harml_rnn_lstm_raw.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import pickle\n",
    "import pprint\n",
    "from scipy.stats import kurtosis, skew\n",
    "from keras.models import Sequential\n",
    "from keras.layers import LSTM, Dense, Dropout, Activation, Flatten\n",
    "from keras.utils import to_categorical\n",
    "from keras.optimizers import Adam\n",
    "from keras.models import load_model\n",
    "from sklearn import metrics\n",
    "import matplotlib.pyplot as plt\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "INPUT_DATA = [\n",
    "    \"body_acc_x_\",\n",
    "    \"body_acc_y_\",\n",
    "    \"body_acc_z_\",\n",
    "    \"body_gyro_x_\",\n",
    "    \"body_gyro_y_\",\n",
    "    \"body_gyro_z_\",\n",
    "    \"total_acc_x_\",\n",
    "    \"total_acc_y_\",\n",
    "    \"total_acc_z_\"\n",
    "]\n",
    "\n",
    "DATA_FILES = [\n",
    "    \"neutral_hazmei_2mins\",\n",
    "    \"jumping_hazmei_1mins\",\n",
    "    \"wave_hands_hazmei_1min\"\n",
    "]\n",
    "\n",
    "LABELS = [\n",
    "    \"HAZMEI'S SEXY POSE\", \n",
    "    \"HAZMEI'S BELLY BOUNCE\", \n",
    "    \"HAZMEI'S FLYING KISS\"\n",
    "] \n",
    "\n",
    "TRAINING_SIZE = 1200\n",
    "TESTING_SIZE = 300\n",
    "WINDOW_SIZE = 1 #1 second\n",
    "WINDOW_READINGS = WINDOW_SIZE * 50 #50Hz"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def normalize_data(data):\n",
    "    data_norm = (data - data.mean()) / (data.max() - data.min())\n",
    "    return data_norm\n",
    "\n",
    "def feature_extraction(x, y, z):\n",
    "    features = [np.mean(x), np.mean(y), np.mean(z), np.std(x), np.std(y), np.std(z)]\n",
    "    #Median Absolute Deviation\n",
    "    features.extend((np.mean(abs(x - features[0])), np.mean(abs(y - features[1])), np.mean(abs(z - features[2]))))\n",
    "    #Jerk Signals\n",
    "    features.extend((np.mean(np.diff(x)), np.mean(np.diff(y)), np.mean(np.diff(z)), np.std(np.diff(x)), np.std(np.diff(y)), np.std(np.diff(z))))\n",
    "    features.extend((np.mean(abs(np.diff(x) - features[9])), np.mean(abs(np.diff(y) - features[10])), np.mean(abs(np.diff(y) - features[11]))))\n",
    "    features.extend((skew(x), skew(y), skew(z), kurtosis(x), kurtosis(y), kurtosis(z)))\n",
    "    features.extend((max(x), max(y), max(z), min(x), min(y), min(z)))\n",
    "    return features\n",
    "\n",
    "def add_noise(data):\n",
    "    data_noise = data + np.random.uniform(size=len(data))\n",
    "    data_noise = data_noise + np.random.laplace(loc=0.0, scale=1.0, size=len(data))\n",
    "    return data_noise\n",
    "\n",
    "def data_augmentation(X):\n",
    "    X_noise = X\n",
    "    for i in range(X.shape[0]):\n",
    "        for j in range(X.shape[2]):\n",
    "            X_noise[i][:][j] = add_noise(X_noise[i][:][j])\n",
    "    return np.concatenate((X, X_noise), axis=0)\n",
    "\n",
    "def feature_selection(X, augmentData=False):\n",
    "    data = []\n",
    "    for i in range(X.shape[0]):\n",
    "        features = []\n",
    "        for j in range(0, X.shape[2], 3):\n",
    "            x = [X[i][u][j] for u in range(X.shape[1])]\n",
    "            y = [X[i][u][j+1] for u in range(X.shape[1])]\n",
    "            z = [X[i][u][j+2] for u in range(X.shape[1])]\n",
    "            \n",
    "            if augmentData:\n",
    "                x_noise = add_noise(x)\n",
    "                y_noise = add_noise(y)\n",
    "                z_noise = add_noise(z)\n",
    "                features.append(feature_extraction(x_noise, y_noise, z_noise))\n",
    "            else:\n",
