 - val_accuracy: 0.6446\n",
825 |       "Epoch 9/1000\n",
826 |       "3754496/3754532 [============================>.] - ETA: 0s - loss: 0.6784 - accuracy: 0.7341\n",
827 |       "Epoch 00009: val_loss improved from 0.63999 to 0.63913, saving model to model_resnet_walkforward_tcn_actual.h5\n",
828 |       "3754532/3754532 [==============================] - 832s 222us/sample - loss: 0.6784 - accuracy: 0.7341 - val_loss: 0.6391 - val_accuracy: 0.6525\n",
829 |       "Epoch 10/1000\n",
830 |       "3754496/3754532 [============================>.] - ETA: 0s - loss: 0.6759 - accuracy: 0.7348\n",
831 |       "Epoch 00010: val_loss improved from 0.63913 to 0.63433, saving model to model_resnet_walkforward_tcn_actual.h5\n",
832 |       "3754532/3754532 [==============================] - 830s 221us/sample - loss: 0.6759 - accuracy: 0.7348 - val_loss: 0.6343 - val_accuracy: 0.7053\n",
833 |       "Epoch 11/1000\n",
834 |       "3754368/3754532 [============================>.] - ETA: 0s - loss: 0.6736 - accuracy: 0.7358\n",
835 |       "Epoch 00011: val_loss did not improve from 0.63433\n",
836 |       "3754532/3754532 [==============================] - 830s 221us/sample - loss: 0.6737 - accuracy: 0.7358 - val_loss: 0.6431 - val_accuracy: 0.6656\n",
837 |       "Epoch 12/1000\n",
838 |       "3754496/3754532 [============================>.] - ETA: 0s - loss: 0.6717 - accuracy: 0.7367\n",
839 |       "Epoch 00012: val_loss improved from 0.63433 to 0.62916, saving model to model_resnet_walkforward_tcn_actual.h5\n",
840 |       "3754532/3754532 [==============================] - 831s 221us/sample - loss: 0.6717 - accuracy: 0.7367 - val_loss: 0.6292 - val_accuracy: 0.7084\n",
841 |       "Epoch 13/1000\n",
842 |       "3754496/3754532 [============================>.] - ETA: 0s - loss: 0.6704 - accuracy: 0.7368\n",
843 |       "Epoch 00013: val_loss did not improve from 0.62916\n",
844 |       "3754532/3754532 [==============================] - 831s 221us/sample - loss: 0.6704 - accuracy: 0.7368 - val_loss: 0.6409 - val_accuracy: 0.7432\n",
845 |       "Epoch 14/1000\n",
846 |       "3754496/3754532 [============================>.] - ETA: 0s - loss: 0.6688 - accuracy: 0.7373\n",
847 |       "Epoch 00014: ReduceLROnPlateau reducing learning rate to 0.004999999888241291.\n",
848 |       "\n",
849 |       "Epoch 00014: val_loss did not improve from 0.62916\n",
850 |       "3754532/3754532 [==============================] - 832s 222us/sample - loss: 0.6688 - accuracy: 0.7373 - val_loss: 0.6355 - val_accuracy: 0.6914\n",
851 |       "Epoch 15/1000\n",
852 |       "3754240/3754532 [============================>.] - ETA: 0s - loss: 0.6536 - accuracy: 0.7419\n",
853 |       "Epoch 00015: val_loss did not improve from 0.62916\n",
854 |       "3754532/3754532 [==============================] - 832s 221us/sample - loss: 0.6536 - accuracy: 0.7419 - val_loss: 0.6311 - val_accuracy: 0.6917\n",
855 |       "Epoch 16/1000\n",
856 |       "3754368/3754532 [============================>.] - ETA: 0s - loss: 0.6521 - accuracy: 0.7426\n",
857 |       "Epoch 00016: ReduceLROnPlateau reducing learning rate to 0.0024999999441206455.\n",
858 |       "\n",
859 |       "Epoch 00016: val_loss did not improve from 0.62916\n",
860 |       "3754532/3754532 [==============================] - 829s 221us/sample - loss: 0.6521 - accuracy: 0.7426 - val_loss: 0.6410 - val_accuracy: 0.7048\n"
861 |      ]
862 |     }
863 |    ],
864 |    "source": [
865 |     "EPOCHS = 1000\n",
866 |     "model_filename = \"model_resnet_walkforward_tcn_actual.h5\"\n",
867 |     "hist_filename = \"hist_model_walkforward_tcn_actual.csv\"\n",
868 |     "early_stop = keras.callbacks.EarlyStopping(\n",
869 |     "                monitor='val_loss',\n",
870 |     "                patience=4)\n",
871 |     "model_save_checkpoint = keras.callbacks.ModelCheckpoint(\n",
872 |     "                            filepath= model_filename,\n",
873 |     "                            save_best_only=True,\n",
874 |     "                            monitor='val_loss',\n",
875 |     "                            verbose=1)\n",
876 |     "reduce_lr = keras.callbacks.ReduceLROnPlateau(\n",
877 |     "                monitor='val_loss',\n",
878 |     "                factor=0.5,\n",
879 |     "                patience=2,\n",
880 |     "                verbose=1,\n",
881 |     "                min_lr=0.0001)\n",
882 |     "callbacks = [reduce_lr, model_save_checkpoint, early_stop]\n",
883 |     "history = m.fit(\n",
884 |     "            X_train_val, y_train_val,\n",
885 |     "            epochs=EPOCHS,\n",
886 |     "            batch_size=128,\n",
887 |     "            callbacks=callbacks,\n",
888 |     "            class_weight = d_class_weights,\n",
889 |     "            validation_data = (X_test, y_test),\n",
890 |     ")\n"
891 |    ]
892 |   },
893 |   {
894 |    "cell_type": "code",
895 |    "execution_count": 16,
896 |    "metadata": {},
897 |    "outputs": [],
898 |    "source": [
899 |     "y_pred = m.predict(X_test).argmax(axis=1)"
900 |    ]
901 |   },
902 |   {
903 |    "cell_type": "code",
904 |    "execution_count": 17,
905 |    "metadata": {},
906 |    "outputs": [
907 |     {
908 |      "data": {
909 |       "image/png": "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\n",
910 |       "text/plain": [
911 |        "<Figure size 432x288 with 2 Axes>"
912 |       ]
913 |      },
914 |      "metadata": {
915 |       "needs_background": "light"
916 |      },
917 |      "output_type": "display_data"
918 |     }
919 |    ],
920 |    "source": [
921 |     "from sklearn.metrics import confusion_matrix\n",
922 |     "\n",
923 |     "cm = confusion_matrix(y_pred=y_pred, y_true=y_test)\n",
924 |     "    \n",
925 |     "cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]\n",
926 |     "\n",
927 |     "ax=sns.heatmap(cm, annot=True, xticklabels=['Down','Stable','Up'], yticklabels=['Down','Stable','Up'], cmap='Blues')\n",
928 |     "ax.set_ylim(3.0, 0)\n",
929 |     "plt.savefig('image')"
930 |    ]
931 |   },
932 |   {
933 |    "cell_type": "code",
934 |    "execution_count": 18,
935 |    "metadata": {},
936 |    "outputs": [
937 |     {
938 |      "name": "stdout",
939 |      "output_type": "stream",
940 |      "text": [
941 |       "             precision    recall  f1-score   support\n",
942 |       "\n",
