notebooks/neural_networks/birnn.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Handwritten Digit Classfication using Bidirectional Recurrent Neural Network"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The tutorial presented will again tackle the MNIST digit classification problem. You can find more information from the previous [MNIST introduction](introduction_to_mnist.ipynb) tutorial. This time, we will build a Bidirectional Recurrent Neural Network. Bidirectional RNNs are the extensions of RNNs which are based on the idea that **the output at time *t* may not only depend on the previous elements in the sequence, but latter elements as well**. **They are just two RNNs stacked on top of each other**(as seen in the following diagram). The output is then computed based on the hidden state of both RNNs."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "![Bi-d RNN](tutorial_images/Bi-directional_RNN/Bi-directional_RNN.jpg)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Populating the interactive namespace from numpy and matplotlib\n"
     ]
    }
   ],
   "source": [
    "import matplotlib\n",
    "matplotlib.use('Agg')\n",
    "%pylab inline\n",
    "\n",
    "import pandas\n",
    "import datetime as dt\n",
    "\n",
    "from bigdl.nn.layer import *\n",
    "from bigdl.nn.criterion import *\n",
    "from bigdl.optim.optimizer import *\n",
    "from bigdl.util.common import *\n",
    "from bigdl.dataset.transformer import *\n",
    "from bigdl.dataset import mnist\n",
    "from utils import get_mnist\n",
    "from pyspark import SparkContext\n",
    "from matplotlib.pyplot import imshow\n",
    "import matplotlib.pyplot as plt\n",
    "sc=SparkContext.getOrCreate(conf=create_spark_conf().setMaster(\"local[4]\").set(\"spark.driver.memory\",\"2g\"))\n",
    "\n",
    "init_engine()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1. Load MNIST dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First, we should get and store MNIST into RDD of Sample."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Note: *edit the \"mnist_path\" accordingly. If the \"mnist_path\" directory does not consist of the mnist data, mnist.read_data_sets method will download the dataset directly to the directory*."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "('Extracting', 'datasets/mnist/train-images-idx3-ubyte.gz')\n",
      "('Extracting', 'datasets/mnist/train-labels-idx1-ubyte.gz')\n",
      "('Extracting', 'datasets/mnist/t10k-images-idx3-ubyte.gz')\n",
      "('Extracting', 'datasets/mnist/t10k-labels-idx1-ubyte.gz')\n",
      "60000\n",
      "10000\n"
     ]
    }
   ],
   "source": [
    "# Get and store MNIST into RDD of Sample, please edit the \"mnist_path\" accordingly.\n",
    "mnist_path = \"datasets/mnist\"\n",
    "(train_data, test_data) = get_mnist(sc, mnist_path)\n",
    "\n",
    "train_data = train_data.map(lambda s: Sample.from_ndarray(np.resize(s.features[0].to_ndarray(), (28, 28)), s.label.to_ndarray()))\n",
    "test_data = test_data.map(lambda s: Sample.from_ndarray(np.resize(s.features[0].to_ndarray(), (28, 28)), s.label.to_ndarray()))\n",
    "print(train_data.count())\n",
    "print(test_data.count())"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2. Hyperparamter setup"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Parameters\n",
    "batch_size = 64\n",
    "\n",
    "# Network Parameters\n",
    "n_input = 28 # MNIST data input (img shape: 28*28)\n",
    "n_hidden = 128 # hidden layer num of features\n",
    "n_classes = 10 # MNIST total classes (0-9 digits)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 3. Model creation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's define our Bi-directional RNN model here. Find more particular information about BiRecurrent() [here](https://bigdl-project.github.io/master/#APIGuide/Layers/Recurrent-Layers/#birecurrent)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "creating: createSequential\n",
      "creating: createJoinTable\n",
      "creating: createBiRecurrent\n",
      "creating: createLSTM\n",
      "creating: createInferReshape\n",
      "creating: createSelect\n",
      "creating: createLinear\n"
     ]
    }
   ],
   "source": [
    "def build_model(input_size, hidden_size, output_size):\n",
    "    model = Sequential()\n",
    "    recurrent = BiRecurrent(JoinTable(3, 3))\n",
    "    recurrent.add(LSTM(input_size, hidden_size))\n",
    "    model.add(InferReshape([-1, input_size], True))\n",
    "    model.add(recurrent)\n",
    "    model.add(Select(2, -1))\n",
    "    model.add(Linear(2*hidden_size, output_size))\n",
    "    return model\n",
