MNIST_Classifier_Tensorflow.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "id": "VDP5_7F-zPGw"
   },
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "import random\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "id": "dd9iyWsazofA"
   },
   "outputs": [],
   "source": [
    "tf.logging.set_verbosity(tf.logging.INFO)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "id": "bqgWxgMczplc"
   },
   "outputs": [],
   "source": [
    "def cnn_model_fn(features,labels,mode):\n",
    "\t#Correct dimensions\n",
    "\tinput_layer = tf.reshape(features[\"x\"], [-1,28,28,1])\n",
    "\n",
    "\t#Define Model\n",
    "\tconv1 = tf.layers.conv2d(\n",
    "\t\tinputs = input_layer,\n",
    "\t\tfilters=32,\n",
    "\t\tkernel_size=[5,5],\n",
    "\t\tpadding=\"same\", \n",
    "\t\tactivation=tf.nn.relu)\n",
    "\n",
    "\tpool1 = tf.layers.max_pooling2d(inputs=conv1, pool_size=[2,2],strides=2)\n",
    "\n",
    "\tconv2 = tf.layers.conv2d(\n",
    "\t\tinputs=pool1,\n",
    "\t\tfilters=64,\n",
    "\t\tkernel_size=[5,5],\n",
    "\t\tpadding=\"same\",\n",
    "\t\tactivation=tf.nn.relu)\n",
    "\n",
    "\tpool2 = tf.layers.max_pooling2d(inputs=conv2, pool_size=[2,2],strides=2)\n",
    "\n",
    "\tpool2_flat = tf.reshape(pool2, [-1,7*7*64])\n",
    "\tdense = tf.layers.dense(inputs=pool2_flat, units=1024,activation=tf.nn.relu)\n",
    "\tdropout = tf.layers.dropout(\n",
    "\t\tinputs=dense, rate=0.4, training=mode == tf.estimator.ModeKeys.TRAIN)\n",
    "\tlogits = tf.layers.dense(inputs=dropout,units=10)\n",
    "\n",
    "\t#Predictions\n",
    "\tpredictions = {\n",
    "\t\t\"classes\":tf.argmax(input=logits,axis=1),\n",
    "\t\t\"probabilities\": tf.nn.softmax(logits, name=\"softmax_tensor\")\n",
    "\t}\n",
    "\n",
    "\t#Predict Mode\n",
    "\tif mode == tf.estimator.ModeKeys.PREDICT :\n",
    "\t\treturn tf.estimator.EstimatorSpec(mode=mode, predictions=predictions)\n",
    "\n",
    "\t#Loss (eval and train both)\n",
    "\t# One hot labels ? \n",
    "\t# onehot_labels = tf.one_hot(indices=tf.cast(labels,tf.int32), depth=10)\n",
    "\tloss = tf.losses.sparse_softmax_cross_entropy(labels=labels, logits=logits) # onehot_labels=onehot_labels\n",
    "\n",
    "\t#Train\n",
    "\tif mode == tf.estimator.ModeKeys.TRAIN :\n",
    "\t\toptimizer = tf.train.GradientDescentOptimizer(learning_rate=0.01)\n",
    "\t\ttrain_op = optimizer.minimize(\n",
    "\t\t\tloss = loss,\n",
    "\t\t\tglobal_step = tf.train.get_global_step())\n",
    "\t\treturn tf.estimator.EstimatorSpec(mode=mode, loss=loss, train_op=train_op)\n",
    "\n",
    "\t#Eval\n",
    "\teval_metric_ops = {\n",
    "\t\t\"accuracy\" : tf.metrics.accuracy(\n",
    "\t\t\tlabels=labels, predictions=predictions[\"classes\"])}\n",
    "\treturn tf.estimator.EstimatorSpec(mode=mode, loss=loss, eval_metric_ops=eval_metric_ops)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "base_uri": "https://localhost:8080/",
     "height": 489
    },
    "colab_type": "code",
    "executionInfo": {
     "elapsed": 10294,
     "status": "ok",
     "timestamp": 1530974011570,
     "user": {
      "displayName": "Dipani Alish Rajesh",
      "photoUrl": "https://lh3.googleusercontent.com/a/default-user=s128",
      "userId": "105760295303045246892"
     },
     "user_tz": -330
