VQA-gluon.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "# Visual Question Answering in MXNET Gluon"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "This is a notebook for implementing visual question answering using MXNET Gluon.\n",
    "\n",
    "This notebook is based on:\n",
    "\n",
    "<a href=\"https://arxiv.org/pdf/1606.01847.pdf\">\"Multimodal Compact Bilinear Pooling for Visual Question Answering and Visual Grounding\"</a>, Akira Fukui, Dong Huk Park, Daylen Yang, Anna Rohrbach, Trevor Darrell and Marcus Rohrbach, EMNLP 2016\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## Table of Content:\n",
    "* [VQA Task](#vqatask)\n",
    "* [VQA Dataset](#dataset)\n",
    "* [Feature Extraction](#feature)\n",
    "* [Pooling methods](#pooling)\n",
    "* [Models](#models)\n",
    "* [Data IO](#dataio)\n",
    "* [Training](#train)\n",
    "* [Testing](#test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## VQA Task <a class=\"anchor\" id=\"vqatask\"></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "Visual question answering(VQA) is a task focusing on providing a natural language answer given any image and any open-ended question. This task requires a deep understanding and reasoning of the image combined with the question: a joint representation of both visual and textual input."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "\n",
    "![](img/pizza.png )\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## VQA Dataset  <a class=\"anchor\" id=\"dataset\"></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "__VQA dataset v1__ is first released by Antol et. al in this <a href=\"https://arxiv.org/pdf/1505.00468v6.pdf\">paper</a>. Based on the type of images: real images and abstract scenes, they have two different dataset. However, due to the inherent structure in our world and bias in language, the learning process is biased. In other words, a specific question tends to have a specific answer regardless of the image. Thus, Goyal et. al. release <a href=\"https://arxiv.org/pdf/1505.00468v6.pdf\">VQA dataset v2</a>. They prepare similar images with same questions but leads to different answers.\n",
    "\n",
    "Johnson et. al. introduce Compositional Language and Elementary Visual Reasoning(__CLEVR__) diagnostic <a href=\"https://arxiv.org/abs/1612.06890\">dataset</a> which focuses more on reasoning. Strub et. al. propose a <a href=\"https://arxiv.org/abs/1611.08481\">__two-player guessing game__</a>: guess a target in a given image with a sequential questions and answers. This requires both visual question answering and spatial reasoning.\n",
    "\n",
    "The <a href=\"https://arxiv.org/pdf/1505.00468v6.pdf\">Task Driven Image Understanding Challenge dataset</a>(__TDIUC__) contains over 1.6 million questions organized into 12 different categories. It contains images and annotations from MSCOCO and Visual genome. The key differences between TDIUC and VQA v1/v2 dataset are: (1)Categorized questions: Each question belongs to one of the 12 categories. This helps us have a task-oriented evaluation. Besides simple accuracy, we can also calculate arithmetic and harmonic means across all per question-type accuracies. \n",
    "(2)Absurd questions: Questions that are totally irrelevant to the image. In this way, it force an algorithm to determine if a question is valid. This also helps to balance the dataset.\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "In this notebook, we will use VQA dataser v1 for demostration. You can download the VQA1.0 dataset from <a href=\"http://www.visualqa.org/vqa_v1_download.html\">VQA</a> website. \n",
    "\n",
    "You need to preprocess the data:\n",
    "\n",
    "(1) Extract the samples from original json files. \n",
    "\n",
    "(2) Filter the samples giving top k answers(k can be 1000, 2000...). This will make the prediction easier."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## Feature Extraction <a class=\"anchor\" id=\"feature\"></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "Usually people use pretrained models to extract features from the image and question.\n",
    "\n",
    "__Image pretrained model__: \n",
    "\n",
    "<a href=\"https://arxiv.org/abs/1409.1556\">VGG</a>: A key aspect of VGG was to use many convolutional blocks with relatively narrow kernels, followed by a max-pooling step and to repeat this block multiple times. \n",
    "\n",