    "                features.append(feature_extraction(x, y, z))\n",
    "\n",
    "        data.append(features)\n",
    "    return np.array(data)\n",
    "\n",
    "def feature_engineering(X, augmentData=False):\n",
    "    if augmentData:\n",
    "        return np.concatenate((feature_selection(X, False), feature_selection(X, True)), axis=0)\n",
    "    else:\n",
    "        return feature_selection(X, False)\n",
    "    \n",
    "def load_X(X_signals_paths):\n",
    "    X_signals = []\n",
    "    \n",
    "    for signal_type_path in X_signals_paths:\n",
    "        file = open(signal_type_path, 'r')\n",
    "        # Read dataset from disk, dealing with text files' syntax\n",
    "        X_signals.append(\n",
    "            [np.array(serie, dtype=np.float32) for serie in \n",
    "             [row.replace('  ', ' ').strip().split(' ') for row in file]\n",
    "            ]\n",
    "        )\n",
    "        file.close()\n",
    "    \n",
    "    return np.transpose(np.array(X_signals), (1, 2, 0))\n",
    "\n",
    "\n",
    "def load_y(y_path):\n",
    "    file = open(y_path, 'r')\n",
    "    # Read dataset from disk, dealing with text file's syntax\n",
    "    y_ = np.array(\n",
    "        [elem for elem in [\n",
    "            row.replace('  ', ' ').strip().split(' ') for row in file\n",
    "        ]], \n",
    "        dtype=np.int32\n",
    "    )\n",
    "    file.close()\n",
    "    return y_ - 1\n",
    "\n",
    "def read_data(X_input_file_paths):\n",
    "    X = np.zeros([1,WINDOW_READINGS,9])\n",
    "    Xt = np.zeros([1,WINDOW_READINGS,9])\n",
    "    y = []\n",
    "    yt = []\n",
    "    count = 0\n",
    "    for file in X_input_file_paths:\n",
    "        raw_data = pd.read_csv(file + \".txt\", header=None, delim_whitespace=True)\n",
    "        data = raw_data.loc[:, raw_data.columns != 0]\n",
    "        data = data.values.reshape(-1, 9)\n",
    "        data = data[:data.shape[0]-(data.shape[0]%WINDOW_READINGS)]\n",
    "        normalized_data = normalize_data(data)\n",
    "        sampled_data = normalized_data.reshape(-1, WINDOW_READINGS, 9)\n",
    "        print(normalized_data.shape)\n",
    "        print(sampled_data.shape)\n",
    "        X = np.concatenate((X, sampled_data[:sampled_data.shape[0]-10]), axis=0)\n",
    "        Xt = np.concatenate((Xt, sampled_data[sampled_data.shape[0]-10:]), axis=0)\n",
    "        y = y + [count]*(sampled_data.shape[0]-10)\n",
    "        yt = yt + [count]*(sampled_data.shape[0]-(sampled_data.shape[0]-10))\n",
    "        np.savetxt('data/data_hazmei/y.txt', y, fmt='%5s', delimiter=\" \")\n",
    "        np.savetxt(file + \"_formatted.csv\", data[0:], delimiter=\",\")\n",
    "        count = count + 1\n",
    "    y = np.array(y)\n",
    "    yt = np.array(yt)\n",
    "    return X[1:], y, Xt[1:], yt\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(5400, 9)\n",
      "(108, 50, 9)\n",
      "(2700, 9)\n",
      "(54, 50, 9)\n",
      "(2800, 9)\n",
      "(56, 50, 9)\n",
      "(188, 50, 9)\n",
      "(188,)\n",
      "(30, 50, 9)\n",
      "(30,)\n"
     ]
    }
   ],
   "source": [
    "X_input_file_paths = ['data/data_hazmei/' + filename for filename in DATA_FILES]\n",
    "X, Y, Xt, Yt = read_data(X_input_file_paths)\n",
    "print(X.shape)\n",
    "print(Y.shape)\n",
    "print(Xt.shape)\n",
    "print(Yt.shape)\n",
    "\n",
    "#raw_data = pd.read_csv('data/data_hazmei/jumping_hazmei_1mins.txt', header=None, delim_whitespace=True)\n",
    "#print(raw_data.shape)\n",
    "#print(raw_data[0:])\n",
    "#pure_data = raw_data.loc[:, raw_data.columns != 0]\n",
