theano/cnn/lasagne-cifar10-inception-example.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": false
   },
   "source": [
    "The goal of this notebook is to have a completly self contained demo (training from scratch) for state of the art performance on CIFAR-10 to serve as a baseline. Sadly it is only at about 84% testing accuracy.\n",
    "\n",
    "Here I ported the MXNet Inception V3 CNN to Lasagne and Theano. It shuffles batches and stops after a fixed number of iterations.\n",
    "\n",
    "To run this demo you need to install theano and lasagne. As of right now the BatchNormLayer does not exist in the standard pip package so the development version is needed.\n",
    "\n",
    "    pip install theano\n",
    "    pip install lasagne\n",
    "    pip install --upgrade https://github.com/Lasagne/Lasagne/archive/master.zip\n",
    "\n",
    "This is based on a bunch of other code:\n",
    "https://github.com/jseppanen/cifar_lasagne/blob/master/cifar_lasagne.ipynb\n",
    "https://github.com/dmlc/mxnet/blob/master/example/image-classification/symbol_inception-bn-28-small.py\n",
    "\n",
    "To convert this notebook into a python script run `jupyter nbconvert --to python lasagne-cifar10-inception-example.ipynb`\n",
    "\n",
    "By Joseph Paul Cohen 2016\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "theano 0.8.2.dev-RELEASE\n",
      "lasagne 0.2.dev1\n"
     ]
    }
   ],
   "source": [
    "import theano, theano.tensor as T\n",
    "import numpy as np, matplotlib.pyplot as plt\n",
    "import os, time, random, sys\n",
    "notebook_start_time = time.time()\n",
    "\n",
    "import lasagne\n",
    "from lasagne.layers.normalization import BatchNormLayer\n",
    "from lasagne.layers import Conv2DLayer, InputLayer, ConcatLayer\n",
    "from lasagne.layers import DenseLayer, Pool2DLayer, FlattenLayer\n",
    "\n",
    "print \"theano\",theano.version.full_version\n",
    "print \"lasagne\",lasagne.__version__"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "#Set seed for random numbers:\n",
    "np.random.seed(1234)\n",
    "lasagne.random.set_rng(np.random.RandomState(1234))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "--2016-11-19 01:22:21--  http://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz\n",
      "Resolving www.cs.toronto.edu (www.cs.toronto.edu)... 128.100.3.30\n",
      "Connecting to www.cs.toronto.edu (www.cs.toronto.edu)|128.100.3.30|:80... connected.\n",
      "HTTP request sent, awaiting response... 416 Requested Range Not Satisfiable\n",
      "\n",
      "    The file is already fully retrieved; nothing to do.\n",
      "\n",
      "cifar-10-batches-py/\n",
      "cifar-10-batches-py/data_batch_4\n",
      "cifar-10-batches-py/readme.html\n",
      "cifar-10-batches-py/test_batch\n",
      "cifar-10-batches-py/data_batch_3\n",
      "cifar-10-batches-py/batches.meta\n",
      "cifar-10-batches-py/data_batch_2\n",
      "cifar-10-batches-py/data_batch_5\n",
      "cifar-10-batches-py/data_batch_1\n"
     ]
    }
   ],
   "source": [
    "##Uncomment these lines to fetch the dataset\n",
    "!wget -c http://www.cs.toronto.edu/~kriz/cifar-10-python.tar.gz\n",
    "!tar -xvf cifar-10-python.tar.gz"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "x_train.shape: (50000, 3072)\n"
     ]
    }
   ],
   "source": [
    "data_dir_cifar10 = os.path.join(\".\", \"cifar-10-batches-py\")\n",
    "\n",
    "def one_hot(x, n):\n",
    "    x = np.array(x)\n",
    "    assert x.ndim == 1\n",
