ideep-trial.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Results\n",
    "1. MLP w/o ideep\n",
    "    - 18.46 (10 epoch)\n",
    "2. MLP w/ ideep\n",
    "    - 25.65 (10 epoch)\n",
    "3. CNN w/o ideep\n",
    "    - 72.0849 (1 epoch)\n",
    "4. CNN w/ ideep\n",
    "    - 23.4223 (1 epoch)\n",
    "5. CNN w/ GPU\n",
    "    - 14.2982 (1 epoch)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import chainer\n",
    "import chainer.links as L\n",
    "import chainer.functions as F\n",
    "from chainer.cuda import to_gpu, to_cpu\n",
    "from chainer.datasets import mnist, split_dataset_random\n",
    "from chainer.optimizers import Adam\n",
    "from chainer.iterators import SerialIterator\n",
    "from chainer.training import StandardUpdater, Trainer, extensions\n",
    "from chainer.training.extensions import LogReport, PrintReport\n",
    "from chainer.backends.intel64 import is_ideep_available"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "GPU_ID = -1\n",
    "BATCH_SIZE = 16\n",
    "EPOCH = 10"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "if GPU_ID >= 0:\n",
    "    pass\n",
    "elif GPU_ID == -2 and is_ideep_available():\n",
    "    chainer.global_config.use_ideep = 'always'\n",
    "else:\n",
    "    chainer.global_config.use_ideep = 'never'\n",
    "\n",
    "chainer.config.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "class MLP(chainer.Chain):\n",
    "    def __init__(self, n_mid_units=16, n_out=10):\n",
    "        super(MLP, self).__init__()\n",
    "        with self.init_scope():\n",
    "            self.l1 = L.Linear(None, n_mid_units)\n",
    "            self.l2 = L.Linear(None, n_mid_units)\n",
    "            self.l3 = L.Linear(None, n_out)\n",
    "\n",
    "    def __call__(self, x, t):\n",
    "        h1 = F.relu(self.l1(x))\n",
    "        h2 = F.relu(self.l2(h1))\n",
    "        y = self.l3(h2)\n",
    "        loss = F.softmax_cross_entropy(y, t)\n",
    "        return loss\n",
    "\n",
    "\n",
    "class CNN(chainer.Chain):\n",
    "    def __init__(self, n_mid_channels=16, n_out=10):\n",
    "        super(CNN, self).__init__()\n",
    "        with self.init_scope():\n",
    "            self.conv1 = L.Convolution2D(None, n_mid_channels, 3)\n",
    "            self.conv2 = L.Convolution2D(None, n_mid_channels, 3)\n",
    "            self.l1 = L.Linear(None, n_out)\n",
    "\n",
    "\n",
    "    def __call__(self, x, t):\n",
    "        h = F.relu(self.conv1(x))\n",
    "        h = F.relu(self.conv2(h))\n",
    "        y = self.l1(h)\n",
    "        loss = F.softmax_cross_entropy(y, t)\n",
    "        return loss"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "train, _ = mnist.get_mnist(ndim=3)\n",
    "train_iter = SerialIterator(train, BATCH_SIZE)\n",
    "\n",
    "# model = MLP()\n",
    "model = CNN()\n",
    "if GPU_ID >= 0:\n",
    "    model.to_gpu(GPU_ID)\n",
    "elif GPU_ID == -2:\n",
    "    model.to_intel64()\n",
    "optimizer = Adam().setup(model)\n",
    "\n",
    "updater = StandardUpdater(train_iter, optimizer, device=GPU_ID)\n",
    "\n",
    "trainer = Trainer(updater, (EPOCH, 'epoch'))\n",
    "trainer.extend(LogReport(trigger=(1, 'epoch')))\n",
    "trainer.extend(PrintReport(entries=['epoch', 'elapsed_time']))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false,
    "scrolled": false
   },
   "outputs": [],
   "source": [
    "trainer.run()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python [conda env:py36-chainer4]",
   "language": "python",
   "name": "conda-env-py36-chainer4-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}