01 1D CNN - Keras.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Introduction\n",
    "In this study, we will start by studying the applicability of using the raw data from the seismometer as an input to the neural network for recognition and prediction of earthquakes. The raw data is a time series data of 100 Hz frequency which discribes the vertical motion of the earth. The figure below show a simple illustration of how it is measured. Note that the current tools rely on a more sophisticated and accurate tools but the concept is still the same.\n",
    "\n",
    "As mentioned above, two problems are studied described in the figure above. Obviously, the second problem is more complicated as there is no known method to do this. In fact, it is not known that the signal contain information for the prediction. The two problems are:\n",
    "1. The recognition of the earthquakes signal after they occur.\n",
    "2. The prediction of earthquakes signal before they occur.\n",
    "\n",
    "<img src=\"Figures/For Notbooks/Seismometer.png\" width=\"700\">"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Dataset\n",
    "The dataset was harvested from servers specifically for this study. After compiling numerous datasets with varying [minimum earthquake magnitude](https://en.wikipedia.org/wiki/Richter_magnitude_scale), lateral coverage, and number of stations, this study area ([The Geysers](https://en.wikipedia.org/wiki/The_Geysers)) of the  is specifically chosen because:\n",
    "1. It contains a spatially dense seismometers coverage.\n",
    "2. The seismicmometers have been operating for a relatively long time.\n",
    "3. The area is tectonically active because of the existance of the geysers.\n",
    "4. Enhanced geothermal system are operating in the area creating more earthquakes.\n",
    "\n",
    "<img src=\"Figures/For Notbooks/Map.png\" width=\"600\">\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Package and helper functions\n",
    "We will use [Keras](https://keras.io/) to build the neural network, [Matplotlib](https://matplotlib.org/) for plotting results if needed, and [Numpy](http://www.numpy.org/) for matrix manipulation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Basic packages\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "# Keras\n",
    "from keras.callbacks import ModelCheckpoint\n",
    "from keras.models import Model, load_model, Sequential\n",
    "from keras.layers import Dense, Activation, Dropout, Input, Masking, TimeDistributed, LSTM, Conv1D\n",
    "from keras.layers import GRU, Bidirectional, BatchNormalization, Reshape, Flatten, MaxPooling1D\n",
    "from keras.optimizers import Adam\n",
    "\n",
    "# Sklearn\n",
    "from sklearn.metrics import confusion_matrix"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We will use this cell to create some helper functions that will be used in the analysis. We decided to have all the functions needed in one script to facilitate readibility and understanding."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "def splitDataset1D(X, Y, proportions=np.array([.8, .2])):\n",
    "    \"\"\"\n",
    "    Splits the dataset into a number of smaller datasets with certain given ratios.\n",
    "    \n",
    "    Argument:\n",
    "    X -- Input dataset\n",
    "    Y -- Labels\n",
    "\n",
    "    Returns:\n",
    "    Xs -- Splitted dataset (dictionary with numbers as keys)\n",
    "    Ys -- Splitted labels  (dictionary with numbers as keys)\n",
    "    \"\"\"\n",
    "    \n",
    "    # Make sure proportions are a numpy array\n",
    "    proportions = np.array(proportions)\n",
    "    \n",
    "    # Handle an option to provide one ratio\n",
    "    if (np.sum(proportions)!=1):\n",
    "        ratios = np.concatenate((proportions, np.array([1-np.sum(proportions)])))\n",
    "\n",
    "    # Create the random indices\n",
    "    m = X.shape[1] \n",
    "    randomIndecies = np.random.permutation(m)\n",
    "    \n",