943 |       "       Down       0.45      0.55      0.50     61142\n",
944 |       "     Stable       0.89      0.73      0.80    350499\n",
945 |       "         Up       0.37      0.71      0.49     53759\n",
946 |       "\n",
947 |       "avg / total       0.77      0.70      0.73    465400\n",
948 |       "\n"
949 |      ]
950 |     }
951 |    ],
952 |    "source": [
953 |     "from sklearn.metrics import classification_report\n",
954 |     "print(classification_report(y_true = y_test,y_pred = y_pred,  target_names=['Down', 'Stable', 'Up']))"
955 |    ]
956 |   },
957 |   {
958 |    "cell_type": "code",
959 |    "execution_count": null,
960 |    "metadata": {},
961 |    "outputs": [],
962 |    "source": []
963 |   }
964 |  ],
965 |  "metadata": {
966 |   "kernelspec": {
967 |    "display_name": "Python 3",
968 |    "language": "python",
969 |    "name": "python3"
970 |   },
971 |   "language_info": {
972 |    "codemirror_mode": {
973 |     "name": "ipython",
974 |     "version": 3
975 |    },
976 |    "file_extension": ".py",
977 |    "mimetype": "text/x-python",
978 |    "name": "python",
979 |    "nbconvert_exporter": "python",
980 |    "pygments_lexer": "ipython3",
981 |    "version": "3.5.3"
982 |   }
983 |  },
984 |  "nbformat": 4,
985 |  "nbformat_minor": 4
986 | }
987 | 


--------------------------------------------------------------------------------
/Walkthrough.ipynb:
--------------------------------------------------------------------------------
  1 | {
  2 |  "cells": [
  3 |   {
  4 |    "cell_type": "code",
  5 |    "execution_count": 1,
  6 |    "metadata": {},
  7 |    "outputs": [],
  8 |    "source": [
  9 |     "import gc\n",
 10 |     "import numpy as np\n",
 11 |     "import pandas as pd\n",
 12 |     "from tqdm import tqdm\n",
 13 |     "from google.cloud import storage\n",
 14 |     "from io import StringIO\n",
 15 |     "%matplotlib inline\n",
 16 |     "import seaborn as sns\n",
 17 |     "import matplotlib.pyplot as plt"
 18 |    ]
 19 |   },
 20 |   {
 21 |    "cell_type": "code",
 22 |    "execution_count": 2,
 23 |    "metadata": {},
 24 |    "outputs": [],
 25 |    "source": [
 26 |     "client = storage.Client()\n",
 27 |     "bucket = client.get_bucket('bucket_l2_snapshot')"
 28 |    ]
 29 |   },
 30 |   {
 31 |    "cell_type": "code",
 32 |    "execution_count": 3,
 33 |    "metadata": {},
 34 |    "outputs": [],
 35 |    "source": [
 36 |     "blob = bucket.blob('l2_snapshot_100ms.csv')\n",
 37 |     "bt = blob.download_as_string()\n",
 38 |     "s = str(bt, 'utf-8')\n",
 39 |     "s = StringIO(s)"
 40 |    ]
 41 |   },
 42 |   {
 43 |    "cell_type": "code",
 44 |    "execution_count": 4,
 45 |    "metadata": {},
 46 |    "outputs": [],
 47 |    "source": [
 48 |     "df = pd.read_csv(s, index_col=0, infer_datetime_format=True)"
 49 |    ]
 50 |   },
 51 |   {
 52 |    "cell_type": "code",
 53 |    "execution_count": 5,
 54 |    "metadata": {},
 55 |    "outputs": [
 56 |     {
 57 |      "data": {
 58 |       "text/html": [
 59 |        "<div>\n",
 60 |        "<style scoped>\n",
 61 |        "    .dataframe tbody tr th:only-of-type {\n",
 62 |        "        vertical-align: middle;\n",
 63 |        "    }\n",
 64 |        "\n",
 65 |        "    .dataframe tbody tr th {\n",
 66 |        "        vertical-align: top;\n",
 67 |        "    }\n",
 68 |        "\n",
 69 |        "    .dataframe thead th {\n",
 70 |        "        text-align: right;\n",
 71 |        "    }\n",
 72 |        "</style>\n",
 73 |        "<table border=\"1\" class=\"dataframe\">\n",
 74 |        "  <thead>\n",
 75 |        "    <tr style=\"text-align: right;\">\n",
 76 |        "      <th></th>\n",
 77 |        "      <th>b1</th>\n",
 78 |        "      <th>b2</th>\n",
 79 |        "      <th>b3</th>\n",
 80 |        "      <th>b4</th>\n",
 81 |        "      <th>b5</th>\n",
 82 |        "      <th>b6</th>\n",
 83 |        "      <th>b7</th>\n",
 84 |        "      <th>b8</th>\n",
 85 |        "      <th>b9</th>\n",
 86 |        "      <th>b10</th>\n",
 87 |        "      <th>...</th>\n",
 88 |        "      <th>aq1</th>\n",
 89 |        "      <th>aq2</th>\n",
 90 |        "      <th>aq3</th>\n",
 91 |        "      <th>aq4</th>\n",
 92 |        "      <th>aq5</th>\n",
 93 |        "      <th>aq6</th>\n",
 94 |        "      <th>aq7</th>\n",
 95 |        "      <th>aq8</th>\n",
 96 |        "      <th>aq9</th>\n",
 97 |        "      <th>aq10</th>\n",
 98 |        "    </tr>\n",
 99 |        "    <tr>\n",
100 |        "      <th>timestamp</th>\n",
101 |        "      <th></th>\n",
102 |        "      <th></th>\n",
103 |        "      <th></th>\n",
104 |        "      <th></th>\n",
105 |        "      <th></th>\n",
106 |        "      <th></th>\n",
107 |        "      <th></th>\n",
108 |        "      <th></th>\n",
109 |        "      <th></th>\n",
110 |        "      <th></th>\n",
111 |        "      <th></th>\n",
112 |        "      <th></th>\n",
113 |        "      <th></th>\n",
114 |        "      <th></th>\n",
115 |        "      <th></th>\n",
116 |        "      <th></th>\n",
117 |        "      <th></th>\n",
118 |        "      <th></th>\n",
119 |        "      <th></th>\n",
120 |        "      <th></th>\n",
121 |        "      <th></th>\n",
122 |        "    </tr>\n",
123 |        "  </thead>\n",
124 |        "  <tbody>\n",
125 |        "    <tr>\n",
126 |        "      <th>2019-11-12 00:00:00.000</th>\n",
127 |        "      <td>8721.53</td>\n",
128 |        "      <td>8720.59</td>\n",
129 |        "      <td>8719.55</td>\n",
130 |        "      <td>8719.50</td>\n",
131 |        "      <td>8719.0</td>\n",
132 |        "      <td>8718.02</td>\n",
133 |        "      <td>8717.87</td>\n",
134 |        "      <td>8717.85</td>\n",
135 |        "      <td>8717.56</td>\n",
136 |        "      <td>8716.06</td>\n",
137 |        "      <td>...</td>\n",
138 |        "      <td>5.802204</td>\n",
139 |        "      <td>2.859</td>\n",
140 |        "      <td>0.12189</td>\n",
141 |        "      <td>0.142575</td>\n",
142 |        "      <td>0.002692</td>\n",
143 |        "      <td>0.46</td>\n",
144 |        "      <td>0.038468</td>\n",
145 |        "      <td>0.200000</td>\n",
146 |        "      <td>0.001890</td>\n",
147 |        "      <td>1.00000</td>\n",
148 |        "    </tr>\n",
149 |        "    <tr>\n",
150 |        "      <th>2019-11-12 00:00:00.100</th>\n",