    "<a href=\"https://arxiv.org/abs/1512.03385\">Resnet</a>: It is a residual learning framework to ease the training of networks that are substantially deep. It reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions.\n",
    "\n",
    "__Question pretrained model__: \n",
    "\n",
    "<a href=\"https://code.google.com/archive/p/word2vec/\">Word2Vec</a>: The word2vec tool takes a text corpus as input and produces the word vectors as output. It first constructs a vocabulary from the training text data and then learns vector representation of words. The model contains 300-dimensional vectors for 3 million words and phrases. \n",
    "\n",
    "<a href=\"https://nlp.stanford.edu/projects/glove/\">Glove</a>: Similar to Word2Vec, it is a word embedding dataset. It contains 100/200/300-dimensional vectors for 2 million words.\n",
    "\n",
    "<a href=\"https://arxiv.org/abs/1506.06726\">Skipthought</a>: This is an encoder-decoder model that tries to reconstruct the surrounding sentences of an encoded passage. Sentences that share semantic and syntactic properties are thus mapped to similar vector representations. Different from the previous two model, this is a sentence based model.\n",
    "\n",
    "<a href=\"https://research.google.com/pubs/pub45610.html\">GNMT encoder</a>: We propose using the encoder of google neural machine translation system to extract the question features. \n",
    "\n",
    "__We will discuss about how to extract the features <a href=\"extract-feature.ipynb\">here</a> in details.__"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## Pooling Methods <a class=\"anchor\" id=\"pooling\"></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "After we extract the features, information of each modality are then combined together, using concatenation, element-wise product or sum operations.\n",
    "\n",
    "Pooling methods are widely used in visual tasks to combine information for various streams into one final feature representation. It inherently can be applied to VQA task. Common pooling methods are average pooling and bilinear pooling. Bilinear pooling requires taking the outer product between two features. However the high dimensional result makes it hard to be applied to huge real image dataset. Thus, different compact pooling methods are proposed to solve this problem.\n",
    "\n",
    "In the following notebook, we will illustrate two different models:(1) Concatenation (2) Multimodel compact bilinear pooling.\n",
    "\n",
    "__We will introduce attention model in this <a href=\"Attention-VQA-gluon.ipynb\">notebook</a>.__"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## Models <a class=\"anchor\" id=\"models\"></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "We define our model with <a href=\"https://mxnet.incubator.apache.org/api/python/gluon.html\">MXNET Gluon</a>. gluon.Block is the basic building block of models. If any operator is not defined under gluon, you can use mxnet.ndarray operators to subsititude. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "from __future__ import print_function\n",
    "import numpy as np\n",
    "import mxnet as mx\n",
    "import mxnet.ndarray as F\n",
    "import mxnet.contrib.ndarray as C\n",
    "import mxnet.gluon as gluon\n",
    "from mxnet.gluon import nn\n",
    "from mxnet import autograd\n",
    "import bisect\n",
    "from IPython.core.display import display, HTML\n",
    "import logging\n",
    "logging.basicConfig(level=logging.INFO)\n",
    "import os\n",
    "from mxnet.test_utils import download\n",
    "import json\n",
    "from IPython.display import HTML, display"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "# Some parameters we are going to use\n",
    "batch_size = 64\n",
    "#ctx = mx.gpu(1)\n",
    "ctx = mx.cpu()\n",
    "compute_size  = batch_size\n",
    "out_dim = 10000\n",
    "gpus = 1"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "In the __first model__, we will concatenate the image and question features and use multilayer perception(MLP) to predict the answer."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "class Net1(gluon.Block):\n",
    "    def __init__(self, **kwargs):\n",
    "        super(Net1, self).__init__(**kwargs)\n",
    "        with self.name_scope():\n",
    "            # layers created in name_scope will inherit name space\n",
    "            # from parent layer.\n",
    "            self.bn = nn.BatchNorm()\n",