    "    return np.eye(n)[x]\n",
    "\n",
    "def _load_batch_cifar10(filename, dtype='float32'):\n",
    "    path = os.path.join(data_dir_cifar10, filename)\n",
    "    batch = np.load(path)\n",
    "    data = batch['data'] / 255.0 # scale between [0, 1]\n",
    "    labels = one_hot(batch['labels'], n=10) # convert labels to one-hot representation\n",
    "    return data.astype(dtype), labels.astype(dtype)\n",
    "\n",
    "def cifar10(dtype='float32', grayscale=True):\n",
    "    x_train = []\n",
    "    t_train = []\n",
    "    for k in xrange(5):\n",
    "        x, t = _load_batch_cifar10(\"data_batch_%d\" % (k + 1), dtype=dtype)\n",
    "        x_train.append(x)\n",
    "        t_train.append(t)\n",
    "\n",
    "    x_train = np.concatenate(x_train, axis=0)\n",
    "    t_train = np.concatenate(t_train, axis=0)\n",
    "\n",
    "    x_test, t_test = _load_batch_cifar10(\"test_batch\", dtype=dtype)\n",
    "\n",
    "    if grayscale:\n",
    "        x_train = _grayscale(x_train)\n",
    "        x_test = _grayscale(x_test)\n",
    "\n",
    "    return x_train, t_train, x_test, t_test\n",
    "\n",
    "# load data\n",
    "x_train, t_train, x_test, t_test = cifar10(dtype=theano.config.floatX,grayscale=False)\n",
    "labels_test = np.argmax(t_test, axis=1)\n",
    "\n",
    "print \"x_train.shape:\",x_train.shape\n",
    "\n",
    "# reshape data\n",
    "x_train = x_train.reshape((x_train.shape[0], 3, 32, 32))\n",
    "x_test = x_test.reshape((x_test.shape[0], 3, 32, 32))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7f007a831250>"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAREAAAETCAYAAAARXDKFAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f007cb32e50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline\n",
    "plt.rcParams['figure.figsize'] = (10, 3)\n",
    "\n",
    "a  = np.rollaxis(np.rollaxis(x_train[0],2,0),2,0)\n",
    "plt.imshow(a,interpolation='none')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "input_var = T.tensor4('inputs')\n",
    "\n",
    "input_shape = (None, 3, 32, 32)\n",
    "network = InputLayer(shape=input_shape, input_var=input_var)\n",
    "\n",
    "# from MXNet \n",
    "def ConvFactory(data, num_filter, filter_size, stride=(1,1), pad=(0, 0)):\n",
    "    data = BatchNormLayer(Conv2DLayer(\n",
    "        data, num_filters=num_filter,\n",
    "        filter_size=filter_size,\n",
    "        stride=stride, pad=pad,\n",
    "        nonlinearity=lasagne.nonlinearities.rectify,\n",
    "        W=lasagne.init.GlorotUniform(gain='relu')))\n",
    "    return data\n",
    "\n",
    "def DownsampleFactory(data, ch_3x3):\n",
    "    conv = ConvFactory(data=data, filter_size=(3, 3), stride=(2, 2), num_filter=ch_3x3, pad=(1, 1))\n",
    "    pool = Pool2DLayer(data, pool_size=3, stride=(2, 2), pad=(1, 1), mode='max')\n",
    "    concat = ConcatLayer([conv, pool])\n",
    "    return concat\n",
    "\n",
    "def SimpleFactory(data, ch_1x1, ch_3x3):\n",
    "    conv1x1 = ConvFactory(data=data, filter_size=(1, 1), pad=(0, 0), num_filter=ch_1x1)\n",
    "    conv3x3 = ConvFactory(data=data, filter_size=(3, 3), pad=(1, 1), num_filter=ch_3x3) \n",
    "    concat = ConcatLayer([conv1x1, conv3x3])\n",
    "    return concat\n",
    "        \n",
    "def get_symbol(network, num_classes = 10):\n",
    "    conv1 = ConvFactory(data=network, filter_size=(3,3), pad=(1,1), num_filter=96)\n",
    "    in3a = SimpleFactory(conv1, 32, 32)\n",
    "    in3b = SimpleFactory(in3a, 32, 48)\n",