151 |        "      <td>8721.53</td>\n",
152 |        "      <td>8720.59</td>\n",
153 |        "      <td>8719.56</td>\n",
154 |        "      <td>8719.55</td>\n",
155 |        "      <td>8719.0</td>\n",
156 |        "      <td>8718.02</td>\n",
157 |        "      <td>8717.87</td>\n",
158 |        "      <td>8717.85</td>\n",
159 |        "      <td>8717.56</td>\n",
160 |        "      <td>8716.06</td>\n",
161 |        "      <td>...</td>\n",
162 |        "      <td>5.802204</td>\n",
163 |        "      <td>2.859</td>\n",
164 |        "      <td>0.12189</td>\n",
165 |        "      <td>0.142575</td>\n",
166 |        "      <td>0.002692</td>\n",
167 |        "      <td>0.46</td>\n",
168 |        "      <td>0.038468</td>\n",
169 |        "      <td>0.200000</td>\n",
170 |        "      <td>0.001890</td>\n",
171 |        "      <td>1.00000</td>\n",
172 |        "    </tr>\n",
173 |        "    <tr>\n",
174 |        "      <th>2019-11-12 00:00:00.200</th>\n",
175 |        "      <td>8721.53</td>\n",
176 |        "      <td>8720.59</td>\n",
177 |        "      <td>8719.61</td>\n",
178 |        "      <td>8719.56</td>\n",
179 |        "      <td>8719.0</td>\n",
180 |        "      <td>8718.02</td>\n",
181 |        "      <td>8717.87</td>\n",
182 |        "      <td>8717.85</td>\n",
183 |        "      <td>8716.06</td>\n",
184 |        "      <td>8716.00</td>\n",
185 |        "      <td>...</td>\n",
186 |        "      <td>5.802204</td>\n",
187 |        "      <td>2.859</td>\n",
188 |        "      <td>0.12189</td>\n",
189 |        "      <td>0.142575</td>\n",
190 |        "      <td>0.002692</td>\n",
191 |        "      <td>0.46</td>\n",
192 |        "      <td>0.038468</td>\n",
193 |        "      <td>0.200000</td>\n",
194 |        "      <td>1.086016</td>\n",
195 |        "      <td>0.00189</td>\n",
196 |        "    </tr>\n",
197 |        "    <tr>\n",
198 |        "      <th>2019-11-12 00:00:00.300</th>\n",
199 |        "      <td>8721.53</td>\n",
200 |        "      <td>8720.59</td>\n",
201 |        "      <td>8719.61</td>\n",
202 |        "      <td>8719.56</td>\n",
203 |        "      <td>8719.0</td>\n",
204 |        "      <td>8718.02</td>\n",
205 |        "      <td>8718.00</td>\n",
206 |        "      <td>8717.87</td>\n",
207 |        "      <td>8717.85</td>\n",
208 |        "      <td>8716.60</td>\n",
209 |        "      <td>...</td>\n",
210 |        "      <td>7.802204</td>\n",
211 |        "      <td>2.859</td>\n",
212 |        "      <td>0.12189</td>\n",
213 |        "      <td>0.142575</td>\n",
214 |        "      <td>0.002692</td>\n",
215 |        "      <td>0.46</td>\n",
216 |        "      <td>0.038468</td>\n",
217 |        "      <td>0.200000</td>\n",
218 |        "      <td>1.086016</td>\n",
219 |        "      <td>0.00189</td>\n",
220 |        "    </tr>\n",
221 |        "    <tr>\n",
222 |        "      <th>2019-11-12 00:00:00.400</th>\n",
223 |        "      <td>8721.53</td>\n",
224 |        "      <td>8720.59</td>\n",
225 |        "      <td>8719.61</td>\n",
226 |        "      <td>8719.56</td>\n",
227 |        "      <td>8719.0</td>\n",
228 |        "      <td>8718.99</td>\n",
229 |        "      <td>8718.02</td>\n",
230 |        "      <td>8718.00</td>\n",
231 |        "      <td>8717.87</td>\n",
232 |        "      <td>8717.85</td>\n",
233 |        "      <td>...</td>\n",
234 |        "      <td>7.802204</td>\n",
235 |        "      <td>2.859</td>\n",
236 |        "      <td>0.12189</td>\n",
237 |        "      <td>0.142575</td>\n",
238 |        "      <td>0.002692</td>\n",
239 |        "      <td>0.46</td>\n",
240 |        "      <td>0.200000</td>\n",
241 |        "      <td>0.038468</td>\n",
242 |        "      <td>1.086016</td>\n",
243 |        "      <td>0.00189</td>\n",
244 |        "    </tr>\n",
245 |        "  </tbody>\n",
246 |        "</table>\n",
247 |        "<p>5 rows × 40 columns</p>\n",
248 |        "</div>"
249 |       ],
250 |       "text/plain": [
251 |        "                              b1       b2       b3       b4      b5       b6  \\\n",
252 |        "timestamp                                                                      \n",
253 |        "2019-11-12 00:00:00.000  8721.53  8720.59  8719.55  8719.50  8719.0  8718.02   \n",
254 |        "2019-11-12 00:00:00.100  8721.53  8720.59  8719.56  8719.55  8719.0  8718.02   \n",
255 |        "2019-11-12 00:00:00.200  8721.53  8720.59  8719.61  8719.56  8719.0  8718.02   \n",
256 |        "2019-11-12 00:00:00.300  8721.53  8720.59  8719.61  8719.56  8719.0  8718.02   \n",
257 |        "2019-11-12 00:00:00.400  8721.53  8720.59  8719.61  8719.56  8719.0  8718.99   \n",
258 |        "\n",
259 |        "                              b7       b8       b9      b10  ...       aq1  \\\n",
260 |        "timestamp                                                    ...             \n",
261 |        "2019-11-12 00:00:00.000  8717.87  8717.85  8717.56  8716.06  ...  5.802204   \n",
262 |        "2019-11-12 00:00:00.100  8717.87  8717.85  8717.56  8716.06  ...  5.802204   \n",
263 |        "2019-11-12 00:00:00.200  8717.87  8717.85  8716.06  8716.00  ...  5.802204   \n",
264 |        "2019-11-12 00:00:00.300  8718.00  8717.87  8717.85  8716.60  ...  7.802204   \n",
265 |        "2019-11-12 00:00:00.400  8718.02  8718.00  8717.87  8717.85  ...  7.802204   \n",
266 |        "\n",
267 |        "                           aq2      aq3       aq4       aq5   aq6       aq7  \\\n",
268 |        "timestamp                                                                     \n",
269 |        "2019-11-12 00:00:00.000  2.859  0.12189  0.142575  0.002692  0.46  0.038468   \n",
270 |        "2019-11-12 00:00:00.100  2.859  0.12189  0.142575  0.002692  0.46  0.038468   \n",
271 |        "2019-11-12 00:00:00.200  2.859  0.12189  0.142575  0.002692  0.46  0.038468   \n",
272 |        "2019-11-12 00:00:00.300  2.859  0.12189  0.142575  0.002692  0.46  0.038468   \n",
273 |        "2019-11-12 00:00:00.400  2.859  0.12189  0.142575  0.002692  0.46  0.200000   \n",
274 |        "\n",
275 |        "                              aq8       aq9     aq10  \n",
276 |        "timestamp                                             \n",
277 |        "2019-11-12 00:00:00.000  0.200000  0.001890  1.00000  \n",
278 |        "2019-11-12 00:00:00.100  0.200000  0.001890  1.00000  \n",