    "            self.dropout = nn.Dropout(0.3)\n",
    "            self.fc1 = nn.Dense(8192,activation=\"relu\")\n",
    "            self.fc2 = nn.Dense(1000)\n",
    "            \n",
    "\n",
    "    def forward(self, x):\n",
    "        x1 = F.L2Normalization(x[0])\n",
    "        x2 = F.L2Normalization(x[1])\n",
    "        z = F.concat(x1,x2,dim=1)\n",
    "        z = self.fc1(z)\n",
    "        z = self.bn(z)\n",
    "        z = self.dropout(z)\n",
    "        z = self.fc2(z)\n",
    "        return z\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "In the __second model__, instead of linearly combine the image and text features, we use count sketch to estimate the outer product of the image and question features. It is also named as multimodel compact bilinear pooling(MCB). This method was proposed in <a href=\"https://arxiv.org/abs/1606.01847\">Multimodal Compact Bilinear Pooling for VQA</a>. \n",
    "\n",
    "Given a vector $a \\in \\mathcal{R}^n$, random hash function $h \\in \\mathcal{R}^n$: $[n] \\to [b]$ and binary variable $s \\in \\mathcal{R}^n$: $[n] \\to \\pm 1$, the count sketch operator $\\psi(a,h,s) \\in \\mathcal{R}^b$ is:\n",
    "\n",
    "$\\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\psi(a,h,s)[j] = \\sum_{h[i] = j}s[i]a[i], \\quad j \\in {1,\\cdots,b}$\n",
    "\n",
    "Let $x$ and $y$ be two separate feature vectors, and their bilinear pooling feature be $x \\otimes y$, then:\n",
    "\n",
    "$\\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\quad \\psi(x \\otimes y, h,s) = \\psi(x,h,s) \\star \\psi(y,h,s)$\n",
    "\n",
    "where $\\star$ is the convolution operator. This can further be simplified by using FFT properties: convolution in time domain equals to elementwise product in frequency domain. We show the procedure below:"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "\n",
    "![](img/mcb.png )\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "One improvement we made is adding ones vectors to each features before count sketch. The intuition is: given input vectors $x,y$, estimating outer product between $[x,1s]$ and $[y, 1s]$ gives us information more than just $x \\otimes y$. It also contains information of $x$ and $y$."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "class Net2(gluon.Block):\n",
    "    def __init__(self, **kwargs):\n",
    "        super(Net2, self).__init__(**kwargs)\n",
    "        with self.name_scope():\n",
    "            # layers created in name_scope will inherit name space\n",
    "            # from parent layer.\n",
    "            self.bn = nn.BatchNorm()\n",
    "            self.dropout = nn.Dropout(0.3)\n",
    "            self.fc1 = nn.Dense(8192,activation=\"relu\")\n",
    "            self.fc2 = nn.Dense(1000)\n",
    "            \n",
    "\n",
    "    def forward(self, x):\n",
    "        x1 = F.L2Normalization(x[0])\n",
    "        x2 = F.L2Normalization(x[1])\n",
    "        text_ones = F.ones((batch_size/gpus, 2048),ctx = ctx)\n",
    "        img_ones = F.ones((batch_size/gpus, 1024),ctx = ctx)\n",
    "        text_data = F.Concat(x1, text_ones,dim = 1)\n",
    "        image_data = F.Concat(x2,img_ones,dim = 1)\n",
    "        # Initialize hash tables\n",
    "        # Random seed to ensure we have same hash function for all samples\n",
    "        np.random.seed(0)\n",
    "        S1 = F.array(np.random.randint(0, 2, (1,3072))*2-1,ctx = ctx)\n",
    "        np.random.seed(0)\n",
    "        H1 = F.array(np.random.randint(0, out_dim,(1,3072)),ctx = ctx)\n",
    "        np.random.seed(1)\n",
    "        S2 = F.array(np.random.randint(0, 2, (1,3072))*2-1,ctx = ctx)\n",
    "        np.random.seed(1)\n",
    "        H2 = F.array(np.random.randint(0, out_dim,(1,3072)),ctx = ctx)\n",
    "        # Count sketch\n",
    "        cs1 = C.count_sketch( data = image_data, s=S1, h = H1 ,name='cs1',out_dim = out_dim) \n",
    "        cs2 = C.count_sketch( data = text_data, s=S2, h = H2 ,name='cs2',out_dim = out_dim) \n",
    "        fft1 = C.fft(data = cs1, name='fft1', compute_size = compute_size) \n",
    "        fft2 = C.fft(data = cs2, name='fft2', compute_size = compute_size) \n",
    "        c = fft1 * fft2\n",
    "        ifft1 = C.ifft(data = c, name='ifft1', compute_size = compute_size) \n",
    "        # MLP\n",
    "        z = self.fc1(ifft1)\n",
    "        z = self.bn(z)\n",