    "    in3c = DownsampleFactory(in3b, 80)\n",
    "    in4a = SimpleFactory(in3c, 112, 48)\n",
    "    in4b = SimpleFactory(in4a, 96, 64)\n",
    "    in4c = SimpleFactory(in4b, 80, 80)\n",
    "    in4d = SimpleFactory(in4c, 48, 96)\n",
    "    in4e = DownsampleFactory(in4d, 96)\n",
    "    in5a = SimpleFactory(in4e, 176, 160)\n",
    "    in5b = SimpleFactory(in5a, 176, 160)\n",
    "    pool = Pool2DLayer(in5b, pool_size=7, stride=1, pad=1, mode='average_exc_pad')\n",
    "    \n",
    "    flatten = FlattenLayer(pool)\n",
    "    dense = DenseLayer(flatten, num_units=num_classes, nonlinearity=lasagne.nonlinearities.rectify)\n",
    "    out = DenseLayer(dense, num_units=num_classes, nonlinearity=lasagne.nonlinearities.softmax)\n",
    "    \n",
    "    return out\n",
    "\n",
    "network = get_symbol(network)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "input_shape: (None, 3, 32, 32) -> output_shape: (None, 10)\n"
     ]
    }
   ],
   "source": [
    "output_shape = lasagne.layers.get_output_shape(network)\n",
    "print \"input_shape:\",input_shape,\"-> output_shape:\",output_shape\n",
    "sys.stdout.flush()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "DONE building output functions\n"
     ]
    }
   ],
   "source": [
    "#Output functions\n",
    "output = lasagne.layers.get_output(network)\n",
    "y = T.argmax(output, axis=1)\n",
    "f_predict = theano.function([input_var], y)\n",
    "print \"DONE building output functions\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "DONE building training functions\n"
     ]
    }
   ],
   "source": [
    "#Training function\n",
    "target_var = T.matrix('target')\n",
    "loss = lasagne.objectives.squared_error(output, target_var)\n",
    "loss = loss.mean()\n",
    "\n",
    "lr = theano.shared(np.array(0, dtype=theano.config.floatX))\n",
    "params = lasagne.layers.get_all_params(network, trainable=True)\n",
    "updates = lasagne.updates.adam(loss, params, learning_rate=lr)\n",
    "#updates = lasagne.updates.nesterov_momentum(loss, params, learning_rate=lr, momentum=0.9)\n",
    "\n",
    "f_train = theano.function([input_var, target_var], loss, updates=updates)\n",
    "print \"DONE building training functions\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "batch_size: 512\n",
      "lr reduced: 0.000900000042748\n",
      "iteration: 0, trainerror: 0.06596, accuracy: 0.60590, seconds: 129\n",
      "iteration: 1, trainerror: 0.04510, accuracy: 0.70870, seconds: 131\n",
      "iteration: 2, trainerror: 0.03566, accuracy: 0.75910, seconds: 131\n",
      "iteration: 3, trainerror: 0.02800, accuracy: 0.77980, seconds: 131\n",
      "iteration: 4, trainerror: 0.02283, accuracy: 0.80120, seconds: 131\n",
      "lr reduced: 0.000810000055935\n",
      "iteration: 5, trainerror: 0.01832, accuracy: 0.81340, seconds: 131\n",
      "iteration: 6, trainerror: 0.01457, accuracy: 0.81830, seconds: 131\n",
      "iteration: 7, trainerror: 0.01209, accuracy: 0.82250, seconds: 131\n",
      "iteration: 8, trainerror: 0.01075, accuracy: 0.82860, seconds: 131\n",
      "iteration: 9, trainerror: 0.00894, accuracy: 0.82930, seconds: 131\n",
      "lr reduced: 0.000729000079446\n",
      "iteration: 10, trainerror: 0.00689, accuracy: 0.83980, seconds: 131\n",
      "iteration: 11, trainerror: 0.00527, accuracy: 0.83270, seconds: 131\n",
      "iteration: 12, trainerror: 0.00439, accuracy: 0.83600, seconds: 132\n",
      "iteration: 13, trainerror: 0.00407, accuracy: 0.83690, seconds: 131\n",