279 |        "2019-11-12 00:00:00.200  0.200000  1.086016  0.00189  \n",
280 |        "2019-11-12 00:00:00.300  0.200000  1.086016  0.00189  \n",
281 |        "2019-11-12 00:00:00.400  0.038468  1.086016  0.00189  \n",
282 |        "\n",
283 |        "[5 rows x 40 columns]"
284 |       ]
285 |      },
286 |      "execution_count": 5,
287 |      "metadata": {},
288 |      "output_type": "execute_result"
289 |     }
290 |    ],
291 |    "source": [
292 |     "df.head()"
293 |    ]
294 |   },
295 |   {
296 |    "cell_type": "code",
297 |    "execution_count": 6,
298 |    "metadata": {},
299 |    "outputs": [],
300 |    "source": [
301 |     "df.index = pd.to_datetime(df.index)"
302 |    ]
303 |   },
304 |   {
305 |    "cell_type": "code",
306 |    "execution_count": 7,
307 |    "metadata": {},
308 |    "outputs": [],
309 |    "source": [
310 |     "def concat_data(base_path, features):\n",
311 |     "    \"\"\"Concatenate all the files in basepath keeping only the\n",
312 |     "    columns specified by features.\n",
313 |     "    \"\"\"\n",
314 |     "    l2_snapshot = pd.DataFrame()\n",
315 |     "    for i, x in enumerate(tqdm(sorted(os.listdir(base_path)))):\n",
316 |     "        if base_path is None:\n",
317 |     "            path = x\n",
318 |     "        else:\n",
319 |     "            path = base_path + x\n",
320 |     "        df_hour = pd.read_parquet(path, columns=features)\n",
321 |     "        l2_snapshot = pd.concat([l2_snapshot, df_hour.dropna()])\n",
322 |     "        \n",
323 |     "    return l2_snapshot\n",
324 |     "\n",
325 |     "def extend_matrix(A, n):\n",
326 |     "    \"\"\"Extend a matrix A by duplicating rows as specified by the list n.\"\"\"\n",
327 |     "    n = n[1:]  # Do not duplicate rows for the first day\n",
328 |     "    A = A[:-1]  # Do not duplicate the last day's row\n",
329 |     "    A = np.repeat(A, repeats=n, axis=0)\n",
330 |     "    return A\n",
331 |     "\n",
332 |     "def normalise_data_per_day(df):\n",
333 |     "    df_mean = df.resample('D').mean()\n",
334 |     "    df_var = df.resample('D').var()\n",
335 |     "    \n",
336 |     "    timestamps_per_day = np.unique(df.index.date, return_counts=True)[1]\n",
337 |     "    mean_array = extend_matrix(df_mean.to_numpy(), timestamps_per_day)\n",
338 |     "    var_array = extend_matrix(df_var.to_numpy(), timestamps_per_day)\n",
339 |     "    \n",
340 |     "    # Drop the rows of the first day\n",
341 |     "    df = df[df.index.date != df.index[0].date()]\n",
342 |     "    \n",
343 |     "    df = (df - mean_array) / np.sqrt(var_array)\n",
344 |     "\n",
345 |     "    return df\n",
346 |     "\n",
347 |     "def balance_classes(y):\n",
348 |     "    unique = np.unique(y, return_counts=True)\n",
349 |     "\n",
350 |     "    # Take smallest number as class size\n",
351 |     "    class_size = np.min(unique[1])\n",
352 |     "    class_size_index = np.argmin(unique[1])\n",
353 |     "    timestamps = np.array([], dtype=int)\n",
354 |     "    for i, category in enumerate(unique[0]):\n",
355 |     "        if i == class_size_index:\n",
356 |     "            continue\n",
357 |     "        index = np.argwhere(y==category)\n",
358 |     "        index = index.reshape(len(index))\n",
359 |     "        random_timestamps = np.random.choice(index, (unique[1][i] - class_size), replace=False)\n",
360 |     "        timestamps = np.concatenate((timestamps, random_timestamps), axis=None)\n",
361 |     "        \n",
362 |     "    return timestamps\n",
363 |     "\n",
364 |     "def generate_y(df_snapshot, T=100, D=40, best_ask='a1', best_bid='b1', k=20, alpha=10e-5):\n",
365 |     "    \"\"\"Return X, y from the snapshot dataframe and the best ask/bid columns.\"\"\"\n",
366 |     "    df = pd.DataFrame()\n",
367 |     "    df['mid_price'] = (df_snapshot[best_ask].to_numpy()+df_snapshot[best_bid].to_numpy())/2\n",
368 |     "\n",
369 |     "    # Create columns delayed by -k to k-1\n",
370 |     "    for i in range(-k, k):\n",
371 |     "        df[i] = df['mid_price'].shift(periods=i)\n",
372 |     "\n",
373 |     "    # Drop first k-1 rows and last k rows\n",
374 |     "    df.drop(range(0,20), axis=0, inplace=True)\n",
375 |     "    df.drop(range(len(df_snapshot)-20,len(df_snapshot)), axis=0, inplace=True)\n",
376 |     "    \n",
377 |     "    # Compute mean of previous k and next k\n",
378 |     "    df['m_b'] = df[range(0,20)].mean(axis=1)\n",
379 |     "    df['m_a'] = df[range(-20,0)].mean(axis=1)\n",
380 |     "    \n",
381 |     "    # Compute label of increasing or decreasing\n",
382 |     "    y_increase = np.where(df['m_b'] > df['m_a'] * (1+alpha), 1, 0)\n",
383 |     "    y_decrease = np.where(df['m_b'] < df['m_a'] * (1-alpha), -1, 0)\n",
384 |     "    y = y_increase + y_decrease\n",
385 |     "\n",
386 |     "    # 100 most recent limit orders used so ignore first 100 timesteps\n",
387 |     "    y = y[T:]\n",
388 |     "    y += 1\n",
389 |     "\n",
390 |     "    return y\n",
391 |     "\n",
392 |     "def generate_preX(df_snapshot):\n",
393 |     "    # First and last 20 can't create labels for as previous and next k=20 needed\n",
394 |     "    preX = df_snapshot.to_numpy()[20:-20]\n",
395 |     "\n",
396 |     "    return preX\n",
397 |     "\n",
398 |     "def generate_X(preX, T=100, D=40):\n",
399 |     "    # For each timestep create matrix of 100 most recent limit orders\n",
400 |     "    X = np.array([preX[t:t+T] for t in range(len(preX)-T)], dtype='float32')\n",
401 |     "        \n",
402 |     "    return X"
403 |    ]
404 |   },
405 |   {
406 |    "cell_type": "code",
407 |    "execution_count": 8,
408 |    "metadata": {},
409 |    "outputs": [],
410 |    "source": [
411 |     "l2_norm = normalise_data_per_day(df)"
412 |    ]
413 |   },
414 |   {
415 |    "cell_type": "code",
416 |    "execution_count": 9,
417 |    "metadata": {},
418 |    "outputs": [
419 |     {
420 |      "data": {
421 |       "text/html": [
422 |        "<div>\n",
423 |        "<style scoped>\n",
424 |        "    .dataframe tbody tr th:only-of-type {\n",
425 |        "        vertical-align: middle;\n",
426 |        "    }\n",
427 |        "\n",
428 |        "    .dataframe tbody tr th {\n",
429 |        "        vertical-align: top;\n",
430 |        "    }\n",
431 |        "\n",
432 |        "    .dataframe thead th {\n",
433 |        "        text-align: right;\n",