    "        z = self.dropout(z)\n",
    "        z = self.fc2(z)\n",
    "        return z\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## Data IO <a class=\"anchor\" id=\"dataio\"></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "MXNET requires a data iterator to feed in the samples to the model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "from VQAtrainIter import VQAtrainIter"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "The function is defined in __VQAtrainIter.py__.\n",
    "\n",
    "The inputs of the data iterator are extracted image and question features. At each step, the data iterator will return a data batch list: question data batch and image data batch. \n",
    "\n",
    "We need to seperate the data batches by the length of the input data because the input questions are in different lengths. The $buckets$ parameter defines the max length you want to keep in the data iterator. Here since we already used pretrained model to extract the question feature, the question length is fixed as the output of the pretrained model.\n",
    "\n",
    "The $layout$ parameter defines the layout of the data iterator output. \"N\" specify where is the data batch dimension is.\n",
    "\n",
    "$reset()$ function is called after every epoch. $next()$ function is call after each batch."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "Here we will use subset of VQA v1 dataset in this tutorial. We extract the image feature from ResNet-152, text feature from GNMT encoder. We have 21537 training samples and 1044 validation samples in this tutorial. Image feature is a 2048-dim vector. Question feature is a 1024-dim vector. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "# Download the dataset\n",
    "dataset_files = {'train': ('train_question.npz','train_img.npz','train_ans.npz'),\n",
    "                 'validation': ('val_question.npz','val_img.npz','val_ans.npz'),\n",
    "                 'test':('test_question_id.npz','test_question.npz','test_img_id.npz','test_img.npz','atoi.json','test_question_txt.json')}\n",
    "\n",
    "train_q, train_i, train_a = dataset_files['train']\n",
    "val_q, val_i, val_a = dataset_files['validation']\n",
    "\n",
    "url_format = 'https://apache-mxnet.s3-accelerate.amazonaws.com/gluon/dataset/VQA-notebook/{}'\n",
    "if not os.path.exists(train_q):\n",
    "    logging.info('Downloading training dataset.')\n",
    "    download(url_format.format(train_q),overwrite=True)\n",
    "    download(url_format.format(train_i),overwrite=True)\n",
    "    download(url_format.format(train_a),overwrite=True)\n",
    "if not os.path.exists(val_q):\n",
    "    logging.info('Downloading validation dataset.')\n",
    "    download(url_format.format(val_q),overwrite=True)\n",
    "    download(url_format.format(val_i),overwrite=True)\n",
    "    download(url_format.format(val_a),overwrite=True)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Total training sample: 21537\n",
      "Total validation sample: 1044\n",
      "WARNING: discarded 0 sentences longer than the largest bucket.\n",
      "WARNING: discarded 0 sentences longer than the largest bucket.\n"
     ]
    }
   ],
   "source": [
    "layout = 'NT'\n",
    "bucket = [1024]\n",
    "\n",
    "train_question = np.load(train_q)['x']\n",
    "val_question = np.load(val_q)['x']\n",
    "train_ans = np.load(train_a)['x']\n",
    "val_ans = np.load(val_a)['x']\n",
    "train_img = np.load(train_i)['x']\n",
    "val_img = np.load(val_i)['x']\n",
    "\n",
    "print(\"Total training sample:\",train_ans.shape[0])   \n",
    "print(\"Total validation sample:\",val_ans.shape[0])   \n",
    "\n",
    "data_train  = VQAtrainIter(train_img, train_question, train_ans, batch_size, buckets = bucket,layout=layout)\n",
    "data_eva = VQAtrainIter(val_img, val_question, val_ans, batch_size, buckets = bucket,layout=layout) \n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## Training  <a class=\"anchor\" id=\"train\"></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "We __initialize the parameters__, __define loss, evaluation metrics and optimizer__ as follows:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "net = Net1()\n",
    "#net = Net2()\n",
    "net.collect_params().initialize(mx.init.Xavier(), ctx=ctx)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "loss = gluon.loss.SoftmaxCrossEntropyLoss()\n",
    "\n",
    "metric = mx.metric.Accuracy()\n",