      "iteration: 14, trainerror: 0.00410, accuracy: 0.83910, seconds: 131\n",
      "lr reduced: 0.000656100048218\n",
      "iteration: 15, trainerror: 0.00331, accuracy: 0.84430, seconds: 131\n",
      "iteration: 16, trainerror: 0.00248, accuracy: 0.84120, seconds: 131\n",
      "iteration: 17, trainerror: 0.00230, accuracy: 0.84440, seconds: 131\n",
      "iteration: 18, trainerror: 0.00218, accuracy: 0.84160, seconds: 131\n",
      "iteration: 19, trainerror: 0.00242, accuracy: 0.84230, seconds: 131\n"
     ]
    }
   ],
   "source": [
    "# train model\n",
    "lr.set_value(0.001)\n",
    "batch_size = 2**9\n",
    "print \"batch_size:\",batch_size\n",
    "\n",
    "stats = []\n",
    "for i in range(0,20):\n",
    "    start_time = time.time()\n",
    "    trainerror = []\n",
    "    \n",
    "    #Decrease learning rate!\n",
    "    if (i % 5) == 0:\n",
    "        lr.set_value((lr.get_value()*0.9).astype(theano.config.floatX))\n",
    "        print \"lr reduced:\", lr.get_value()\n",
    "    \n",
    "    #Shuffle batches!\n",
    "    order = range(0,len(x_train))\n",
    "    random.shuffle(order)\n",
    "    x_train = x_train[order]\n",
    "    t_train = t_train[order]\n",
    "    \n",
    "    #Start batch training!\n",
    "    for start in range(0, len(x_train), batch_size):\n",
    "        x_batch = x_train[start:start + batch_size]\n",
    "        t_batch = t_train[start:start + batch_size]\n",
    "        cost = f_train(x_batch, t_batch)\n",
    "        trainerror.append(cost)\n",
    "        #print \"batch error: %.5f\" % cost\n",
    "    elapsed_time = time.time() - start_time\n",
    "    \n",
    "    ## Calculate test error (just for logging)\n",
    "    pred = np.array([])\n",
    "    for start in range(0, len(x_test), batch_size):\n",
    "        pred = np.append(pred,f_predict(x_test[start:start + batch_size]))\n",
    "    testacc = np.mean(pred == labels_test)\n",
    "    \n",
    "    epocherror = np.mean(trainerror)\n",
    "    stats.append((testacc,epocherror))\n",
    "    \n",
    "    print \"iteration: %d, trainerror: %.5f, accuracy: %.5f, seconds: %d\" % (i, epocherror,testacc, elapsed_time)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7f0051c2a390>]"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA0UAAAEXCAYAAABiXu9gAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f00762254d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "statsnp = np.rollaxis(np.asarray(stats),0,2)\n",
    "plt.plot(statsnp[1])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false,
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[<matplotlib.lines.Line2D at 0x7f0051b64110>]"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA0UAAAEXCAYAAABiXu9gAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nJKlL+z8xgWMKQDPmAwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f0051d0f410>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(statsnp[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "('Max Testing Accuracy', 0.84440000000000004)"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "\"Max Testing Accuracy\", max(statsnp[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Notebook took 48.0419564486 minutes\n"
     ]
    }
   ],
   "source": [
    "notebook_elapsed_time = time.time() - notebook_start_time\n",
    "print \"Notebook took\", notebook_elapsed_time/60.0, \"minutes\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "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
}