04-compare-cpu-gpu.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Why GPUs matter\n",
    "\n",
    "Let's compare GPU vs CPU performance at training a network...\n",
    "\n",
    "Of course, this notebook will only run on a machine with a TensorFlow-supported gpu. I tested in on a AWS g2 instance."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# for DSX, need to switch to the right directory. Detect using path name.\n",
    "s = %pwd\n",
    "if s.startswith('/gpfs'):\n",
    "    %cd ~/deep-learning-workshop/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import time\n",
    "\n",
    "# for making plots prettier\n",
    "import seaborn as sns \n",
    "sns.set_style('white')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from __future__ import print_function\n",
    "np.random.seed(1337)  # for reproducibility"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "# We'll need to use tensorflow directly to tell it to train on CPU or GPU\n",
    "import tensorflow as tf\n",
    "\n",
    "from keras.datasets import mnist\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, Dropout, Activation, Flatten\n",
    "from keras.layers import Convolution2D, MaxPooling2D\n",
    "from keras.utils import np_utils"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "batch_size = 128\n",
    "nb_classes = 10\n",
    "nb_epoch = 3\n",
    "\n",
    "# input image dimensions\n",
    "img_rows, img_cols = 28, 28\n",
    "# number of convolutional filters to use\n",
    "nb_filters = 64\n",
    "# size of pooling area for max pooling\n",
    "pool_size = (2, 2)\n",
    "# convolution kernel size\n",
    "kernel_size = (3, 3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "x_train shape: (50000, 28, 28, 1)\n",
      "50000 train samples\n",
      "10000 valid samples\n",
      "10000 test samples\n"
     ]
    }
   ],
   "source": [
    "import notmnist\n",
    "notmnist_path = \"~/data/notmnist/notMNIST.pickle\"\n",
    "\n",
    "# the data, shuffled and split between train, validation, and test sets\n",
    "(x_train, y_train), (x_valid, y_valid), (x_test, y_test) = notmnist.load_data(notmnist_path)\n",
    "\n",
    "# Reshape inputs to be flat, convert labels to 1-hot encoding.\n",
    "\n",
    "# Reshape to have an extra dimension (useless for our black&white images, useful for RGB)\n",
    "x_train = x_train.reshape(-1, img_rows, img_cols, 1)\n",
    "x_valid = x_valid.reshape(-1, img_rows, img_cols, 1)\n",
    "x_test = x_test.reshape(-1, img_rows, img_cols, 1)\n",
    "input_shape = (img_rows, img_cols, 1)\n",
    "print(\"x_train shape:\", x_train.shape)\n",
    "print(x_train.shape[0], 'train samples')\n",
    "print(x_valid.shape[0], 'valid samples')\n",
    "print(x_test.shape[0], 'test samples')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# convert class vectors to binary class matrices\n",
    "y_train = np_utils.to_categorical(y_train, nb_classes)\n",
    "y_valid = np_utils.to_categorical(y_valid, nb_classes)\n",
    "y_test = np_utils.to_categorical(y_test, nb_classes)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Helper function to make same model twice, once on CPU, then on GPU\n",
    "def make_model():\n",
    "    model = Sequential()\n",
    "    \n",
    "    # Here, we're making _convolutional_ network. We'll learn the details\n",
    "    # in future sessions. The key is that it requires lots of parallelizable \n",
    "    # computation...\n",
    "    model.add(Convolution2D(nb_filters, kernel_size[0], kernel_size[1],\n",
    "                            border_mode='valid',\n",
    "                            input_shape=input_shape))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(Convolution2D(nb_filters, kernel_size[0], kernel_size[1]))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(Convolution2D(nb_filters, kernel_size[0], kernel_size[1]))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=pool_size))\n",
    "    model.add(Dropout(0.25))\n",
    "\n",
    "    model.add(Flatten())\n",
    "    model.add(Dense(128))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(Dropout(0.5))\n",