    "\n",
    "def evaluate_accuracy(data_iterator, net):\n",
    "    numerator = 0.\n",
    "    denominator = 0.\n",
    "    \n",
    "    data_iterator.reset()\n",
    "    for i, batch in enumerate(data_iterator):\n",
    "        with autograd.record():\n",
    "            data1 = batch.data[0].as_in_context(ctx)\n",
    "            data2 = batch.data[1].as_in_context(ctx)\n",
    "            data = [data1,data2]\n",
    "            label = batch.label[0].as_in_context(ctx)\n",
    "            output = net(data)\n",
    "        \n",
    "        metric.update([label], [output])\n",
    "    return metric.get()[1]\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "trainer = gluon.Trainer(net.collect_params(), 'sgd', {'learning_rate': 0.01})"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "After the initializations, we can start the training. It takes one batch a time. After we run over all batches(one epoch), we calculate the training/validation accuracy and save the network parameters if the validation accuracy is higher than the previous one. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:root:Best validation acc found. Checkpointing...\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 0. Loss: 2.94325249822, Train_acc 0.445534446023, Eval_acc 0.384765625\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:root:Best validation acc found. Checkpointing...\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1. Loss: 2.00672398202, Train_acc 0.483775745739, Eval_acc 0.442807404891\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:root:Best validation acc found. Checkpointing...\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 2. Loss: 1.64857072872, Train_acc 0.517578125, Eval_acc 0.481336805556\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:root:Best validation acc found. Checkpointing...\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 3. Loss: 1.33320239985, Train_acc 0.551225142045, Eval_acc 0.515741604478\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:root:Best validation acc found. Checkpointing...\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 4. Loss: 1.12062701052, Train_acc 0.582794744318, Eval_acc 0.549607180478\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:root:Best validation acc found. Checkpointing...\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 5. Loss: 1.05326851985, Train_acc 0.609456380208, Eval_acc 0.581116272523\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:root:Best validation acc found. Checkpointing...\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 6. Loss: 0.800970615745, Train_acc 0.632894937094, Eval_acc 0.607833059211\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:root:Best validation acc found. Checkpointing...\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 7. Loss: 0.730079422974, Train_acc 0.653958407315, Eval_acc 0.631401209677\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:root:Best validation acc found. Checkpointing...\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 8. Loss: 0.644547770061, Train_acc 0.672037760417, Eval_acc 0.652459613347\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:root:Best validation acc found. Checkpointing...\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 9. Loss: 0.537664141449, Train_acc 0.687650923295, Eval_acc 0.670662884736\n"
     ]
    }
   ],
   "source": [
    "epochs = 10\n",
    "moving_loss = 0.\n",
    "best_eva = 0\n",
    "for e in range(epochs):\n",
    "    data_train.reset()\n",
    "    for i, batch in enumerate(data_train):\n",
    "        data1 = batch.data[0].as_in_context(ctx)\n",
    "        data2 = batch.data[1].as_in_context(ctx)\n",
    "        data = [data1,data2]\n",
    "        label = batch.label[0].as_in_context(ctx)\n",
    "        with autograd.record():\n",
    "            output = net(data)\n",
    "            cross_entropy = loss(output, label)\n",
    "            cross_entropy.backward()\n",
    "        trainer.step(data[0].shape[0])\n",
    "        \n",
    "        ##########################\n",
    "        #  Keep a moving average of the losses\n",
    "        ##########################\n",
    "        if i == 0:\n",
    "            moving_loss = np.mean(cross_entropy.asnumpy()[0])\n",
    "        else:\n",
    "            moving_loss = .99 * moving_loss + .01 * np.mean(cross_entropy.asnumpy()[0])\n",
