1DCNN_demo.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from six.moves import cPickle as pickle\n",
    "import keras\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Conv1D, MaxPooling1D, Flatten, Dense, Dropout\n",
    "from keras.callbacks import ModelCheckpoint"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Loading the data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "pfile = r\"./data/ABCD_Datasets.pickle\"\n",
    "with (open(pfile, \"rb\")) as openfile:\n",
    "    while True:\n",
    "        try:\n",
    "            ABCD_Datasets = pickle.load(openfile)\n",
    "        except EOFError:\n",
    "            break\n",
    "\n",
    "X_train_D = ABCD_Datasets[\"train_datasets\"][\"D\"]\n",
    "Y_train_D = ABCD_Datasets[\"train_labels\"][\"D\"]\n",
    "X_test_D = ABCD_Datasets[\"test_datasets\"][\"D\"]\n",
    "Y_test_D = ABCD_Datasets[\"test_labels\"][\"D\"]\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "one hot-encoding and creating one-hot encoding, quick check"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "x_train shape: (17800, 2048, 2)\n",
      "17800 train samples\n",
      "750 test samples\n",
      "2000 validation samples\n"
     ]
    }
   ],
   "source": [
    "# one-hot encode the labels\n",
    "num_classes = 10\n",
    "Y_train_D_hot = keras.utils.to_categorical(Y_train_D-1, num_classes)\n",
    "Y_test_D_hot = keras.utils.to_categorical(Y_test_D-1, num_classes)\n",
    "\n",
    "# break training set into training and validation sets\n",
    "(X_train, X_valid) = X_train_D[2000:], X_train_D[:2000]\n",
    "(Y_train, Y_valid) = Y_train_D_hot[2000:], Y_train_D_hot[:2000]\n",
    "X_test = X_test_D\n",
    "Y_test = Y_test_D_hot\n",
    "\n",
    "# print shape of training set\n",
    "print('x_train shape:', X_train.shape)\n",
    "\n",
    "# print number of training, validation, and test images\n",
    "print(X_train.shape[0], 'train samples')\n",
    "print(X_test.shape[0], 'test samples')\n",
    "print(X_valid.shape[0], 'validation samples')\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's start with simple methods.\n",
    "1. Rondom guess\n",
    "2. logistic regression (or 2 layers NN model)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "flatten_2 (Flatten)          (None, 2048)              0         \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 10)                20490     \n",
      "=================================================================\n",
      "Total params: 20,490\n",
      "Trainable params: 20,490\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "Train on 17800 samples, validate on 2000 samples\n",
      "Epoch 1/10\n",
      "17800/17800 [==============================] - 2s 131us/step - loss: 2.3613 - acc: 0.2061 - val_loss: 2.3690 - val_acc: 0.2060\n",
      "Epoch 2/10\n",
      "17800/17800 [==============================] - 2s 120us/step - loss: 2.2478 - acc: 0.2508 - val_loss: 2.3459 - val_acc: 0.2165\n",
      "Epoch 3/10\n",
      "17800/17800 [==============================] - 2s 117us/step - loss: 2.1805 - acc: 0.3015 - val_loss: 2.3461 - val_acc: 0.2445\n",
      "Epoch 4/10\n",
      "17800/17800 [==============================] - 2s 118us/step - loss: 2.1348 - acc: 0.3290 - val_loss: 2.3254 - val_acc: 0.2475\n",
      "Epoch 5/10\n",
      "17800/17800 [==============================] - 2s 118us/step - loss: 2.0981 - acc: 0.3531 - val_loss: 2.3203 - val_acc: 0.2575\n",
      "Epoch 6/10\n",
      "17800/17800 [==============================] - 2s 116us/step - loss: 2.0670 - acc: 0.3692 - val_loss: 2.2941 - val_acc: 0.2695\n",
      "Epoch 7/10\n",
      "17800/17800 [==============================] - 2s 116us/step - loss: 2.0406 - acc: 0.3784 - val_loss: 2.2946 - val_acc: 0.2580\n",
      "Epoch 8/10\n",
      "17800/17800 [==============================] - 2s 116us/step - loss: 2.0173 - acc: 0.3835 - val_loss: 2.2880 - val_acc: 0.2700\n",
      "Epoch 9/10\n",
      "17800/17800 [==============================] - 2s 116us/step - loss: 1.9996 - acc: 0.3903 - val_loss: 2.2928 - val_acc: 0.2645\n",
      "Epoch 10/10\n",
      "17800/17800 [==============================] - 2s 117us/step - loss: 1.9841 - acc: 0.3957 - val_loss: 2.2980 - val_acc: 0.2700\n",
      "\n",
      " Logistic Regression Test accuracy: 0.308000000636\n"
     ]
    }
   ],
   "source": [
    "# only using 1 channel now\n",
    "CNNch = 1\n",
    "\n",
    "# epch\n",
    "ne = 10\n",
    "\n",
    "# define the model\n",
    "modelLog = Sequential()\n",
    "modelLog.add(Flatten(input_shape = (2048,CNNch)))\n",
    "modelLog.add(Dense(10, activation='softmax'))\n",