    "#print(pure_data.shape)\n",
    "#print(pure_data[0:])\n",
    "#data = pure_data.values.reshape(-1, 9)\n",
    "#data = data[:data.shape[0]-(data.shape[0]%50)]\n",
    "#data = normalize_data(data)\n",
    "#print(data.shape)\n",
    "#print(data[0:])\n",
    "#cropped_data = data.reshape(-1, 50, 9)\n",
    "#print(cropped_data.shape)\n",
    "#print(cropped_data)\n",
    "#np.savetxt(\"data/data_hazmei/formatted_data.csv\", data[0:], delimiter=\",\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'\\nX_train_input_paths = [\"data/train/\" + signal + \"train.txt\" for signal in INPUT_DATA]\\nX_test_input_paths = [\"data/test/\" + signal + \"test.txt\" for signal in INPUT_DATA]\\nX_train = load_X(X_train_input_paths)\\nX_test = load_X(X_test_input_paths)\\n\\ny_train_path = \"data/y_train.txt\"\\ny_test_path = \"data/y_test.txt\"\\nY_train = load_y(y_train_path).ravel()\\nY_test = load_y(y_test_path).ravel()\\n\\nprint(X_train.shape)\\nprint(Y_train.shape)\\nprint(X_test.shape)\\nprint(Y_test.shape)\\n\\nprint(X_train[0])\\nprint(Y_train[0])\\n\\nX_train = X_train[0:TRAINING_SIZE]\\nY_train = Y_train[0:TRAINING_SIZE]\\nX_test = X_test[0:TESTING_SIZE]\\nY_test = Y_test[0:TESTING_SIZE]\\n\\nprint(X_train.shape)\\nprint(Y_train.shape)\\nprint(X_test.shape)\\nprint(Y_test.shape)\\n'"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "'''\n",
    "X_train_input_paths = [\"data/train/\" + signal + \"train.txt\" for signal in INPUT_DATA]\n",
    "X_test_input_paths = [\"data/test/\" + signal + \"test.txt\" for signal in INPUT_DATA]\n",
    "X_train = load_X(X_train_input_paths)\n",
    "X_test = load_X(X_test_input_paths)\n",
    "\n",
    "y_train_path = \"data/y_train.txt\"\n",
    "y_test_path = \"data/y_test.txt\"\n",
    "Y_train = load_y(y_train_path).ravel()\n",
    "Y_test = load_y(y_test_path).ravel()\n",
    "\n",
    "print(X_train.shape)\n",
    "print(Y_train.shape)\n",
    "print(X_test.shape)\n",
    "print(Y_test.shape)\n",
    "\n",
    "print(X_train[0])\n",
    "print(Y_train[0])\n",
    "\n",
    "X_train = X_train[0:TRAINING_SIZE]\n",
    "Y_train = Y_train[0:TRAINING_SIZE]\n",
    "X_test = X_test[0:TESTING_SIZE]\n",
    "Y_test = Y_test[0:TESTING_SIZE]\n",
    "\n",
    "print(X_train.shape)\n",
    "print(Y_train.shape)\n",
    "print(X_test.shape)\n",
    "print(Y_test.shape)\n",
    "'''"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(188, 50, 9)\n",
      "(30, 50, 9)\n",
      "(188, 3)\n"
     ]
    }
   ],
   "source": [
    "#Raw Data with Data Augmentation\n",
    "'''\n",
    "X = data_augmentation(X)\n",
    "Xt = data_augmentation(Xt)\n",
    "Y = np.concatenate((Y, Y), axis=0)\n",
    "Yt = np.concatenate((Yt, Yt), axis=0)\n",
    "'''\n",
    "\n",
    "#Feature Selection & Data Augmentation\n",
    "'''\n",
    "X = feature_engineering(X, False)\n",
    "Xt = feature_engineering(Xt, False)\n",
    "#Y = np.concatenate((Y, Y), axis=0)\n",
    "#Yt = np.concatenate((Yt, Yt), axis=0)\n",
    "'''\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "Yhot = to_categorical(Y)\n",
    "\n",
    "print(X.shape)\n",
    "print(Xt.shape)\n",
    "print(Yhot.shape)\n",
    "#print(X[0])\n",
    "\n",
    "np.savetxt(\"data/test_data_format.csv\", Xt[0], delimiter=\",\")\n",
    "\n",
    "#X = X.reshape(X.shape[0], X.shape[1]*X.shape[2])\n",
    "#Xt = Xt.reshape(Xt.shape[0], Xt.shape[1]*Xt.shape[2])\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/30\n",
      "188/188 [==============================] - 1s - loss: 1.0959 - acc: 0.4202     \n",
      "Epoch 2/30\n",
      "188/188 [==============================] - 1s - loss: 1.0725 - acc: 0.5213     \n",