    "    # Calculate the number of boundaries of indecies in each set\n",
    "    nSamples = np.round(np.array(proportions) * m)\n",
    "    nSamples = np.concatenate((np.array([0]), np.round(proportions * m)))\n",
    "    nSamples = np.cumsum(nSamples)\n",
    "    nSamples = nSamples.astype(int)\n",
    "    \n",
    "    # Extract teh sets\n",
    "    Xs = []\n",
    "    Ys = []\n",
    "    for i in range(0, proportions.size):\n",
    "        chosenSamples = randomIndecies[nSamples[i]:nSamples[i+1]]\n",
    "        Xs.append(X[:, chosenSamples, :])\n",
    "        Ys.append(Y[chosenSamples, :])\n",
    "        \n",
    "    return Xs, Ys\n",
    "\n",
    "# ======================================================================\n",
    "def normalizeData(data, axis=None, keepdims=True):\n",
    "    \"\"\"\n",
    "    Computes the zscore of data along a certain axis.\n",
    "    \n",
    "    Argument:\n",
    "    data -- Data to be normalized\n",
    "    axis -- Axis to normalized (if no axis is given, we\n",
    "\n",
    "    Returns:\n",
    "    normalizedData -- normalized dataset\n",
    "    Ys -- Splitted labels  (dictionary with numbers as keys)\n",
    "    \"\"\"\n",
    "    \n",
    "    # Zscore normalization\n",
    "    normalizedData = (data - np.mean(data, axis = axis, keepdims= keepdims)) / np.std(data, axis = axis, keepdims = keepdims)\n",
    "    return normalizedData\n",
    "\n",
    "# ======================================================================\n",
    "def plotAccuracy(history, isError=True):\n",
    "    trainValues = np.array(history.history[\"acc\"])\n",
    "    validValues = np.array(history.history[\"val_acc\"])\n",
    "    \n",
    "    if (isError == True):\n",
    "        trainValues = 1-trainValues\n",
    "        validValues = 1-validValues\n",
    "    \n",
    "    trainAxe, = plt.plot(trainValues, label=\"Training set\")\n",
    "    validAxe, = plt.plot(validValues, label=\"Validation set\")\n",
    "    \n",
    "    if (isError == True):\n",
    "        plt.ylabel('Error (fraction)')\n",
    "    else:\n",
    "        plt.ylabel('Accuracy (fraction)')\n",
    "\n",
    "    plt.xlabel('Epoch')\n",
    "    \n",
    "    plt.legend(handles=[trainAxe, validAxe])\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now, we will load the dataset. The dataset is already saved in binary numpy format using another script written specifically for this study. The data loaded is the time series data and the labels. Note that we have two datasets."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "folderName = \"Datasets\"\n",
    "datasetNumber = 9\n",
    "testPercent = .1\n",
    "\n",
    "data = np.load(folderName + \"\\Data_D\" + str(datasetNumber) + \".npy\")\n",
    "label = np.load(folderName + \"\\Label_D\" + str(datasetNumber) + \".npy\")\n",
    "\n",
    "# for recognition dataset (12,13,14)\n",
    "if (datasetNumber == 12 or datasetNumber == 13 or datasetNumber == 14):\n",
    "    data = data[25000:35000,:,:]\n",
    "\n",
    "# for prediction dataset (9,10,11)\n",
    "if (datasetNumber == 9 or datasetNumber == 10 or datasetNumber == 11):\n",
    "    data = data[29500:30000,:,:]\n",
    "\n",
    "# Calculate the training percentage\n",
    "trainingPercent = 1 - testPercent;\n",
    "\n",
    "# Normalize the dataset\n",
    "data = normalizeData(data, axis=1)\n",
    "\n",
    "# Split the dataset\n",
    "Xs, Ys = splitDataset1D(data, label, proportions=[trainingPercent, testPercent])\n",
    "\n",
    "xTrain = np.swapaxes(Xs[0],0,1)\n",
    "xTest = np.swapaxes(Xs[1],0,1)\n",
    "\n",
    "yTrain = Ys[0][:,0]\n",
    "yTest = Ys[1][:,0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Neural network model\n",
    "\n",
    "The model that will be used for this 1D convolutional network is as discribed in the figure."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_17 (InputLayer)        (None, 1000, 1)           0         \n",
      "_________________________________________________________________\n",