434 |        "    }\n",
435 |        "</style>\n",
436 |        "<table border=\"1\" class=\"dataframe\">\n",
437 |        "  <thead>\n",
438 |        "    <tr style=\"text-align: right;\">\n",
439 |        "      <th></th>\n",
440 |        "      <th>b1</th>\n",
441 |        "      <th>b2</th>\n",
442 |        "      <th>b3</th>\n",
443 |        "      <th>b4</th>\n",
444 |        "      <th>b5</th>\n",
445 |        "      <th>b6</th>\n",
446 |        "      <th>b7</th>\n",
447 |        "      <th>b8</th>\n",
448 |        "      <th>b9</th>\n",
449 |        "      <th>b10</th>\n",
450 |        "      <th>...</th>\n",
451 |        "      <th>aq1</th>\n",
452 |        "      <th>aq2</th>\n",
453 |        "      <th>aq3</th>\n",
454 |        "      <th>aq4</th>\n",
455 |        "      <th>aq5</th>\n",
456 |        "      <th>aq6</th>\n",
457 |        "      <th>aq7</th>\n",
458 |        "      <th>aq8</th>\n",
459 |        "      <th>aq9</th>\n",
460 |        "      <th>aq10</th>\n",
461 |        "    </tr>\n",
462 |        "    <tr>\n",
463 |        "      <th>timestamp</th>\n",
464 |        "      <th></th>\n",
465 |        "      <th></th>\n",
466 |        "      <th></th>\n",
467 |        "      <th></th>\n",
468 |        "      <th></th>\n",
469 |        "      <th></th>\n",
470 |        "      <th></th>\n",
471 |        "      <th></th>\n",
472 |        "      <th></th>\n",
473 |        "      <th></th>\n",
474 |        "      <th></th>\n",
475 |        "      <th></th>\n",
476 |        "      <th></th>\n",
477 |        "      <th></th>\n",
478 |        "      <th></th>\n",
479 |        "      <th></th>\n",
480 |        "      <th></th>\n",
481 |        "      <th></th>\n",
482 |        "      <th></th>\n",
483 |        "      <th></th>\n",
484 |        "      <th></th>\n",
485 |        "    </tr>\n",
486 |        "  </thead>\n",
487 |        "  <tbody>\n",
488 |        "    <tr>\n",
489 |        "      <th>2019-11-13 00:00:00.000</th>\n",
490 |        "      <td>2.069669</td>\n",
491 |        "      <td>2.066221</td>\n",
492 |        "      <td>2.049359</td>\n",
493 |        "      <td>2.032164</td>\n",
494 |        "      <td>2.045755</td>\n",
495 |        "      <td>2.056132</td>\n",
496 |        "      <td>2.062144</td>\n",
497 |        "      <td>2.072841</td>\n",
498 |        "      <td>2.064472</td>\n",
499 |        "      <td>2.076723</td>\n",
500 |        "      <td>...</td>\n",
501 |        "      <td>-0.630327</td>\n",
502 |        "      <td>-0.585376</td>\n",
503 |        "      <td>-0.603213</td>\n",
504 |        "      <td>-0.584939</td>\n",
505 |        "      <td>-0.568453</td>\n",
506 |        "      <td>-0.547841</td>\n",
507 |        "      <td>-0.503235</td>\n",
508 |        "      <td>-0.459792</td>\n",
509 |        "      <td>-0.545609</td>\n",
510 |        "      <td>-0.581027</td>\n",
511 |        "    </tr>\n",
512 |        "    <tr>\n",
513 |        "      <th>2019-11-13 00:00:00.100</th>\n",
514 |        "      <td>2.096553</td>\n",
515 |        "      <td>2.099017</td>\n",
516 |        "      <td>2.099095</td>\n",
517 |        "      <td>2.093216</td>\n",
518 |        "      <td>2.076695</td>\n",
519 |        "      <td>2.060169</td>\n",
520 |        "      <td>2.073992</td>\n",
521 |        "      <td>2.084426</td>\n",
522 |        "      <td>2.090621</td>\n",
523 |        "      <td>2.101545</td>\n",
524 |        "      <td>...</td>\n",
525 |        "      <td>-0.630484</td>\n",
526 |        "      <td>-0.585376</td>\n",
527 |        "      <td>-0.603213</td>\n",
528 |        "      <td>-0.584939</td>\n",
529 |        "      <td>-0.568453</td>\n",
530 |        "      <td>-0.547841</td>\n",
531 |        "      <td>-0.559809</td>\n",
532 |        "      <td>-0.556070</td>\n",
533 |        "      <td>-0.552547</td>\n",
534 |        "      <td>-0.576830</td>\n",
535 |        "    </tr>\n",
536 |        "    <tr>\n",
537 |        "      <th>2019-11-13 00:00:00.200</th>\n",
538 |        "      <td>2.096553</td>\n",
539 |        "      <td>2.099017</td>\n",
540 |        "      <td>2.099095</td>\n",
541 |        "      <td>2.093216</td>\n",
542 |        "      <td>2.076695</td>\n",
543 |        "      <td>2.060169</td>\n",
544 |        "      <td>2.073992</td>\n",
545 |        "      <td>2.084426</td>\n",
546 |        "      <td>2.090621</td>\n",
547 |        "      <td>2.101545</td>\n",
548 |        "      <td>...</td>\n",
549 |        "      <td>-0.630484</td>\n",
550 |        "      <td>-0.585376</td>\n",
551 |        "      <td>-0.603213</td>\n",
552 |        "      <td>-0.584939</td>\n",
553 |        "      <td>-0.568453</td>\n",
554 |        "      <td>-0.547841</td>\n",
555 |        "      <td>-0.559809</td>\n",
556 |        "      <td>-0.556070</td>\n",
557 |        "      <td>-0.552547</td>\n",
558 |        "      <td>-0.576830</td>\n",
559 |        "    </tr>\n",
560 |        "    <tr>\n",
561 |        "      <th>2019-11-13 00:00:00.300</th>\n",
562 |        "      <td>2.096553</td>\n",
563 |        "      <td>2.085576</td>\n",
564 |        "      <td>2.096676</td>\n",
565 |        "      <td>2.062824</td>\n",
566 |        "      <td>2.046024</td>\n",
567 |        "      <td>2.059900</td>\n",
568 |        "      <td>2.070222</td>\n",
569 |        "      <td>2.076344</td>\n",
570 |        "      <td>2.087117</td>\n",
571 |        "      <td>2.078882</td>\n",
572 |        "      <td>...</td>\n",
573 |        "      <td>-0.630484</td>\n",
574 |        "      <td>-0.585376</td>\n",
575 |        "      <td>-0.603213</td>\n",
576 |        "      <td>-0.584939</td>\n",
577 |        "      <td>-0.568453</td>\n",
578 |        "      <td>-0.547841</td>\n",
579 |        "      <td>-0.559809</td>\n",
580 |        "      <td>-0.556070</td>\n",
581 |        "      <td>-0.552547</td>\n",
582 |        "      <td>-0.576830</td>\n",
583 |        "    </tr>\n",
584 |        "    <tr>\n",
585 |        "      <th>2019-11-13 00:00:00.400</th>\n",
586 |        "      <td>2.096553</td>\n",
587 |        "      <td>2.085576</td>\n",
588 |        "      <td>2.096676</td>\n",
589 |        "      <td>2.062824</td>\n",
590 |        "      <td>2.046024</td>\n",
591 |        "      <td>2.059900</td>\n",
592 |        "      <td>2.070222</td>\n",
593 |        "      <td>2.076344</td>\n",
594 |        "      <td>2.064472</td>\n",
595 |        "      <td>2.076723</td>\n",