    "    model.add(Dense(nb_classes))\n",
    "    model.add(Activation('softmax'))\n",
    "    return model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# CPU"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "with tf.device('/cpu:0'):\n",
    "    model_cpu = make_model()\n",
    "\n",
    "model_cpu.compile(loss='categorical_crossentropy',\n",
    "              optimizer='adadelta',\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 50000 samples, validate on 10000 samples\n",
      "Epoch 1/3\n",
      "50000/50000 [==============================] - 235s - loss: 0.7324 - acc: 0.7780 - val_loss: 0.4591 - val_acc: 0.8679\n",
      "Epoch 2/3\n",
      "50000/50000 [==============================] - 234s - loss: 0.4728 - acc: 0.8623 - val_loss: 0.4107 - val_acc: 0.8789\n",
      "Epoch 3/3\n",
      "50000/50000 [==============================] - 234s - loss: 0.4246 - acc: 0.8744 - val_loss: 0.3865 - val_acc: 0.8869\n"
     ]
    }
   ],
   "source": [
    "t0 = time.time()\n",
    "history = model_cpu.fit(x_train, y_train,\n",
    "                    batch_size=batch_size, nb_epoch=nb_epoch,\n",
    "                    verbose=1, validation_data=(x_valid, y_valid))\n",
    "t1 = time.time()\n",
    "\n",
    "total_cpu = t1-t0"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# GPU"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "with tf.device('/gpu:0'):\n",
    "    model_gpu = make_model()\n",
    "    \n",
    "model_gpu.compile(loss='categorical_crossentropy',\n",
    "              optimizer='adadelta',\n",
    "              metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 50000 samples, validate on 10000 samples\n",
      "Epoch 1/3\n",
      "50000/50000 [==============================] - 32s - loss: 0.7178 - acc: 0.7836 - val_loss: 0.4525 - val_acc: 0.8674\n",
      "Epoch 2/3\n",
      "50000/50000 [==============================] - 31s - loss: 0.4773 - acc: 0.8586 - val_loss: 0.4078 - val_acc: 0.8817\n",
      "Epoch 3/3\n",
      "50000/50000 [==============================] - 31s - loss: 0.4263 - acc: 0.8737 - val_loss: 0.3859 - val_acc: 0.8868\n"
     ]
    }
   ],
   "source": [
    "t0 = time.time()\n",
    "history = model_gpu.fit(x_train, y_train,\n",
    "                    batch_size=batch_size, nb_epoch=nb_epoch,\n",
    "                    verbose=1, validation_data=(x_valid, y_valid))\n",
    "t1 = time.time()\n",
    "total_gpu = t1-t0\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU:\n",
      "Test score: 0.193013398772\n",
      "Test accuracy: 0.9468\n",
      "GPU:\n",
      "Test score: 0.196715823737\n",
      "Test accuracy: 0.9444\n"
     ]
    }
   ],
   "source": [
    "# Ensure they got the same (modulo random noise) results\n",
    "def eval_model(model):\n",
    "    score = model.evaluate(x_test, y_test, verbose=0)\n",
    "    print('Test score:', score[0])\n",
    "    print('Test accuracy:', score[1])\n",
    "    \n",
    "print(\"CPU:\")\n",
    "eval_model(model_cpu)\n",
    "print(\"GPU:\")\n",
    "eval_model(model_gpu)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Times:\n",
      "\tCPU: 704.2s\n",
      "\tGPU: 96.6s\n"
     ]
    }
   ],
   "source": [
    "print(\"Times:\\n\\tCPU: {:.1f}s\\n\\tGPU: {:.1f}s\".format(total_cpu, total_gpu))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAhMAAAFkCAYAAACabLnAAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nRWFCREREvKIwISIiIl75fyqKEFDgTvklAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fca245b3510>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.set_context(\"notebook\", font_scale=2)\n",
    "\n",
    "fig, ax = plt.subplots()\n",
    "ax.bar([0,1], [total_cpu, total_gpu])\n",
    "ax.set_ylabel('seconds')\n",
    "sns.despine(fig)\n",
    "ax.set_xticks([0.4, 1.4])\n",
    "ax.set_xticklabels(['CPU','GPU']);\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This is just an example with a small model and our tiny 28x28 images. More complex models can see an even larger difference."
   ]
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python [default]",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}