    "\n",
    "modelLog.summary()\n",
    "# total param 2,566,642\n",
    "# ref  CNN       53,830\n",
    "modelLog.compile(loss='categorical_crossentropy', optimizer='rmsprop', \n",
    "                  metrics=['accuracy'])\n",
    "\n",
    "hist = modelLog.fit(X_train[:,:,0:CNNch], Y_train, batch_size=32, epochs=ne,\n",
    "          validation_data=(X_valid[:,:,0:CNNch], Y_valid), \n",
    "          verbose=1, shuffle=True)\n",
    "\n",
    "score = modelLog.evaluate(X_test[:,:,0:CNNch], Y_test, verbose=0)\n",
    "print('\\n', 'Logistic Regression Test accuracy:', score[1])\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "1D-CNN\n",
    "\n",
    "From the moddel suggested in the paper \n",
    "* the 2nd layer # of filter(kernel) is reduced from 32 to 16.\n",
    "* the 3rd layer # of filter(kernel) is reduced from 64 to 32.\n",
    "* the 4th layer # of filter(kernel) is reduced from 64 to 32.\n",
    "* the 5th layer # of filter(kernel) is reduced from 64 to 32.\n",
    "* the last layer fullly connected layer is reduced from 100 to 50.\n",
    "\n",
    "Just to make the number of variable even smaller than the logistic regression above.\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv1d_6 (Conv1D)            (None, 128, 16)           2064      \n",
      "_________________________________________________________________\n",
      "max_pooling1d_6 (MaxPooling1 (None, 64, 16)            0         \n",
      "_________________________________________________________________\n",
      "conv1d_7 (Conv1D)            (None, 64, 16)            784       \n",
      "_________________________________________________________________\n",
      "max_pooling1d_7 (MaxPooling1 (None, 32, 16)            0         \n",
      "_________________________________________________________________\n",
      "conv1d_8 (Conv1D)            (None, 32, 32)            1568      \n",
      "_________________________________________________________________\n",
      "max_pooling1d_8 (MaxPooling1 (None, 16, 32)            0         \n",
      "_________________________________________________________________\n",
      "dropout_4 (Dropout)          (None, 16, 32)            0         \n",
      "_________________________________________________________________\n",
      "conv1d_9 (Conv1D)            (None, 16, 32)            3104      \n",
      "_________________________________________________________________\n",
      "max_pooling1d_9 (MaxPooling1 (None, 8, 32)             0         \n",
      "_________________________________________________________________\n",
      "dropout_5 (Dropout)          (None, 8, 32)             0         \n",
      "_________________________________________________________________\n",
      "conv1d_10 (Conv1D)           (None, 8, 32)             3104      \n",
      "_________________________________________________________________\n",
      "max_pooling1d_10 (MaxPooling (None, 4, 32)             0         \n",
      "_________________________________________________________________\n",
      "flatten_4 (Flatten)          (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_5 (Dense)              (None, 50)                6450      \n",
      "_________________________________________________________________\n",
      "dropout_6 (Dropout)          (None, 50)                0         \n",
      "_________________________________________________________________\n",
      "dense_6 (Dense)              (None, 10)                510       \n",
      "=================================================================\n",
      "Total params: 17,584\n",
      "Trainable params: 17,584\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "Train on 17800 samples, validate on 2000 samples\n",
      "Epoch 1/10\n",
      "17800/17800 [==============================] - 6s 310us/step - loss: 0.6830 - acc: 0.7360 - val_loss: 0.0371 - val_acc: 0.9920\n",
      "\n",
      "Epoch 00001: val_loss improved from inf to 0.03707, saving model to CNNC2.weights.best.hdf5\n",
      "Epoch 2/10\n",
      "17800/17800 [==============================] - 5s 273us/step - loss: 0.0607 - acc: 0.9807 - val_loss: 0.0218 - val_acc: 0.9935\n",
      "\n",
      "Epoch 00002: val_loss improved from 0.03707 to 0.02179, saving model to CNNC2.weights.best.hdf5\n",
      "Epoch 3/10\n",
      "17800/17800 [==============================] - 5s 270us/step - loss: 0.0348 - acc: 0.9889 - val_loss: 0.2573 - val_acc: 0.9205\n",
      "\n",
      "Epoch 00003: val_loss did not improve\n",
      "Epoch 4/10\n",
      "17800/17800 [==============================] - 5s 267us/step - loss: 0.0298 - acc: 0.9910 - val_loss: 0.0023 - val_acc: 0.9995\n",
