mnist_baseline.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### MNIST CNN with same topology as FT_MNIST to compare"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "cellView": "both",
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "id": "apJbCsBHl-2A"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/mathew/anaconda/envs/egpu/lib/python2.7/site-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n",
      "  from ._conv import register_converters as _register_converters\n",
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "#Choose GPUs and CPUs\n",
    "\n",
    "#If want to choose GPU to use\n",
    "#import os\n",
    "#os.environ[\"CUDA_VISIBLE_DEVICES\"]=\"1\"\n",
    "#This will make it so that only the 0th GPU is visible to TensorFlow.\n",
    "#In your case, you can choose any in the range [0, 3].\n",
    "#If you wanted, for example, the 0th and 2nd GPUs to both be visible, replace \"0\" with \"0,2\"\n",
    "\n",
    "import tensorflow as tf\n",
    "from keras import backend as K\n",
    "import keras\n",
    "\n",
    "GPU=True\n",
    "CPU=False\n",
    "num_cores = 8\n",
    "\n",
    "if GPU:\n",
    "    num_GPU = 1\n",
    "    num_CPU = 2\n",
    "if CPU:\n",
    "    num_CPU = num_cores\n",
    "    num_GPU = 0\n",
    "\n",
    "config = tf.ConfigProto(intra_op_parallelism_threads=num_cores,\\\n",
    "        inter_op_parallelism_threads=num_cores, allow_soft_placement=True,\\\n",
    "        device_count = {'CPU' : num_CPU, 'GPU' : num_GPU},\\\n",
    "#        gpu_options=tf.GPUOptions(per_process_gpu_memory_fraction=0.5))\n",
    "                       )\n",
    "config.gpu_options.allow_growth = True\n",
    "\n",
    "session = tf.Session(config=config)\n",
    "K.set_session(session)\n",
    "\n",
    "from __future__ import print_function\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "\n",
    "\n",
    "from keras.layers import Conv2D, MaxPool2D\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, Dropout, Activation, Flatten\n",
    "from keras.utils import to_categorical\n",
    "# Config the matplotlib backend as plotting inline in IPython\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Reload all data from before"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "train_dataset shape: (60000, 28, 28, 1)\n",
      "60000 train samples\n",
      "10000 test samples\n"
     ]
    }
   ],
   "source": [
    "\n",
    "from keras.datasets import mnist\n",
    "\n",
    "\n",
    "# input image dimensions\n",
    "img_rows, img_cols = 28, 28\n",
    "num_classes=10\n",
    "\n",
    "# the data, split between train and test sets\n",
    "(x_train, y_train), (x_test, y_test) = mnist.load_data()\n",
    "\n",
    "if K.image_data_format() == 'channels_first':\n",
    "    x_train = x_train.reshape(x_train.shape[0], 1, img_rows, img_cols)\n",
    "    x_test = x_test.reshape(x_test.shape[0], 1, img_rows, img_cols)\n",
    "    input_shape = (1, img_rows, img_cols)\n",
    "else:\n",
    "    x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 1)\n",
    "    x_test = x_test.reshape(x_test.shape[0], img_rows, img_cols, 1)\n",
    "    input_shape = (img_rows, img_cols, 1)\n",
    "\n",
    "train_dataset = x_train.astype('float32')\n",
    "test_dataset = x_test.astype('float32')\n",
    "train_dataset /= 255\n",
    "test_dataset /= 255\n",
    "print('train_dataset shape:', x_train.shape)\n",
    "print(x_train.shape[0], 'train samples')\n",
    "print(x_test.shape[0], 'test samples')\n",
    "\n",
    "# convert class vectors to binary class matrices\n",
    "train_labels = keras.utils.to_categorical(y_train, num_classes)\n",
    "test_labels = keras.utils.to_categorical(y_test, num_classes)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Pad numpy array \n",
    "Want get 36x36 which is how much the multiMNIST will be"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(60000, 36, 36, 1)\n",
      "(10000, 36, 36, 1)\n"
     ]
    }
   ],
   "source": [
    "train_dataset=np.pad(train_dataset, ((0,0),(4,4),(4,4),(0,0)), 'constant', constant_values=(0,0))\n",
    "print (train_dataset.shape)\n",
    "\n",
    "test_dataset=np.pad(test_dataset, ((0,0),(4,4),(4,4),(0,0)), 'constant', constant_values=(0,0))\n",
    "print (test_dataset.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# O.K now try CNN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(60000, 36, 36, 1)\n",
      "(60000, 36, 36, 1)\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_1 (Conv2D)            (None, 36, 36, 32)        320       \n",