    "rnn_model = build_model(n_input, n_hidden, n_classes)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 4. Optimizer Setup"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's create an optimizer for training. As presented in the code, we are trying to optimize a [CrossEntropyCriterion](https://bigdl-project.github.io/master/#APIGuide/Losses/#crossentropycriterion) and use [Adam](https://bigdl-project.github.io/master/#APIGuide/Optimizers/Optim-Methods/#adam) to update the weights. Also in order to enable visualization support, we need to [generate summary info in BigDL](https://bigdl-project.github.io/master/#ProgrammingGuide/visualization/)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "creating: createCrossEntropyCriterion\n",
      "creating: createAdam\n",
      "creating: createMaxEpoch\n",
      "creating: createOptimizer\n",
      "creating: createEveryEpoch\n",
      "creating: createTop1Accuracy\n",
      "creating: createTrainSummary\n",
      "creating: createSeveralIteration\n",
      "creating: createValidationSummary\n",
      "saving logs to  rnn-20171221-170110\n"
     ]
    }
   ],
   "source": [
    "# Create an Optimizer\n",
    "\n",
    "criterion = CrossEntropyCriterion()\n",
    "optimizer = Optimizer(\n",
    "    model=rnn_model,\n",
    "    training_rdd=train_data,\n",
    "    criterion=criterion,\n",
    "    optim_method=Adam(),\n",
    "    end_trigger=MaxEpoch(5),\n",
    "    batch_size=batch_size)\n",
    "\n",
    "# Set the validation logic\n",
    "optimizer.set_validation(\n",
    "    batch_size=batch_size,\n",
    "    val_rdd=test_data,\n",
    "    trigger=EveryEpoch(),\n",
    "    val_method=[Top1Accuracy()]\n",
    ")\n",
    "\n",
    "app_name='rnn-'+dt.datetime.now().strftime(\"%Y%m%d-%H%M%S\")\n",
    "train_summary = TrainSummary(log_dir='/tmp/bigdl_summaries',\n",
    "                                     app_name=app_name)\n",
    "train_summary.set_summary_trigger(\"Parameters\", SeveralIteration(50))\n",
    "val_summary = ValidationSummary(log_dir='/tmp/bigdl_summaries',\n",
    "                                        app_name=app_name)\n",
    "optimizer.set_train_summary(train_summary)\n",
    "optimizer.set_val_summary(val_summary)\n",
    "print(\"saving logs to \",app_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Optimization Done.\n",
      "CPU times: user 40 ms, sys: 16 ms, total: 56 ms\n",
      "Wall time: 8min 37s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "# Boot training process\n",
    "trained_model = optimizer.optimize()\n",
    "print(\"Optimization Done.\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 5. Loss visualization"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "After training, we can draw the preformance curves from the previous `train_summary` and `val_summary` variables."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAtMAAAK7CAYAAAAwUwMGAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nJUmSpIZ+AxS4bYdND+HIAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8cdf7e4d90>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "loss = np.array(train_summary.read_scalar(\"Loss\"))\n",
    "top1 = np.array(val_summary.read_scalar(\"Top1Accuracy\"))\n",
    "\n",
    "plt.figure(figsize = (12,12))\n",
    "plt.subplot(2,1,1)\n",
    "plt.plot(loss[:,0],loss[:,1],label='loss')\n",
    "plt.xlim(0,loss.shape[0]+10)\n",
    "plt.grid(True)\n",
    "plt.title(\"loss\")\n",
    "plt.subplot(2,1,2)\n",
    "plt.plot(top1[:,0],top1[:,1],label='top1')\n",
    "plt.xlim(0,loss.shape[0]+10)\n",
    "plt.title(\"top1 accuracy\")\n",
    "plt.grid(True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 6. Prediction on test data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now, let's see the prediction results on test data by our trained model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def map_predict_label(l):\n",
    "    return np.array(l).argmax()\n",
    "def map_groundtruth_label(l):\n",
    "    return int(l[0] - 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Ground Truth labels:\n",
      "7, 2, 1, 0, 4, 1, 4, 9\n",
      "Predicted labels:\n",
      "7, 2, 1, 0, 4, 1, 4, 9\n",
      "CPU times: user 60 ms, sys: 4 ms, total: 64 ms\n",
      "Wall time: 3.48 s\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAABMCAYAAAB9PUwnAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\ngC6EEDYhAV0IIWxCAroQQtjE/wFIye0NfFiqDQAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8cdf7f49d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%%time\n",
    "predictions = trained_model.predict(test_data)\n",
    "imshow(np.column_stack([np.array(s.features[0].to_ndarray()).reshape(28,28) for s in test_data.take(8)]),cmap='gray'); plt.axis('off')\n",
    "print('Ground Truth labels:')\n",
    "print(', '.join(str(map_groundtruth_label(s.label.to_ndarray())) for s in test_data.take(8)))\n",
    "print('Predicted labels:')\n",
    "print(', '.join(str(map_predict_label(s)) for s in predictions.take(8)))"
   ]
  }
 ],
 "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": "2.7.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}