CNN Deeplearning v2.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import os\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "import seaborn as sns\n",
    "import gc\n",
    "from tqdm import tqdm\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from sklearn.metrics import classification_report"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "paths = 'coinbase_btc_usd/coinbase/btc_usd/l2_snapshots/100ms/'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "register = {}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "l2_snapshot = pd.DataFrame()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(566480, 200)\n",
      "(533024, 200)\n",
      "(517431, 200)\n",
      "(529814, 200)\n",
      "(537967, 200)\n",
      "(511850, 200)\n",
      "(545960, 200)\n",
      "(561058, 200)\n",
      "(535563, 200)\n"
     ]
    }
   ],
   "source": [
    "count = 0\n",
    "day = pd.DataFrame()\n",
    "i = 0\n",
    "for x in os.listdir(paths):\n",
    "    path = paths+x\n",
    "    temp = pd.read_parquet(path)\n",
    "    if count%24 == 0:\n",
    "        day = pd.read_parquet(path)\n",
    "    else:\n",
    "        day = pd.concat([day,temp])\n",
    "  \n",
    "    i+=1\n",
    "    count +=1\n",
    "\n",
    "    if count%24 == 0:\n",
    "        flag = not register\n",
    "        day = day.dropna()\n",
    "        print(day.shape)\n",
    "        result = []\n",
    "        for cols in day.columns:\n",
    "            values = day[cols].to_numpy()\n",
    "            mean = np.mean(values)\n",
    "            std = np.std(values)\n",
    "            result.append([cols,[mean,std]])\n",
    "        register[count/24] = result\n",
    "        if not flag:\n",
    "            for l in range(200):\n",
    "                prev_stat = register[(count/24)-1]\n",
    "                col_name = prev_stat[l][0]\n",
    "                col_mean = prev_stat[l][1][0]\n",
    "                col_std = prev_stat[l][1][1]\n",
    "                values = day[col_name].to_numpy()\n",
    "                values = (values - col_mean)/col_std\n",
    "                day[col_name] = values\n",
    "            l2_snapshot = pd.concat([day,l2_snapshot])\n",
    "        del day\n",
    "  \n",
    "    if i>240:\n",
    "        break"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "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']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "def generate_features_labels(df,ask, bid,k=20, alpha = 10e-5):\n",
    "    df_ = df.copy()\n",
    "    df_['mid_price'] = (df_[ask].to_numpy()+df_[bid].to_numpy())/2\n",
    "    df_['target'] = 1\n",
    "    index = df_.columns.get_loc('mid_price')\n",
    "    target_index = df_.columns.get_loc('target')\n",
    "    shape = df_.shape[0]\n",
    "    y = df_['target'].to_numpy()\n",
    "    for i in tqdm(range(k,shape-k)):\n",
    "        if i==k:\n",
    "            m_b = np.mean(df_.iloc[(i-k):i, index].to_numpy())\n",
    "            m_a = np.mean(df_.iloc[i+1:(i+k+1), index].to_numpy())\n",
    "            val = df_.iloc[i-k,index]\n",
    "            valB = df_.iloc[i+1, index]\n",
    "        else:\n",
    "            curr_val = df_.iloc[i-1,index]\n",
    "            curr_valB = df_.iloc[i+k, index]\n",
    "            m_b = (m_b*k+(curr_val)-(val))/k\n",
    "            m_a = (m_a*k+(curr_valB - valB))/k\n",
    "            val = df_.iloc[(i-k),index]\n",
    "            valB = df_.iloc[(i+1), index]\n",
    "\n",
    "        if (m_b > m_a*(1+alpha)):\n",
    "            y[i] = 2\n",
    "        if (m_b < m_a*(1-alpha)):\n",
    "            y[i] = 0\n",
    "\n",
    "    y = y[k:shape-k]\n",
    "    X = df.iloc[k:shape-k,:].to_numpy()\n",
    "    return X,y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 4272627/4272627 [02:53<00:00, 24654.56it/s]\n",
      "100%|██████████| 4272627/4272627 [02:55<00:00, 24407.77it/s]\n",
      "100%|██████████| 4272627/4272627 [03:02<00:00, 23402.77it/s]\n",
      "100%|██████████| 4272627/4272627 [03:02<00:00, 23407.06it/s]\n",
      "100%|██████████| 4272627/4272627 [02:57<00:00, 24138.81it/s]\n",
      "100%|██████████| 4272627/4272627 [02:56<00:00, 24214.54it/s]\n",
      "100%|██████████| 4272627/4272627 [02:54<00:00, 24419.84it/s]\n",
      "100%|██████████| 4272627/4272627 [02:55<00:00, 24315.06it/s]\n",
      "100%|██████████| 4272627/4272627 [03:00<00:00, 23713.46it/s]\n",
      "100%|██████████| 4272627/4272627 [02:55<00:00, 24276.69it/s]\n"