      "_________________________________________________________________\n",
      "conv2d_2 (Conv2D)            (None, 36, 36, 32)        9248      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2 (None, 18, 18, 32)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_3 (Conv2D)            (None, 18, 18, 64)        18496     \n",
      "_________________________________________________________________\n",
      "conv2d_4 (Conv2D)            (None, 18, 18, 64)        36928     \n",
      "_________________________________________________________________\n",
      "max_pooling2d_2 (MaxPooling2 (None, 9, 9, 64)          0         \n",
      "_________________________________________________________________\n",
      "conv2d_5 (Conv2D)            (None, 9, 9, 128)         73856     \n",
      "_________________________________________________________________\n",
      "conv2d_6 (Conv2D)            (None, 9, 9, 128)         147584    \n",
      "_________________________________________________________________\n",
      "max_pooling2d_3 (MaxPooling2 (None, 4, 4, 128)         0         \n",
      "_________________________________________________________________\n",
      "flatten_1 (Flatten)          (None, 2048)              0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 600)               1229400   \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 600)               0         \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 10)                6010      \n",
      "=================================================================\n",
      "Total params: 1,521,842\n",
      "Trainable params: 1,521,842\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n",
      "Train on 54000 samples, validate on 6000 samples\n",
      "Epoch 1/10\n",
      "53504/54000 [============================>.] - ETA: 0s - loss: 0.2733 - acc: 0.9086Epoch 00000: saving model to mn_models/weights.00.hdf5\n",
      "54000/54000 [==============================] - 6s - loss: 0.2715 - acc: 0.9092 - val_loss: 0.0471 - val_acc: 0.9857\n",
      "Epoch 2/10\n",
      "53504/54000 [============================>.] - ETA: 0s - loss: 0.0564 - acc: 0.9828Epoch 00001: saving model to mn_models/weights.01.hdf5\n",
      "54000/54000 [==============================] - 6s - loss: 0.0563 - acc: 0.9828 - val_loss: 0.0313 - val_acc: 0.9917\n",
      "Epoch 3/10\n",
      "53504/54000 [============================>.] - ETA: 0s - loss: 0.0387 - acc: 0.9882Epoch 00002: saving model to mn_models/weights.02.hdf5\n",
      "54000/54000 [==============================] - 6s - loss: 0.0385 - acc: 0.9883 - val_loss: 0.0298 - val_acc: 0.9917\n",
      "Epoch 4/10\n",
      "53504/54000 [============================>.] - ETA: 0s - loss: 0.0284 - acc: 0.9913Epoch 00003: saving model to mn_models/weights.03.hdf5\n",
      "54000/54000 [==============================] - 6s - loss: 0.0285 - acc: 0.9913 - val_loss: 0.0296 - val_acc: 0.9922\n",
      "Epoch 5/10\n",
      "53504/54000 [============================>.] - ETA: 0s - loss: 0.0242 - acc: 0.9927Epoch 00004: saving model to mn_models/weights.04.hdf5\n",
      "54000/54000 [==============================] - 6s - loss: 0.0242 - acc: 0.9928 - val_loss: 0.0237 - val_acc: 0.9933\n",
      "Epoch 6/10\n",
      "53504/54000 [============================>.] - ETA: 0s - loss: 0.0185 - acc: 0.9939Epoch 00005: saving model to mn_models/weights.05.hdf5\n",
      "54000/54000 [==============================] - 6s - loss: 0.0183 - acc: 0.9939 - val_loss: 0.0248 - val_acc: 0.9933\n",
      "Epoch 7/10\n",
      "53504/54000 [============================>.] - ETA: 0s - loss: 0.0161 - acc: 0.9951Epoch 00006: saving model to mn_models/weights.06.hdf5\n",
      "54000/54000 [==============================] - 6s - loss: 0.0161 - acc: 0.9950 - val_loss: 0.0255 - val_acc: 0.9937\n",
      "Epoch 8/10\n",
      "53504/54000 [============================>.] - ETA: 0s - loss: 0.0146 - acc: 0.9953Epoch 00007: saving model to mn_models/weights.07.hdf5\n",
      "54000/54000 [==============================] - 6s - loss: 0.0146 - acc: 0.9954 - val_loss: 0.0239 - val_acc: 0.9932\n",
      "Epoch 9/10\n",
      "53504/54000 [============================>.] - ETA: 0s - loss: 0.0134 - acc: 0.9955Epoch 00008: saving model to mn_models/weights.08.hdf5\n",
      "54000/54000 [==============================] - 6s - loss: 0.0134 - acc: 0.9955 - val_loss: 0.0259 - val_acc: 0.9943\n",
      "Epoch 10/10\n",
      "53504/54000 [============================>.] - ETA: 0s - loss: 0.0109 - acc: 0.9961Epoch 00009: saving model to mn_models/weights.09.hdf5\n",