      "conv1d_49 (Conv1D)           (None, 997, 8)            40        \n",
      "_________________________________________________________________\n",
      "max_pooling1d_49 (MaxPooling (None, 499, 8)            0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_49 (Batc (None, 499, 8)            32        \n",
      "_________________________________________________________________\n",
      "activation_49 (Activation)   (None, 499, 8)            0         \n",
      "_________________________________________________________________\n",
      "dropout_49 (Dropout)         (None, 499, 8)            0         \n",
      "_________________________________________________________________\n",
      "conv1d_50 (Conv1D)           (None, 493, 16)           528       \n",
      "_________________________________________________________________\n",
      "max_pooling1d_50 (MaxPooling (None, 247, 16)           0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_50 (Batc (None, 247, 16)           64        \n",
      "_________________________________________________________________\n",
      "activation_50 (Activation)   (None, 247, 16)           0         \n",
      "_________________________________________________________________\n",
      "dropout_50 (Dropout)         (None, 247, 16)           0         \n",
      "_________________________________________________________________\n",
      "conv1d_51 (Conv1D)           (None, 241, 32)           2080      \n",
      "_________________________________________________________________\n",
      "max_pooling1d_51 (MaxPooling (None, 121, 32)           0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_51 (Batc (None, 121, 32)           128       \n",
      "_________________________________________________________________\n",
      "activation_51 (Activation)   (None, 121, 32)           0         \n",
      "_________________________________________________________________\n",
      "dropout_51 (Dropout)         (None, 121, 32)           0         \n",
      "_________________________________________________________________\n",
      "flatten_17 (Flatten)         (None, 3872)              0         \n",
      "_________________________________________________________________\n",
      "dense_17 (Dense)             (None, 1)                 3873      \n",
      "=================================================================\n",
      "Total params: 6,745\n",
      "Trainable params: 6,633\n",
      "Non-trainable params: 112\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "def constructModel(inputShape):\n",
    "    \"\"\"\n",
    "    Function creating the model's graph in Keras.\n",
    "    \n",
    "    Argument:\n",
    "    input_shape -- shape of the model's input data (using Keras conventions)\n",
    "\n",
    "    Returns:\n",
    "    model -- Keras model instance\n",
    "    \"\"\"\n",
    "    \n",
    "    xInput = Input(shape = inputShape)\n",
    "    \n",
    "    X = Conv1D(8, kernel_size=4, strides=1, dilation_rate=1)(xInput)       \n",
    "    X = MaxPooling1D(pool_size=2, strides=None, padding='same')(X)\n",
    "    X = BatchNormalization()(X)                                          \n",
    "    X = Activation('relu')(X)                                              \n",
    "    X = Dropout(.05)(X)                                                  \n",
    "\n",
    "    X = Conv1D(16, kernel_size=4, strides=1, dilation_rate=2)(X)       \n",
    "    X = MaxPooling1D(pool_size=2, strides=None, padding='same')(X)\n",
    "    X = BatchNormalization()(X)                          \n",
    "    X = Activation('relu')(X)                             \n",
    "    X = Dropout(.05)(X)                                  \n",
    "\n",
    "    X = Conv1D(32, kernel_size=4, strides=1, dilation_rate=2)(X)    \n",
    "    X = MaxPooling1D(pool_size=2, strides=None, padding='same')(X)\n",
    "    X = BatchNormalization()(X)                            \n",