      "Epoch 3/30\n",
      "188/188 [==============================] - 1s - loss: 1.0332 - acc: 0.5213     \n",
      "Epoch 4/30\n",
      "188/188 [==============================] - 1s - loss: 0.9898 - acc: 0.5213     \n",
      "Epoch 5/30\n",
      "188/188 [==============================] - 1s - loss: 0.9749 - acc: 0.5213     \n",
      "Epoch 6/30\n",
      "188/188 [==============================] - 1s - loss: 0.9375 - acc: 0.5213     \n",
      "Epoch 7/30\n",
      "188/188 [==============================] - 1s - loss: 0.8829 - acc: 0.5266     \n",
      "Epoch 8/30\n",
      "188/188 [==============================] - 1s - loss: 0.8082 - acc: 0.5904     \n",
      "Epoch 9/30\n",
      "188/188 [==============================] - 1s - loss: 0.6769 - acc: 0.7287     \n",
      "Epoch 10/30\n",
      "188/188 [==============================] - 1s - loss: 0.4843 - acc: 0.8777     \n",
      "Epoch 11/30\n",
      "188/188 [==============================] - 1s - loss: 0.2761 - acc: 0.9468     \n",
      "Epoch 12/30\n",
      "188/188 [==============================] - 1s - loss: 0.1455 - acc: 0.9734     \n",
      "Epoch 13/30\n",
      "188/188 [==============================] - 1s - loss: 0.0848 - acc: 0.9734     \n",
      "Epoch 14/30\n",
      "188/188 [==============================] - 1s - loss: 0.0658 - acc: 0.9787     \n",
      "Epoch 15/30\n",
      "188/188 [==============================] - 1s - loss: 0.0531 - acc: 0.9681     \n",
      "Epoch 16/30\n",
      "188/188 [==============================] - 1s - loss: 0.0485 - acc: 0.9840     \n",
      "Epoch 17/30\n",
      "188/188 [==============================] - 1s - loss: 0.0494 - acc: 0.9787     \n",
      "Epoch 18/30\n",
      "188/188 [==============================] - 1s - loss: 0.0358 - acc: 0.9840     \n",
      "Epoch 19/30\n",
      "188/188 [==============================] - 1s - loss: 0.0381 - acc: 0.9894     \n",
      "Epoch 20/30\n",
      "188/188 [==============================] - 1s - loss: 0.0328 - acc: 0.9894     \n",
      "Epoch 21/30\n",
      "188/188 [==============================] - 1s - loss: 0.0279 - acc: 0.9894     \n",
      "Epoch 22/30\n",
      "188/188 [==============================] - 1s - loss: 0.0266 - acc: 0.9894     \n",
      "Epoch 23/30\n",
      "188/188 [==============================] - 1s - loss: 0.0218 - acc: 0.9947     \n",
      "Epoch 24/30\n",
      "188/188 [==============================] - 1s - loss: 0.0216 - acc: 0.9947     \n",
      "Epoch 25/30\n",
      "188/188 [==============================] - 1s - loss: 0.0212 - acc: 0.9894     \n",
      "Epoch 26/30\n",
      "188/188 [==============================] - 1s - loss: 0.0174 - acc: 0.9947     \n",
      "Epoch 27/30\n",
      "188/188 [==============================] - 1s - loss: 0.0168 - acc: 0.9947     \n",
      "Epoch 28/30\n",
      "188/188 [==============================] - 1s - loss: 0.0166 - acc: 0.9947     \n",
      "Epoch 29/30\n",
      "188/188 [==============================] - 1s - loss: 0.0110 - acc: 1.0000     \n",
      "Epoch 30/30\n",
      "188/188 [==============================] - 1s - loss: 0.0116 - acc: 0.9894     \n",
      "160/188 [========================>.....] - ETA: 0s\n",
      "loss: 0.01%\n",
      "\n",
      "acc: 100.00%\n"
     ]
    }
   ],
   "source": [
    "#Neural Network\n",
    "\n",
    "data_dim = X.shape[2]\n",
    "timesteps = X.shape[1]\n",
    "num_classes = Yhot.shape[1]\n",
    "b_size = 32\n",
    "\n",
    "model = Sequential()\n",