     ]
    }
   ],
   "source": [
    "# GridSearch\n",
    "min_ = float('inf')\n",
    "minI = 0\n",
    "for i in np.logspace(0,-8,10, endpoint = True):\n",
    "    X,y = generate_features_labels(l2_snapshot, ask='a1', bid = 'b1', alpha=i)\n",
    "    y_ = pd.Series(y)\n",
    "    temp = y_.value_counts()\n",
    "    a,b,c = temp[0], temp[1], temp[2]\n",
    "    sum_ = a+b+c\n",
    "    l2 = ((1/3-(a/sum_))*100)**2 + ((1/3-(b/sum_))*100)**2 + ((1/3-(c/sum_))*100)**2\n",
    "    if l2 < min_:\n",
    "        min_ = l2\n",
    "        minI = i\n",
    "    del X,y,temp"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "7.742636826811277e-08 347.51453285264193\n"
     ]
    }
   ],
   "source": [
    "print(minI, min_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 4272627/4272627 [02:55<00:00, 24396.68it/s]\n"
     ]
    }
   ],
   "source": [
    "X,y = generate_features_labels(l2_snapshot_, ask='a1', bid = 'b1', alpha=minI)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "y = y.astype('int')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f982a437d30>"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.countplot(x = y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "T = 100\n",
    "D = 40\n",
    "N = len(X) - T"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "trainPart = 3176046"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train = np.zeros((trainPart, T, D), dtype = 'float16')\n",
    "y_train = np.zeros(trainPart, dtype = 'float16')\n",
    "\n",
    "#Preparing the time series data using timestep of 100 and no of features = 200\n",
    "for t in range(trainPart):\n",
    "    X_train[t, :, :] = X[t:t+T]\n",
    "    y_train[t] = y[t+T] \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X_train - Before: 23.663386702537537 GB float16\n"
     ]
    }
   ],
   "source": [
    "print('X_train - Before: {} GB'.format(X_train.nbytes/1024**3), X_train.dtype)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "y_train - Before: 0.005915846675634384 GB float16\n"
     ]
    }
   ],
   "source": [
    "print('y_train - Before: {} GB'.format(y_train.nbytes/1024**3), y_train.dtype)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X_test - Before: 8.169420063495636 GB float16\n",
      "y_test - Before: 0.002042355015873909 GB float16\n"
     ]
    }
   ],
   "source": [
    "X_test = np.zeros((N - trainPart, T, D), dtype='float16')\n",
    "y_test = np.zeros(N - trainPart, dtype='float16')\n",
    "\n",
    "#Preparing the time series data using timestep of 100 and no of features = 40\n",
    "for k in range(N - trainPart):\n",
    "    t = k + trainPart\n",
    "    X_test[k, :, :] = X[t:t+T]\n",
    "    y_test[k] = y[t+T]\n",
    "\n",
    "print('X_test - Before: {} GB'.format(X_test.nbytes/1024**3), X_test.dtype)\n",
    "print('y_test - Before: {} GB'.format(y_test.nbytes/1024**3), y_test.dtype)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "2.1.0\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "print(tf.__version__)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "from tensorflow.keras.layers import Input, Conv1D, Conv2D, Flatten, MaxPooling1D, MaxPooling2D, Dense, LeakyReLU, Bidirectional\n",
    "from tensorflow.keras.models import Model\n",
    "from tensorflow.keras.optimizers import SGD, Adam\n",
    "from tensorflow.compat.v1.keras.layers import CuDNNLSTM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "X_train = np.expand_dims(X_train, -1)\n",
    "X_test = np.expand_dims(X_test, -1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "i = Input(shape=X_train[0].shape)\n",
    "x = Conv2D(16, (4,D), activation=LeakyReLU(alpha=0.01))(i)\n",
    "x = tf.keras.layers.Reshape(target_shape=(T-3,16))(x)\n",
    "x = Conv1D(16, 4, activation=LeakyReLU(alpha=0.01))(x)\n",
    "x = tf.keras.layers.BatchNormalization()(x)\n",
    "x = MaxPooling1D(2)(x)\n",
    "x = Conv1D(32, 3, activation=LeakyReLU(alpha=0.01))(x)\n",
    "x = tf.keras.layers.BatchNormalization()(x)\n",
    "x = MaxPooling1D(2)(x)\n",
    "x = Bidirectional(CuDNNLSTM(64, return_sequences = False))(x)\n",
    "x = Dense(32, activation=LeakyReLU(alpha=0.01))(x)\n",
    "x = Dense(32, activation=LeakyReLU(alpha=0.01))(x)\n",
    "x = Dense(3, activation='softmax')(x)\n",
    "\n",
    "model = Model(i,x)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_1 (InputLayer)         [(None, 100, 40, 1)]      0         \n",
      "_________________________________________________________________\n",