    "        #if i % 200 == 0:\n",
    "        #    print(\"Epoch %s, batch %s. Moving avg of loss: %s\" % (e, i, moving_loss))   \n",
    "    eva_accuracy = evaluate_accuracy(data_eva, net)\n",
    "    train_accuracy = evaluate_accuracy(data_train, net)\n",
    "    print(\"Epoch %s. Loss: %s, Train_acc %s, Eval_acc %s\" % (e, moving_loss, train_accuracy, eva_accuracy))\n",
    "    if eva_accuracy > best_eva:\n",
    "            best_eva = eva_accuracy\n",
    "            logging.info('Best validation acc found. Checkpointing...')\n",
    "            net.save_params('vqa-mlp-%d.params'%(e))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "## Testing  <a class=\"anchor\" id=\"test\"></a>"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "After the training loop over Net1 or Net2, we can try it on test data. Here we have 10 test samples. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "test = True\n",
    "if test:\n",
    "    test_q_id, test_q, test_i_id, test_i, atoi,text = dataset_files['test']\n",
    "\n",
    "if test and not os.path.exists(test_q):     \n",
    "    logging.info('Downloading test dataset.')\n",
    "    download(url_format.format(test_q_id),overwrite=True)\n",
    "    download(url_format.format(test_q),overwrite=True)\n",
    "    download(url_format.format(test_i_id),overwrite=True)\n",
    "    download(url_format.format(test_i),overwrite=True)\n",
    "    download(url_format.format(atoi),overwrite=True)\n",
    "    download(url_format.format(text),overwrite=True)\n",
    "\n",
    "if test:\n",
    "    test_question = np.load(\"test_question.npz\")['x']\n",
    "    test_img = np.load(\"test_img.npz\")['x']\n",
    "    test_question_id = np.load(\"test_question_id.npz\")['x']\n",
    "    test_img_id = np.load(\"test_img_id.npz\")['x']\n",
    "    #atoi = np.load(\"atoi.json\")['x']    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "We pass the test data iterator to the trained model. The output is a list of predicted answers. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING: discarded 0 sentences longer than the largest bucket.\n"
     ]
    }
   ],
   "source": [
    "data_test = VQAtrainIter(test_img, test_question, np.zeros((test_img.shape[0],1)), 10, buckets = bucket,layout=layout)\n",
    "for i, batch in enumerate(data_test):\n",
    "    with autograd.record():\n",
    "        data1 = batch.data[0].as_in_context(ctx)\n",
    "        data2 = batch.data[1].as_in_context(ctx)\n",
    "        data = [data1,data2]\n",
    "        #label = batch.label[0].as_in_context(ctx)\n",
    "        #label_one_hot = nd.one_hot(label, 10)\n",
    "        output = net(data)\n",
    "output = np.argmax(output.asnumpy(), axis = 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "We randomly select one sample to show the testing result."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Question: Is the bird looking at the camera?\n"
     ]
    }
   ],
   "source": [
    "idx = np.random.randint(10)\n",
    "question = json.load(open(text))\n",
    "print(\"Question:\", question[idx])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:root:Downloading training dataset.\n",
      "INFO:root:downloaded https://apache-mxnet.s3-accelerate.amazonaws.com/gluon/dataset/VQA-notebook/test_images/COCO_test2015_000000504864.jpg into COCO_test2015_000000504864.jpg successfully\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Image:\n"
     ]
    },
    {
     "data": {
      "image/jpeg": "/9j/4AAQSkZJRgABAQEASABIAAD/2wBDAAMCAgMCAgMDAwMEAwMEBQgFBQQEBQoHBwYIDAoMDAsK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\n",
      "text/plain": [
       "<IPython.core.display.Image object>"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "image_name = 'COCO_test2015_' + str(int(test_img_id[idx])).zfill(12)+'.jpg'\n",
    "if not os.path.exists(image_name):\n",
    "    logging.info('Downloading training dataset.')\n",
    "    download(url_format.format('test_images/'+image_name),overwrite=True)\n",
    "\n",
    "from IPython.display import Image\n",
    "print(\"Image:\")\n",
    "Image(filename=image_name) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Answer: no\n"
     ]
    }
   ],
   "source": [
    "dataset = json.load(open('atoi.json'))\n",
    "ans = dataset['ix_to_ans'][str(output[idx]+1)]\n",
    "print(\"Answer:\", ans)"
   ]
  }
 ],
 "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": 2
}