596 |        "      <td>...</td>\n",
597 |        "      <td>-0.630484</td>\n",
598 |        "      <td>-0.585376</td>\n",
599 |        "      <td>-0.603213</td>\n",
600 |        "      <td>-0.584939</td>\n",
601 |        "      <td>-0.568453</td>\n",
602 |        "      <td>-0.547841</td>\n",
603 |        "      <td>-0.559809</td>\n",
604 |        "      <td>-0.556070</td>\n",
605 |        "      <td>-0.552547</td>\n",
606 |        "      <td>-0.576830</td>\n",
607 |        "    </tr>\n",
608 |        "  </tbody>\n",
609 |        "</table>\n",
610 |        "<p>5 rows × 40 columns</p>\n",
611 |        "</div>"
612 |       ],
613 |       "text/plain": [
614 |        "                               b1        b2        b3        b4        b5  \\\n",
615 |        "timestamp                                                                   \n",
616 |        "2019-11-13 00:00:00.000  2.069669  2.066221  2.049359  2.032164  2.045755   \n",
617 |        "2019-11-13 00:00:00.100  2.096553  2.099017  2.099095  2.093216  2.076695   \n",
618 |        "2019-11-13 00:00:00.200  2.096553  2.099017  2.099095  2.093216  2.076695   \n",
619 |        "2019-11-13 00:00:00.300  2.096553  2.085576  2.096676  2.062824  2.046024   \n",
620 |        "2019-11-13 00:00:00.400  2.096553  2.085576  2.096676  2.062824  2.046024   \n",
621 |        "\n",
622 |        "                               b6        b7        b8        b9       b10  \\\n",
623 |        "timestamp                                                                   \n",
624 |        "2019-11-13 00:00:00.000  2.056132  2.062144  2.072841  2.064472  2.076723   \n",
625 |        "2019-11-13 00:00:00.100  2.060169  2.073992  2.084426  2.090621  2.101545   \n",
626 |        "2019-11-13 00:00:00.200  2.060169  2.073992  2.084426  2.090621  2.101545   \n",
627 |        "2019-11-13 00:00:00.300  2.059900  2.070222  2.076344  2.087117  2.078882   \n",
628 |        "2019-11-13 00:00:00.400  2.059900  2.070222  2.076344  2.064472  2.076723   \n",
629 |        "\n",
630 |        "                         ...       aq1       aq2       aq3       aq4  \\\n",
631 |        "timestamp                ...                                           \n",
632 |        "2019-11-13 00:00:00.000  ... -0.630327 -0.585376 -0.603213 -0.584939   \n",
633 |        "2019-11-13 00:00:00.100  ... -0.630484 -0.585376 -0.603213 -0.584939   \n",
634 |        "2019-11-13 00:00:00.200  ... -0.630484 -0.585376 -0.603213 -0.584939   \n",
635 |        "2019-11-13 00:00:00.300  ... -0.630484 -0.585376 -0.603213 -0.584939   \n",
636 |        "2019-11-13 00:00:00.400  ... -0.630484 -0.585376 -0.603213 -0.584939   \n",
637 |        "\n",
638 |        "                              aq5       aq6       aq7       aq8       aq9  \\\n",
639 |        "timestamp                                                                   \n",
640 |        "2019-11-13 00:00:00.000 -0.568453 -0.547841 -0.503235 -0.459792 -0.545609   \n",
641 |        "2019-11-13 00:00:00.100 -0.568453 -0.547841 -0.559809 -0.556070 -0.552547   \n",
642 |        "2019-11-13 00:00:00.200 -0.568453 -0.547841 -0.559809 -0.556070 -0.552547   \n",
643 |        "2019-11-13 00:00:00.300 -0.568453 -0.547841 -0.559809 -0.556070 -0.552547   \n",
644 |        "2019-11-13 00:00:00.400 -0.568453 -0.547841 -0.559809 -0.556070 -0.552547   \n",
645 |        "\n",
646 |        "                             aq10  \n",
647 |        "timestamp                          \n",
648 |        "2019-11-13 00:00:00.000 -0.581027  \n",
649 |        "2019-11-13 00:00:00.100 -0.576830  \n",
650 |        "2019-11-13 00:00:00.200 -0.576830  \n",
651 |        "2019-11-13 00:00:00.300 -0.576830  \n",
652 |        "2019-11-13 00:00:00.400 -0.576830  \n",
653 |        "\n",
654 |        "[5 rows x 40 columns]"
655 |       ]
656 |      },
657 |      "execution_count": 9,
658 |      "metadata": {},
659 |      "output_type": "execute_result"
660 |     }
661 |    ],
662 |    "source": [
663 |     "l2_norm.head()"
664 |    ]
665 |   },
666 |   {
667 |    "cell_type": "code",
668 |    "execution_count": 10,
669 |    "metadata": {},
670 |    "outputs": [
671 |     {
672 |      "name": "stdout",
673 |      "output_type": "stream",
674 |      "text": [
675 |       "preX Shape:  (4220032, 40)\n",
676 |       "X shape, y shape:  (4219932, 100, 40) (4219932,)\n",
677 |       "Unique y's:  [0 1 2]\n"
678 |      ]
679 |     }
680 |    ],
681 |    "source": [
682 |     "T = 100\n",
683 |     "D = 40\n",
684 |     "y = generate_y(l2_norm, T=T, D=D, best_ask='a1', best_bid = 'b1', alpha=0.002)\n",
685 |     "preX = generate_preX(l2_norm)\n",
686 |     "print(\"preX Shape: \", preX.shape)\n",
687 |     "X = generate_X(preX)\n",
688 |     "print(\"X shape, y shape: \", X.shape, y.shape)\n",
689 |     "print(\"Unique y's: \", np.unique(y))\n",
690 |     "\n",
691 |     "del preX\n",
692 |     "gc.collect()\n",
693 |     "\n",
694 |     "# First and last 20 are removed to create labels and then last 100 as previous 100 is required for input matrix\n",
695 |     "X_index = l2_norm[20:-120].index\n",
696 |     "\n",
697 |     "# Number of data points for the last day\n",
698 |     "num_test = np.unique(X_index.day, return_counts=True)[1][-1]\n",
699 |     "\n",
700 |     "# Split the data into the first seven days and the last day\n",
701 |     "X_train_val = X[:-num_test]\n",
702 |     "y_train_val = y[:-num_test]\n",
703 |     "X_test = X[-num_test:]\n",
704 |     "y_test = y[-num_test:]"
705 |    ]
706 |   },
707 |   {
708 |    "cell_type": "code",
709 |    "execution_count": 11,
710 |    "metadata": {},
711 |    "outputs": [],
712 |    "source": [
713 |     "import tensorflow as tf\n",
714 |     "from tensorflow.keras.layers import Dense\n",
715 |     "from tensorflow.keras import Input, Model\n",
716 |     "from tcn import TCN, tcn_full_summary"
717 |    ]
718 |   },
719 |   {
720 |    "cell_type": "code",
721 |    "execution_count": 12,
722 |    "metadata": {},
723 |    "outputs": [],
724 |    "source": [
725 |     "from tensorflow import keras"
726 |    ]
727 |   },
728 |   {
729 |    "cell_type": "code",
730 |    "execution_count": 22,
731 |    "metadata": {},
732 |    "outputs": [
733 |     {
734 |      "name": "stdout",
735 |      "output_type": "stream",
736 |      "text": [