    "    X = Activation('relu')(X)                             \n",
    "    X = Dropout(.05)(X)                                 \n",
    "    \n",
    "    X = Flatten()(X)\n",
    "    X = Dense(1, activation = \"sigmoid\")(X) \n",
    "\n",
    "    model = Model(inputs = xInput, outputs = X)\n",
    "    \n",
    "    return model\n",
    "\n",
    "model = constructModel(inputShape = (xTrain.shape[1], 1))\n",
    "model.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Training the model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 2577 samples, validate on 645 samples\n",
      "Epoch 1/40\n",
      "2577/2577 [==============================] - 6s 2ms/step - loss: 0.6995 - acc: 0.5367 - val_loss: 0.6842 - val_acc: 0.5659\n",
      "Epoch 2/40\n",
      "2577/2577 [==============================] - 3s 986us/step - loss: 0.7030 - acc: 0.5537 - val_loss: 0.6837 - val_acc: 0.5488\n",
      "Epoch 3/40\n",
      "2577/2577 [==============================] - 2s 957us/step - loss: 0.7161 - acc: 0.5320 - val_loss: 0.6819 - val_acc: 0.5659\n",
      "Epoch 4/40\n",
      "2577/2577 [==============================] - 2s 965us/step - loss: 0.6947 - acc: 0.4913 - val_loss: 0.6835 - val_acc: 0.4992\n",
      "Epoch 5/40\n",
      "2577/2577 [==============================] - 2s 967us/step - loss: 0.7036 - acc: 0.5530 - val_loss: 0.6893 - val_acc: 0.5659\n",
      "Epoch 6/40\n",
      "2577/2577 [==============================] - 2s 967us/step - loss: 0.6900 - acc: 0.5537 - val_loss: 0.6871 - val_acc: 0.5581\n",
      "Epoch 7/40\n",
      "2577/2577 [==============================] - 2s 953us/step - loss: 0.6899 - acc: 0.5386 - val_loss: 0.6893 - val_acc: 0.5659\n",
      "Epoch 8/40\n",
      "2577/2577 [==============================] - 2s 962us/step - loss: 0.6841 - acc: 0.5499 - val_loss: 0.6763 - val_acc: 0.5612\n",
      "Epoch 9/40\n",
      "2577/2577 [==============================] - 3s 972us/step - loss: 0.6939 - acc: 0.5468 - val_loss: 0.6785 - val_acc: 0.5674\n",
      "Epoch 10/40\n",
      "2577/2577 [==============================] - 2s 965us/step - loss: 0.6974 - acc: 0.5456 - val_loss: 0.6980 - val_acc: 0.5628\n",
      "Epoch 11/40\n",
      "2577/2577 [==============================] - 2s 969us/step - loss: 0.6906 - acc: 0.5386 - val_loss: 0.6848 - val_acc: 0.5519\n",
      "Epoch 12/40\n",
      "2577/2577 [==============================] - 3s 972us/step - loss: 0.6839 - acc: 0.5452 - val_loss: 0.6834 - val_acc: 0.5643\n",
      "Epoch 13/40\n",
      "2577/2577 [==============================] - 2s 962us/step - loss: 0.6800 - acc: 0.5464 - val_loss: 0.6868 - val_acc: 0.5643\n",
      "Epoch 14/40\n",
      "2577/2577 [==============================] - 2s 961us/step - loss: 0.6851 - acc: 0.5479 - val_loss: 0.6968 - val_acc: 0.5628\n",
      "Epoch 15/40\n",
      "2577/2577 [==============================] - 2s 967us/step - loss: 0.6953 - acc: 0.5518 - val_loss: 0.6821 - val_acc: 0.5659\n",
      "Epoch 16/40\n",
      "2577/2577 [==============================] - 2s 970us/step - loss: 0.6771 - acc: 0.5537 - val_loss: 0.6809 - val_acc: 0.5674\n",
      "Epoch 17/40\n",
      "2577/2577 [==============================] - 3s 971us/step - loss: 0.6758 - acc: 0.5506 - val_loss: 0.6801 - val_acc: 0.5643\n",
      "Epoch 18/40\n",
      "2577/2577 [==============================] - 2s 966us/step - loss: 0.6773 - acc: 0.5634 - val_loss: 0.6814 - val_acc: 0.5643\n",
      "Epoch 19/40\n",
      "2577/2577 [==============================] - 2s 955us/step - loss: 0.6884 - acc: 0.5514 - val_loss: 0.6866 - val_acc: 0.5628\n",
      "Epoch 20/40\n",
      "2577/2577 [==============================] - 2s 963us/step - loss: 0.6787 - acc: 0.5514 - val_loss: 0.6810 - val_acc: 0.5612\n",
      "Epoch 21/40\n",
      "2577/2577 [==============================] - 3s 974us/step - loss: 0.6761 - acc: 0.5572 - val_loss: 0.6810 - val_acc: 0.5581\n",
      "Epoch 22/40\n",
      "2577/2577 [==============================] - 3s 977us/step - loss: 0.6749 - acc: 0.5588 - val_loss: 0.6812 - val_acc: 0.5550\n",