      "54000/54000 [==============================] - 6s - loss: 0.0109 - acc: 0.9961 - val_loss: 0.0216 - val_acc: 0.9940\n",
      "Test performance\n",
      " 9728/10000 [============================>.] - ETA: 0s[0.01975472029054181, 0.9938]\n"
     ]
    }
   ],
   "source": [
    "dims=train_dataset.shape\n",
    "print (dims)\n",
    "trainee=train_dataset.reshape(dims[0],dims[1],dims[2],1)\n",
    "dims=test_dataset.shape\n",
    "testee=test_dataset.reshape(dims[0],dims[1],dims[2],1)\n",
    "print (trainee.shape)\n",
    "\n",
    "\n",
    "K.clear_session()\n",
    "w_size=3\n",
    "np.random.seed(123)\n",
    "model=Sequential()\n",
    "\n",
    "model.add(Conv2D(32,(w_size,w_size), activation='relu', input_shape=(36,36,1),\n",
    "                     data_format='channels_last', padding='same'))\n",
    "model.add(Conv2D(32,(w_size,w_size), activation='relu', padding='same'))\n",
    "model.add(MaxPool2D(pool_size=(2,2)))\n",
    "model.add(Conv2D(64,(w_size,w_size), activation='relu', padding='same'))\n",
    "model.add(Conv2D(64,(w_size,w_size), activation='relu', padding='same'))\n",
    "model.add(MaxPool2D(pool_size=(2,2)))\n",
    "model.add(Conv2D(128,(w_size,w_size), activation='relu', padding='same'))\n",
    "model.add(Conv2D(128,(w_size,w_size), activation='relu', padding='same'))\n",
    "model.add(MaxPool2D(pool_size=(2,2)))\n",
    "model.add(Flatten())\n",
    "model.add(Dense(600, activation='relu'))\n",
    "model.add(Dropout(0.5))\n",
    "model.add(Dense(10, activation='softmax'))\n",
    "\n",
    "\n",
    "\n",
    "model.compile(loss='categorical_crossentropy',\n",
    "              optimizer='adam',\n",
    "              metrics=['accuracy'])\n",
    "\n",
    "print (model.summary())\n",
    "\n",
    "checkpoints=keras.callbacks.ModelCheckpoint('mn_models/weights.{epoch:02d}.hdf5', \n",
    "                                            monitor='val_loss', verbose=1, \n",
    "                                            save_best_only=False, \n",
    "                                            save_weights_only=False, \n",
    "                                            mode='auto', period=1)\n",
    "\n",
    "history=model.fit(trainee, train_labels, batch_size=256, verbose=1, epochs=10, \n",
    "                  validation_split = 0.1\n",
    "                  , callbacks=[checkpoints])\n",
    "print (\"Test performance\")\n",
    "score=model.evaluate(testee, test_labels, verbose=1)\n",
    "print (score)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x432 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "hist=history\n",
    "epochs=np.asarray(history.epoch)+1\n",
    "\n",
    "%matplotlib inline\n",
    "import matplotlib\n",
    "plt.style.use('seaborn-white')\n",
    "matplotlib.rc('font',family='Times New Roman')\n",
    "matplotlib.rcParams['font.size'] = 20\n",
    "\n",
    "f, axarr = plt.subplots(2, sharex=True, figsize=(8, 6))\n",
    "\n",
    "axarr[0].set(ylabel='Loss')\n",
    "axarr[0].plot(epochs,hist.history['loss'], 'C3o', label='Training')\n",
    "axarr[0].plot(epochs,hist.history['val_loss'], 'C3-', label='Validation')\n",
    "axarr[0].grid()\n",
    "axarr[0].legend(loc='center right', bbox_to_anchor=(1.0, 0.5))\n",
    "\n",
    "\n",
    "axarr[1].plot(epochs,hist.history['acc'], 'C0o', label='Training')\n",
    "axarr[1].plot(epochs,hist.history['val_acc'], 'C0-', label='Validation')\n",
    "plt.legend(loc='center right', bbox_to_anchor=(1.0, 0.5))\n",
    "plt.xlabel('Epochs')\n",
    "plt.ylabel('Accuracy')\n",
    "plt.tight_layout()\n",
    "plt.grid()\n",
    "plt.tick_params(which='both', width=2)\n",
    "plt.tick_params(which='major', length=7)\n",
    "plt.tick_params(which='minor', length=5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " 9792/10000 [============================>.] - ETA: 0s[0.01975472029054181, 0.9938]\n"
     ]
    }
   ],
   "source": [
    "from keras.models import Model, load_model\n",
    "model=load_model('mn_models/weights.09.hdf5')\n",
    "model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])\n",
    "score=model.evaluate(testee, test_labels, verbose=1)\n",
    "print (score)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "colab": {
   "default_view": {},
   "name": "1_notmnist.ipynb",
   "provenance": [],
   "version": "0.3.2",
   "views": {}
  },
  "kernelspec": {
   "display_name": "Python [conda env:egpu]",
   "language": "python",
   "name": "conda-env-egpu-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.15"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}