737 |       "INFO:tensorflow:Using MirroredStrategy with devices ('/job:localhost/replica:0/task:0/device:GPU:0', '/job:localhost/replica:0/task:0/device:GPU:1', '/job:localhost/replica:0/task:0/device:GPU:2', '/job:localhost/replica:0/task:0/device:GPU:3')\n"
738 |      ]
739 |     }
740 |    ],
741 |    "source": [
742 |     "strategy = tf.distribute.MirroredStrategy()\n",
743 |     "with strategy.scope():\n",
744 |     "    i = Input(batch_shape=(None, 100, 40))\n",
745 |     "    o = TCN(return_sequences=False, use_skip_connections=True, dropout_rate=0.2)(i)\n",
746 |     "    o = Dense(3, activation='softmax')(o)\n",
747 |     "    m = Model(inputs=[i], outputs=[o])\n",
748 |     "    opt = keras.optimizers.Adam(learning_rate=0.001)\n",
749 |     "    m.compile(optimizer=opt, loss='sparse_categorical_crossentropy', metrics=['accuracy'])"
750 |    ]
751 |   },
752 |   {
753 |    "cell_type": "code",
754 |    "execution_count": 23,
755 |    "metadata": {},
756 |    "outputs": [],
757 |    "source": [
758 |     "from sklearn.utils.class_weight import compute_class_weight\n",
759 |     "class_weights = compute_class_weight('balanced', np.unique(y_train_val), y_train_val)\n",
760 |     "d_class_weights = dict(enumerate(class_weights))"
761 |    ]
762 |   },
763 |   {
764 |    "cell_type": "code",
765 |    "execution_count": 24,
766 |    "metadata": {},
767 |    "outputs": [
768 |     {
769 |      "name": "stdout",
770 |      "output_type": "stream",
771 |      "text": [
772 |       "WARNING:tensorflow:sample_weight modes were coerced from\n",
773 |       "  ...\n",
774 |       "    to  \n",
775 |       "  ['...']\n",
776 |       "WARNING:tensorflow:sample_weight modes were coerced from\n",
777 |       "  ...\n",
778 |       "    to  \n",
779 |       "  ['...']\n",
780 |       "Train on 3754532 samples, validate on 465400 samples\n",
781 |       "Epoch 1/1000\n",
782 |       "INFO:tensorflow:batch_all_reduce: 28 all-reduces with algorithm = nccl, num_packs = 1, agg_small_grads_max_bytes = 0 and agg_small_grads_max_group = 10\n",
783 |       "INFO:tensorflow:batch_all_reduce: 28 all-reduces with algorithm = nccl, num_packs = 1, agg_small_grads_max_bytes = 0 and agg_small_grads_max_group = 10\n",
784 |       "3754240/3754532 [============================>.] - ETA: 0s - loss: 0.7837 - accuracy: 0.6785\n",
785 |       "Epoch 00001: val_loss improved from inf to 0.64774, saving model to model_resnet_walkforward_tcn.h5\n",
786 |       "3754532/3754532 [==============================] - 506s 135us/sample - loss: 0.7837 - accuracy: 0.6785 - val_loss: 0.6477 - val_accuracy: 0.7247\n",
787 |       "Epoch 2/1000\n",
788 |       "3754496/3754532 [============================>.] - ETA: 0s - loss: 0.6901 - accuracy: 0.7326\n",
789 |       "Epoch 00002: val_loss improved from 0.64774 to 0.63731, saving model to model_resnet_walkforward_tcn.h5\n",
790 |       "3754532/3754532 [==============================] - 433s 115us/sample - loss: 0.6901 - accuracy: 0.7326 - val_loss: 0.6373 - val_accuracy: 0.7113\n",
791 |       "Epoch 3/1000\n",
792 |       "3754112/3754532 [============================>.] - ETA: 0s - loss: 0.6794 - accuracy: 0.7365\n",
793 |       "Epoch 00003: val_loss did not improve from 0.63731\n",
794 |       "3754532/3754532 [==============================] - 434s 116us/sample - loss: 0.6794 - accuracy: 0.7365 - val_loss: 0.6388 - val_accuracy: 0.7474\n",
795 |       "Epoch 4/1000\n",
796 |       "3754496/3754532 [============================>.] - ETA: 0s - loss: 0.6734 - accuracy: 0.7384\n",
797 |       "Epoch 00004: val_loss improved from 0.63731 to 0.63683, saving model to model_resnet_walkforward_tcn.h5\n",
798 |       "3754532/3754532 [==============================] - 434s 116us/sample - loss: 0.6734 - accuracy: 0.7384 - val_loss: 0.6368 - val_accuracy: 0.7244\n",
799 |       "Epoch 5/1000\n",
800 |       "3754112/3754532 [============================>.] - ETA: 0s - loss: 0.6691 - accuracy: 0.7396\n",
801 |       "Epoch 00005: val_loss improved from 0.63683 to 0.62580, saving model to model_resnet_walkforward_tcn.h5\n",
802 |       "3754532/3754532 [==============================] - 433s 115us/sample - loss: 0.6691 - accuracy: 0.7396 - val_loss: 0.6258 - val_accuracy: 0.7162\n",
803 |       "Epoch 6/1000\n",
804 |       "3754496/3754532 [============================>.] - ETA: 0s - loss: 0.6663 - accuracy: 0.7403\n",
805 |       "Epoch 00006: val_loss did not improve from 0.62580\n",
806 |       "3754532/3754532 [==============================] - 434s 116us/sample - loss: 0.6663 - accuracy: 0.7403 - val_loss: 0.6288 - val_accuracy: 0.7256\n",
807 |       "Epoch 7/1000\n",
808 |       "3754112/3754532 [============================>.] - ETA: 0s - loss: 0.6634 - accuracy: 0.7409\n",
809 |       "Epoch 00007: ReduceLROnPlateau reducing learning rate to 0.00020000000949949026.\n",
810 |       "\n",
811 |       "Epoch 00007: val_loss did not improve from 0.62580\n",
812 |       "3754532/3754532 [==============================] - 434s 115us/sample - loss: 0.6634 - accuracy: 0.7409 - val_loss: 0.6315 - val_accuracy: 0.7172\n",
813 |       "Epoch 8/1000\n",
814 |       "3754112/3754532 [============================>.] - ETA: 0s - loss: 0.6404 - accuracy: 0.7496\n",
815 |       "Epoch 00008: val_loss did not improve from 0.62580\n",
816 |       "3754532/3754532 [==============================] - 436s 116us/sample - loss: 0.6404 - accuracy: 0.7496 - val_loss: 0.6264 - val_accuracy: 0.7098\n",
817 |       "Epoch 9/1000\n",
818 |       "3754496/3754532 [============================>.] - ETA: 0s - loss: 0.6339 - accuracy: 0.7503\n",
819 |       "Epoch 00009: ReduceLROnPlateau reducing learning rate to 0.0001.\n",
820 |       "\n",
821 |       "Epoch 00009: val_loss did not improve from 0.62580\n",
822 |       "3754532/3754532 [==============================] - 435s 116us/sample - loss: 0.6339 - accuracy: 0.7503 - val_loss: 0.6287 - val_accuracy: 0.6852\n"
823 |      ]
824 |     }
825 |    ],
826 |    "source": [
827 |     "EPOCHS = 1000\n",
828 |     "model_filename = \"model_resnet_walkforward_tcn.h5\"\n",
829 |     "hist_filename = \"hist_model_walkforward_tcn.csv\"\n",
830 |     "early_stop = keras.callbacks.EarlyStopping(\n",
831 |     "                monitor='val_loss',\n",
832 |     "                patience=4)\n",
833 |     "model_save_checkpoint = keras.callbacks.ModelCheckpoint(\n",
834 |     "                            filepath= model_filename,\n",