      "Epoch 23/40\n",
      "2577/2577 [==============================] - 2s 967us/step - loss: 0.6782 - acc: 0.5530 - val_loss: 0.6808 - val_acc: 0.5612\n",
      "Epoch 24/40\n",
      "2577/2577 [==============================] - 3s 971us/step - loss: 0.6840 - acc: 0.5592 - val_loss: 0.6825 - val_acc: 0.5612\n",
      "Epoch 25/40\n",
      "2577/2577 [==============================] - 3s 1ms/step - loss: 0.6750 - acc: 0.5596 - val_loss: 0.6792 - val_acc: 0.5612\n",
      "Epoch 26/40\n",
      "2577/2577 [==============================] - 3s 1ms/step - loss: 0.6740 - acc: 0.5568 - val_loss: 0.6785 - val_acc: 0.5612\n",
      "Epoch 27/40\n",
      "2577/2577 [==============================] - 3s 1ms/step - loss: 0.6713 - acc: 0.5650 - val_loss: 0.6782 - val_acc: 0.5643\n",
      "Epoch 28/40\n",
      "2577/2577 [==============================] - 3s 1ms/step - loss: 0.6728 - acc: 0.5662 - val_loss: 0.6775 - val_acc: 0.5659\n",
      "Epoch 29/40\n",
      "2577/2577 [==============================] - 2s 970us/step - loss: 0.6752 - acc: 0.5600 - val_loss: 0.6791 - val_acc: 0.5674\n",
      "Epoch 30/40\n",
      "2577/2577 [==============================] - 2s 969us/step - loss: 0.6714 - acc: 0.5615 - val_loss: 0.6791 - val_acc: 0.5628\n",
      "Epoch 31/40\n",
      "2577/2577 [==============================] - 2s 965us/step - loss: 0.6726 - acc: 0.5623 - val_loss: 0.6789 - val_acc: 0.5659\n",
      "Epoch 32/40\n",
      "2577/2577 [==============================] - 3s 975us/step - loss: 0.6747 - acc: 0.5541 - val_loss: 0.6781 - val_acc: 0.5705\n",
      "Epoch 33/40\n",
      "2577/2577 [==============================] - 2s 969us/step - loss: 0.6757 - acc: 0.5468 - val_loss: 0.6773 - val_acc: 0.5581\n",
      "Epoch 34/40\n",
      "2577/2577 [==============================] - 2s 970us/step - loss: 0.6745 - acc: 0.5681 - val_loss: 0.6777 - val_acc: 0.5628\n",
      "Epoch 35/40\n",
      "2577/2577 [==============================] - 3s 995us/step - loss: 0.6736 - acc: 0.5576 - val_loss: 0.6791 - val_acc: 0.5612\n",
      "Epoch 36/40\n",
      "2577/2577 [==============================] - 3s 984us/step - loss: 0.6725 - acc: 0.5669 - val_loss: 0.6788 - val_acc: 0.5612\n",
      "Epoch 37/40\n",
      "2577/2577 [==============================] - 2s 968us/step - loss: 0.6781 - acc: 0.5662 - val_loss: 0.6792 - val_acc: 0.5597\n",
      "Epoch 38/40\n",
      "2577/2577 [==============================] - 3s 971us/step - loss: 0.6733 - acc: 0.5755 - val_loss: 0.6796 - val_acc: 0.5659\n",
      "Epoch 39/40\n",
      "2577/2577 [==============================] - 2s 965us/step - loss: 0.6755 - acc: 0.5685 - val_loss: 0.6795 - val_acc: 0.5628\n",
      "Epoch 40/40\n",
      "2577/2577 [==============================] - 2s 963us/step - loss: 0.6738 - acc: 0.5596 - val_loss: 0.6779 - val_acc: 0.5659\n"
     ]
    }
   ],
   "source": [
    "model = constructModel(inputShape = (xTrain.shape[1], 1))\n",
    "opt = Adam(lr=0.002, beta_1=0.9, beta_2=0.999, decay=0.01)\n",
    "model.compile(loss='binary_crossentropy', optimizer=opt, metrics=[\"accuracy\"])\n",
    "history = model.fit(xTrain, yTrain, batch_size = 256, epochs=40,  validation_split=.2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "358/358 [==============================] - 0s 302us/step\n",
      "Test set accuracy =  0.5418994416737689\n",
      "Guess Accuracy =  0.5332402234636872\n",
      "[[0.51675978 0.0027933 ]\n",
      " [0.45530726 0.02513966]]\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x26142c98cc0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "loss, acc = model.evaluate(xTest, yTest)\n",
    "print(\"Test set accuracy = \", acc)\n",
    "print(\"Guess Accuracy = \", 1-np.sum(label)/label.size)\n",
    "\n",
    "yPred = model.predict(xTest) > .5                      \n",
    "print(confusion_matrix(yTest, yPred)/yTest.size)\n",
    "\n",
    "plotAccuracy(history, False)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}