    "model.add(LSTM(16, return_sequences=True, input_shape=(timesteps, data_dim)))\n",
    "model.add(LSTM(16, return_sequences=True))\n",
    "#model.add(LSTM(16, return_sequences=True))\n",
    "model.add(Flatten())\n",
    "model.add(Dense(num_classes, activation='softmax'))\n",
    "\n",
    "\n",
    "model.compile(loss='categorical_crossentropy',\n",
    "              optimizer='adam',\n",
    "              metrics=['accuracy'])\n",
    "\n",
    "model.fit(X, Yhot, epochs=30, batch_size=b_size, validation_split=0.0, shuffle=True)\n",
    "result = model.evaluate(X, Yhot)\n",
    "print(\"\\n%s: %.2f%%\" % (model.metrics_names[0], result[0]))\n",
    "print(\"\\n%s: %.2f%%\" % (model.metrics_names[1], result[1]*100))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[0 0 0 0 0 0 0 0 0 0 1 1 1 1 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 2]\n",
      "Accuracy Rate:\n",
      "1.0\n",
      "[[10  0  0]\n",
      " [ 0 10  0]\n",
      " [ 0  0 10]]\n",
      "Precision: 100.0%\n",
      "Recall: 100.0%\n",
      "f1_score: 100.0%\n",
      "\n",
      "Confusion Matrix:\n",
      "[[10  0  0]\n",
      " [ 0 10  0]\n",
      " [ 0  0 10]]\n",
      "\n",
      "Confusion matrix (normalised to % of total test data):\n",
      "[[ 33.33333588   0.           0.        ]\n",
      " [  0.          33.33333588   0.        ]\n",
      " [  0.           0.          33.33333588]]\n",
      "Note: training and testing data is not equally distributed amongst classes, \n",
      "so it is normal that more than a 6th of the data is correctly classifier in the last category.\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAzYAAAM+CAYAAADIHJQsAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nSZIkqfcsbCRJkiT13v8HnL7KPxqIWWsAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x129935f28>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "Y_output = model.predict(Xt)\n",
    "Y_pred = np.argmax(Y_output, axis=1)\n",
    "print(np.array(Y_pred))\n",
    "print(\"Accuracy Rate:\")\n",
    "print(metrics.accuracy_score(Yt, Y_pred))\n",
    "print(metrics.confusion_matrix(Yt, Y_pred))\n",
    "\n",
    "print(\"Precision: {}%\".format(100*metrics.precision_score(Yt, Y_pred, average=\"weighted\")))\n",
    "print(\"Recall: {}%\".format(100*metrics.recall_score(Yt, Y_pred, average=\"weighted\")))\n",
    "print(\"f1_score: {}%\".format(100*metrics.f1_score(Yt, Y_pred, average=\"weighted\")))\n",
    "\n",
    "print(\"\")\n",
    "print(\"Confusion Matrix:\")\n",
    "confusion_matrix = metrics.confusion_matrix(Yt, Y_pred)\n",
    "print(confusion_matrix)\n",
    "normalised_confusion_matrix = np.array(confusion_matrix, dtype=np.float32)/np.sum(confusion_matrix)*100\n",
    "\n",
    "print(\"\")\n",
    "print(\"Confusion matrix (normalised to % of total test data):\")\n",
    "print(normalised_confusion_matrix)\n",
    "print(\"Note: training and testing data is not equally distributed amongst classes, \")\n",
    "print(\"so it is normal that more than a 6th of the data is correctly classifier in the last category.\")\n",
    "\n",
    "# Plot Results: \n",
    "width = 12\n",
    "height = 12\n",
    "plt.figure(figsize=(width, height))\n",
    "plt.imshow(\n",
    "    normalised_confusion_matrix, \n",
    "    interpolation='nearest', \n",
    "    cmap=plt.cm.rainbow\n",
    ")\n",
    "plt.title(\"Confusion matrix \\n(normalised to % of total test data)\")\n",
    "plt.colorbar()\n",
    "tick_marks = np.arange(len(LABELS))\n",
    "plt.xticks(tick_marks, LABELS, rotation=90)\n",
    "plt.yticks(tick_marks, LABELS)\n",
    "plt.tight_layout()\n",
    "plt.ylabel('True label')\n",