835 |     "                            save_best_only=True,\n",
836 |     "                            monitor='val_loss',\n",
837 |     "                            verbose=1)\n",
838 |     "reduce_lr = keras.callbacks.ReduceLROnPlateau(\n",
839 |     "                monitor='val_loss',\n",
840 |     "                factor=0.2,\n",
841 |     "                patience=2,\n",
842 |     "                verbose=1,\n",
843 |     "                min_lr=0.0001)\n",
844 |     "callbacks = [reduce_lr, model_save_checkpoint, early_stop]\n",
845 |     "history = m.fit(\n",
846 |     "            X_train_val, y_train_val,\n",
847 |     "            epochs=EPOCHS,\n",
848 |     "            batch_size=128,\n",
849 |     "            callbacks=callbacks,\n",
850 |     "            class_weight = d_class_weights,\n",
851 |     "            validation_data = (X_test, y_test),\n",
852 |     ")\n"
853 |    ]
854 |   },
855 |   {
856 |    "cell_type": "code",
857 |    "execution_count": 25,
858 |    "metadata": {},
859 |    "outputs": [],
860 |    "source": [
861 |     "y_pred = m.predict(X_test).argmax(axis=1)"
862 |    ]
863 |   },
864 |   {
865 |    "cell_type": "code",
866 |    "execution_count": 32,
867 |    "metadata": {},
868 |    "outputs": [
869 |     {
870 |      "data": {
871 |       "image/png": "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\n",
872 |       "text/plain": [
873 |        "<Figure size 432x288 with 2 Axes>"
874 |       ]
875 |      },
876 |      "metadata": {
877 |       "needs_background": "light"
878 |      },
879 |      "output_type": "display_data"
880 |     }
881 |    ],
882 |    "source": [
883 |     "from sklearn.metrics import confusion_matrix\n",
884 |     "\n",
885 |     "cm = confusion_matrix(y_pred=y_pred, y_true=y_test)\n",
886 |     "    \n",
887 |     "cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]\n",
888 |     "\n",
889 |     "ax=sns.heatmap(cm, annot=True, xticklabels=['Down','Stable','Up'], yticklabels=['Down','Stable','Up'], cmap='Blues')\n",
890 |     "ax.set_ylim(3.0, 0)\n",
891 |     "plt.savefig('image')"
892 |    ]
893 |   },
894 |   {
895 |    "cell_type": "code",
896 |    "execution_count": 27,
897 |    "metadata": {},
898 |    "outputs": [
899 |     {
900 |      "name": "stdout",
901 |      "output_type": "stream",
902 |      "text": [
903 |       "             precision    recall  f1-score   support\n",
904 |       "\n",
905 |       "          0       0.42      0.63      0.50     61142\n",
906 |       "          1       0.90      0.69      0.78    350499\n",
907 |       "          2       0.36      0.71      0.48     53759\n",
908 |       "\n",
909 |       "avg / total       0.77      0.69      0.71    465400\n",
910 |       "\n"
911 |      ]
912 |     }
913 |    ],
914 |    "source": [
915 |     "from sklearn.metrics import classification_report\n",
916 |     "print(classification_report(y_test,y_pred))"
917 |    ]
918 |   },
919 |   {
920 |    "cell_type": "code",
921 |    "execution_count": 19,
922 |    "metadata": {},
923 |    "outputs": [
924 |     {
925 |      "data": {
926 |       "text/plain": [
927 |        "array([1, 1, 1, ..., 1, 1, 1])"
928 |       ]
929 |      },
930 |      "execution_count": 19,
931 |      "metadata": {},
932 |      "output_type": "execute_result"
933 |     }
934 |    ],
935 |    "source": [
936 |     "y_pred"
937 |    ]
938 |   },
939 |   {
940 |    "cell_type": "code",
941 |    "execution_count": null,
942 |    "metadata": {},
943 |    "outputs": [],
944 |    "source": []
945 |   }
946 |  ],
947 |  "metadata": {
948 |   "kernelspec": {
949 |    "display_name": "Python 3",
950 |    "language": "python",
951 |    "name": "python3"
952 |   },
953 |   "language_info": {
954 |    "codemirror_mode": {
955 |     "name": "ipython",
956 |     "version": 3
957 |    },
958 |    "file_extension": ".py",
959 |    "mimetype": "text/x-python",
960 |    "name": "python",
961 |    "nbconvert_exporter": "python",
962 |    "pygments_lexer": "ipython3",
963 |    "version": "3.5.3"
964 |   }
965 |  },
966 |  "nbformat": 4,
967 |  "nbformat_minor": 4
968 | }
969 | 


--------------------------------------------------------------------------------
/paper/main.tex:
--------------------------------------------------------------------------------
 1 | \documentclass[11pt,a4paper,english]{article}
 2 | \usepackage[T1]{fontenc}
 3 | \usepackage[utf8]{inputenc}
 4 | \usepackage{babel}
 5 | \usepackage{blindtext}
 6 |  
 7 | \title{Deep learning orderbook models for digital assets}
 8 | \author{%
 9 |     Mattijs de-Paepe \\
10 |     \small Cambridge University\\
11 |     \and
12 |     Vivek \\
13 |     \small Cambridge University\\
14 |     \and
15 |     Rakshit Jha \\
16 |     \small Cambridge University\\
17 |     \and
18 |     Ruizhou \\
19 |     \small Cambridge University\\
20 |     \and
21 |     Samuel Holt \\
22 |     \small Fifthrow Technologies\\
23 |     \and
24 |     James West\\
25 |     \small Globe Research\\
26 | }
27 |  
28 | \begin{document}
29 |     \maketitle
30 | 
31 |     \begin{abstract}
32 |     Digital assets (cryptocurrencies) have a far more relaxed regulatory oversight than traditional financial instruments, which means that their orderbooks have a variety of complex phenomena that don’t occur in traditional markets. This is an investigation on how to build a deep learning model to forecast price action based on order books for cryptocurrencies.
33 |     \end{abstract}
34 |     \newpage
35 | 
36 |     \tableofcontents\newpage
37 |     \section{Introduction}
38 |     Digital assets (cryptocurrencies) have a far more relaxed regulatory oversight than traditional financial instruments, which means that their orderbooks have a variety of complex phenomena that don’t occur in traditional markets.
39 |     This is an investigation on how to build a deep learning model to forecast price action based on order books for cryptocurrencies.
40 |     Firstly implementing prior work, and investigating how to improve such models.
41 | 
42 | 
43 |     Extension, determine features that are relevant and are the most important influencing factors for this model.
44 |     \section{Data}
45 |     \section{Methodology}
46 |     \section{Empirical Study}
47 |     \section{Conclusion}
48 | \end{document}
49 | 


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