      "\n",
      "Epoch 00004: val_loss improved from 0.02179 to 0.00233, saving model to CNNC2.weights.best.hdf5\n",
      "Epoch 5/10\n",
      "17800/17800 [==============================] - 5s 277us/step - loss: 0.0191 - acc: 0.9943 - val_loss: 0.0128 - val_acc: 0.9960\n",
      "\n",
      "Epoch 00005: val_loss did not improve\n",
      "Epoch 6/10\n",
      "17800/17800 [==============================] - 5s 274us/step - loss: 0.0146 - acc: 0.9958 - val_loss: 0.0063 - val_acc: 0.9980\n",
      "\n",
      "Epoch 00006: val_loss did not improve\n",
      "Epoch 7/10\n",
      "17800/17800 [==============================] - 5s 269us/step - loss: 0.0133 - acc: 0.9959 - val_loss: 0.0013 - val_acc: 0.9995\n",
      "\n",
      "Epoch 00007: val_loss improved from 0.00233 to 0.00126, saving model to CNNC2.weights.best.hdf5\n",
      "Epoch 8/10\n",
      "17800/17800 [==============================] - 5s 272us/step - loss: 0.0129 - acc: 0.9962 - val_loss: 0.0027 - val_acc: 0.9990\n",
      "\n",
      "Epoch 00008: val_loss did not improve\n",
      "Epoch 9/10\n",
      "17800/17800 [==============================] - 5s 267us/step - loss: 0.0101 - acc: 0.9975 - val_loss: 0.0041 - val_acc: 0.9985\n",
      "\n",
      "Epoch 00009: val_loss did not improve\n",
      "Epoch 10/10\n",
      "17800/17800 [==============================] - 5s 271us/step - loss: 0.0107 - acc: 0.9976 - val_loss: 6.9428e-05 - val_acc: 1.0000\n",
      "\n",
      "Epoch 00010: val_loss improved from 0.00126 to 0.00007, saving model to CNNC2.weights.best.hdf5\n",
      "\n",
      " CNN Test accuracy: 1.0\n",
      "\n",
      " CNN train accuracy: 1.0\n",
      "\n",
      " CNN validation accuracy: 1.0\n"
     ]
    }
   ],
   "source": [
    "CNNch = 2\n",
    "\n",
    "# epch\n",
    "ne = 10\n",
    "\n",
    "modelC2 = Sequential()\n",
    "#1\n",
    "modelC2.add(Conv1D(filters=16, kernel_size=64,strides = 16, padding='same', activation='relu', \n",
    "                        input_shape=(2048, CNNch)))\n",
    "modelC2.add(MaxPooling1D(pool_size=2))\n",
    "#2\n",
    "modelC2.add(Conv1D(filters=16, kernel_size=3, strides = 1, padding='same', activation='relu'))\n",
    "modelC2.add(MaxPooling1D(pool_size=2))\n",
    "#3\n",
    "modelC2.add(Conv1D(filters=32, kernel_size=3, strides = 1, padding='same', activation='relu'))\n",
    "modelC2.add(MaxPooling1D(pool_size=2))\n",
    "modelC2.add(Dropout(0.2))\n",
    "#4\n",
    "modelC2.add(Conv1D(filters=32, kernel_size=3, strides = 1, padding='same', activation='relu'))\n",
    "modelC2.add(MaxPooling1D(pool_size=2))\n",
    "modelC2.add(Dropout(0.2))\n",
    "#5\n",
    "modelC2.add(Conv1D(filters=32, kernel_size=3, strides = 1, padding='same', activation='relu'))\n",
    "#paper no padding?, Yes, to make 5th layer output 6 width and 3 after pooling\n",
    "#-> same seems to perform little better because of more parameter? \n",
    "# little diffrernt from the paper but keep it as padding = 'same'\n",
    "modelC2.add(MaxPooling1D(pool_size=2))  \n",
    "\n",
    "modelC2.add(Flatten())\n",
    "modelC2.add(Dense(50, activation='relu'))\n",
    "modelC2.add(Dropout(0.2))\n",
    "modelC2.add(Dense(10, activation='softmax'))\n",
    "\n",
    "modelC2.summary()\n",
    "\n",
    "\n",
    "# compile the model\n",
    "modelC2.compile(loss='categorical_crossentropy', optimizer='rmsprop', \n",
    "                  metrics=['accuracy'])\n",
    "\n",
    "# train the model\n",
    "checkpointer = ModelCheckpoint(filepath='CNNC2.weights.best.hdf5', verbose=1, \n",
    "                               save_best_only=True)\n",
    "\n",
    "hist = modelC2.fit(X_train[:,:,0:CNNch], Y_train, batch_size=32, epochs=ne,\n",
    "          validation_data=(X_valid[:,:,0:CNNch], Y_valid), callbacks=[checkpointer], \n",
    "          verbose=1, shuffle=True)\n",
    "\n",
    "# load the weights that yielded the best validation accuracy\n",
    "modelC2.load_weights('CNNC2.weights.best.hdf5')\n",
    "\n",
    "# evaluate and print test accuracy\n",
    "score = modelC2.evaluate(X_test[:,:,0:CNNch], Y_test, verbose=0)\n",
    "print('\\n', 'CNN Test accuracy:', score[1])\n",
    "\n",
    "score = modelC2.evaluate(X_train[:,:,0:CNNch], Y_train, verbose=0)\n",
    "print('\\n', 'CNN train accuracy:', score[1])\n",
    "\n",
    "score = modelC2.evaluate(X_valid[:,:,0:CNNch], Y_valid, verbose=0)\n",
    "print('\\n', 'CNN validation accuracy:', score[1])"
   ]
  },
  {
   "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": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}