    "plt.xlabel('Predicted label')\n",
    "plt.show()\n",
    "\n",
    "np.savetxt(\"data/accuracy.csv\", [metrics.accuracy_score(Yt, Y_pred)], delimiter=\",\")\n",
    "np.savetxt(\"data/confusion_matrix.csv\", metrics.confusion_matrix(Yt, Y_pred), delimiter=\",\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "model.save('data/har_rnn_lstm.h5')  # creates a HDF5 file 'my_model.h5'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(120, 9)\n",
      "(3, 40, 9)\n",
      "[[[  3.30674392e-03   4.08453215e-03   3.11877709e-02 ...,   1.01641703e+00\n",
      "    -1.25868201e-01   1.02473199e-01]\n",
      "  [  7.57194078e-03   5.11844782e-03   2.35671904e-02 ...,   1.02074504e+00\n",
      "    -1.24368198e-01   9.75658968e-02]\n",
      "  [  5.40711684e-03   5.89226792e-03   2.10004691e-02 ...,   1.01864302e+00\n",
      "    -1.23138197e-01   9.76466537e-02]\n",
      "  ..., \n",
      "  [  2.28138897e-03  -4.51477198e-03  -2.35145297e-02 ...,   1.01842904e+00\n",
      "    -1.24303102e-01   1.02193497e-01]\n",
      "  [  3.65372794e-03  -5.92592312e-03  -1.84824299e-02 ...,   1.01989496e+00\n",
      "    -1.25629902e-01   1.07519001e-01]\n",
      "  [  2.31263903e-03  -5.06280502e-03  -1.73001904e-02 ...,   1.01864696e+00\n",
      "    -1.24689601e-01   1.08948000e-01]]\n",
      "\n",
      " [[ -4.04219390e-05  -2.24447693e-03  -1.92251801e-02 ...,   1.01638699e+00\n",
      "    -1.21801399e-01   1.07225001e-01]\n",
      "  [  5.34225372e-04  -7.86087825e-04  -2.20926702e-02 ...,   1.01705301e+00\n",
      "    -1.20280102e-01   1.04516402e-01]\n",
      "  [  2.96326308e-03  -1.21828204e-03  -2.44254097e-02 ...,   1.01957202e+00\n",
      "    -1.20656200e-01   1.02301002e-01]\n",
      "  ..., \n",
      "  [ -1.17679394e-03  -2.98620295e-03  -1.44583303e-02 ...,   1.01778698e+00\n",
      "    -1.23681098e-01   1.00359797e-01]\n",
      "  [  1.64613593e-04  -4.76858579e-03  -1.50742196e-02 ...,   1.01916695e+00\n",
      "    -1.25553101e-01   9.92000178e-02]\n",
      "  [  7.50828010e-04  -3.32420901e-03  -1.20167900e-02 ...,   1.01978898e+00\n",
      "    -1.24198698e-01   1.01714201e-01]]\n",
      "\n",
      " [[  3.88662796e-04  -1.19253702e-03  -1.17171500e-02 ...,   1.01946199e+00\n",
      "    -1.22157499e-01   1.01471998e-01]\n",
      "  [  1.32751698e-03  -2.34180293e-03  -1.14236400e-02 ...,   1.02043295e+00\n",
      "    -1.23397402e-01   1.01225898e-01]\n",
      "  [  2.05252506e-03  -6.11669291e-03  -9.45955236e-03 ...,   1.02118897e+00\n",
      "    -1.27262995e-01   1.02653198e-01]\n",
      "  ..., \n",
      "  [ -1.14748406e-03   1.71443899e-04   2.64786393e-03 ...,   1.01844501e+00\n",
      "    -1.24069601e-01   1.00385197e-01]\n",
      "  [ -2.22265502e-04   1.57418102e-03   2.38105701e-03 ...,   1.01937199e+00\n",
      "    -1.22745097e-01   9.98735502e-02]\n",
      "  [  1.57550001e-03   3.07018892e-03  -2.26975698e-03 ...,   1.02117097e+00\n",
      "    -1.21325999e-01   9.49874073e-02]]]\n"
     ]
    }
   ],
   "source": [
    "raw_data = pd.read_csv('data/sample_data_format.csv', skiprows=range(0, 7))\n",
    "print(raw_data.shape)\n",
    "cropped_data = raw_data.values.reshape(-1, 40, 9)\n",
    "print(cropped_data.shape)\n",
    "print(cropped_data)\n",
    "pickle.dump(cropped_data, open('data/cropped_data_format.pkl', 'wb'))\n",
    "np.savetxt(\"data/cropped_data_format_2.csv\", cropped_data[2], delimiter=\",\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "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": "3.6.1"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}