      "conv2d (Conv2D)              (None, 97, 1, 16)         2576      \n",
      "_________________________________________________________________\n",
      "reshape (Reshape)            (None, 97, 16)            0         \n",
      "_________________________________________________________________\n",
      "conv1d (Conv1D)              (None, 94, 16)            1040      \n",
      "_________________________________________________________________\n",
      "batch_normalization (BatchNo (None, 94, 16)            64        \n",
      "_________________________________________________________________\n",
      "max_pooling1d (MaxPooling1D) (None, 47, 16)            0         \n",
      "_________________________________________________________________\n",
      "conv1d_1 (Conv1D)            (None, 45, 32)            1568      \n",
      "_________________________________________________________________\n",
      "batch_normalization_1 (Batch (None, 45, 32)            128       \n",
      "_________________________________________________________________\n",
      "max_pooling1d_1 (MaxPooling1 (None, 22, 32)            0         \n",
      "_________________________________________________________________\n",
      "bidirectional (Bidirectional (None, 128)               50176     \n",
      "_________________________________________________________________\n",
      "dense (Dense)                (None, 32)                4128      \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 32)                1056      \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 3)                 99        \n",
      "=================================================================\n",
      "Total params: 60,835\n",
      "Trainable params: 60,739\n",
      "Non-trainable params: 96\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "model.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.compile(optimizer='adam',\n",
    "              loss='sparse_categorical_crossentropy',\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 3176046 samples, validate on 1096481 samples\n",
      "Epoch 1/4\n",
      "3176046/3176046 [==============================] - 1623s 511us/sample - loss: 0.4721 - accuracy: 0.8082 - val_loss: 0.4457 - val_accuracy: 0.8195\n",
      "Epoch 2/4\n",
      "3176046/3176046 [==============================] - 1616s 509us/sample - loss: 0.4309 - accuracy: 0.8272 - val_loss: 0.4443 - val_accuracy: 0.8223\n",
      "Epoch 3/4\n",
      "3176046/3176046 [==============================] - 1624s 511us/sample - loss: 0.4118 - accuracy: 0.8345 - val_loss: 0.4485 - val_accuracy: 0.8240\n",
      "Epoch 4/4\n",
      "3176046/3176046 [==============================] - 1621s 511us/sample - loss: 0.3971 - accuracy: 0.8399 - val_loss: 0.4598 - val_accuracy: 0.8224\n"
     ]
    }
   ],
   "source": [
    "EPOCHS = 4\n",
    "BATCH_SIZE = 32\n",
    "r = model.fit(\n",
    "  X_train, y_train,\n",
    "  batch_size=BATCH_SIZE,\n",
    "  epochs=EPOCHS,\n",
    "  validation_data=(X_test, y_test),\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7f977e25f940>"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.lineplot(x=range(1,EPOCHS+1),y=r.history['loss'])\n",
    "sns.lineplot(x=range(1,EPOCHS+1),y=r.history['val_loss'])\n",
    "plt.title('Model Cross Entropy Loss')\n",
    "plt.ylabel('Cross Entropy Loss')\n",
    "plt.xlabel('Epochs')\n",
    "plt.legend(['Training', 'Testing'], loc='upper left')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "y_pred = model.predict(X_test).argmax(axis=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.metrics import confusion_matrix\n",
    "\n",
    "cm = confusion_matrix(y_pred=y_pred, y_true=y_test)\n",
    "    \n",
    "cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]\n",
    "    \n",
    "ax=sns.heatmap(cm, annot=True, xticklabels=[0,1,2], yticklabels=[0,1,2], cmap='Blues')\n",
    "ax.set_ylim(3.0, 0)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "             precision    recall  f1-score   support\n",
      "\n",
      "        0.0       0.85      0.87      0.86    566555\n",
      "        1.0       0.81      0.95      0.87    304953\n",
      "        2.0       0.73      0.53      0.62    224973\n",
      "\n",
      "avg / total       0.82      0.82      0.82   1096481\n",
      "\n"
     ]
    }
   ],
   "source": [
    "print(classification_report(y_test,y_pred))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}