    },
    "id": "CYMyPLU-zspE",
    "outputId": "1ee8b5a2-d390-4960-aa4b-eb72af7a5f7c"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Extracting MNIST_data/train-images-idx3-ubyte.gz\n",
      "Extracting MNIST_data/train-labels-idx1-ubyte.gz\n",
      "Extracting MNIST_data/t10k-images-idx3-ubyte.gz\n",
      "Extracting MNIST_data/t10k-labels-idx1-ubyte.gz\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.examples.tutorials.mnist import input_data\n",
    "mnist = input_data.read_data_sets(\"MNIST_data/\", one_hot=False)\n",
    "train_data = mnist.train.images\n",
    "train_labels = np.asarray(mnist.train.labels, dtype=np.int32)\n",
    "eval_data = mnist.test.images\n",
    "eval_labels = np.asarray(mnist.test.labels, dtype=np.int32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "base_uri": "https://localhost:8080/",
     "height": 663
    },
    "colab_type": "code",
    "executionInfo": {
     "elapsed": 3814,
     "status": "ok",
     "timestamp": 1530974015452,
     "user": {
      "displayName": "Dipani Alish Rajesh",
      "photoUrl": "https://lh3.googleusercontent.com/a/default-user=s128",
      "userId": "105760295303045246892"
     },
     "user_tz": -330
    },
    "id": "zWz9cHvZYUjE",
    "outputId": "aeacbce4-ec1c-47dd-c5a3-6c51c8e2e92c"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "16 Random images and their labels\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 16 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#Preview random image\n",
    "fig = plt.figure(figsize=(8,8))\n",
    "plt.subplots_adjust(top=1.2)\n",
    "col = 4\n",
    "row = 4\n",
    "print(str(col*row)+\" Random images and their labels\")\n",
    "ri = random.sample(range(0,len(train_data)),col*row)\n",
    "rii = 0\n",
    "for i in range(1, col*row+1):\n",
    "  img = np.copy(train_data[ri[rii]])\n",
    "  img = img.reshape((28,28))\n",
    "  img = np.asarray(img, dtype=np.float32)\n",
    "  fig.add_subplot(row,col,i)\n",
    "  x_label = 'Label : ' + str(train_labels[ri[rii]])\n",
    "  plt.xlabel(x_label)\n",
    "  plt.imshow(img, interpolation = 'nearest')\n",
    "  rii = rii + 1\n",
    "_ = plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "base_uri": "https://localhost:8080/",
     "height": 72
    },
    "colab_type": "code",
    "executionInfo": {
     "elapsed": 1586,
     "status": "ok",
     "timestamp": 1530974017113,
     "user": {
      "displayName": "Dipani Alish Rajesh",
      "photoUrl": "https://lh3.googleusercontent.com/a/default-user=s128",
      "userId": "105760295303045246892"
     },
     "user_tz": -330
    },
    "id": "EoWs7JZA06Ea",
    "outputId": "401a5ba9-f8ef-423f-bfc7-404fc73dde67"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Using default config.\n",
      "INFO:tensorflow:Using config: {'_model_dir': '/tmp/mnist_convnet_modelpr', '_tf_random_seed': None, '_save_summary_steps': 100, '_save_checkpoints_steps': None, '_save_checkpoints_secs': 600, '_session_config': None, '_keep_checkpoint_max': 5, '_keep_checkpoint_every_n_hours': 10000, '_log_step_count_steps': 100, '_train_distribute': None, '_service': None, '_cluster_spec': <tensorflow.python.training.server_lib.ClusterSpec object at 0x7fad06a7ba20>, '_task_type': 'worker', '_task_id': 0, '_global_id_in_cluster': 0, '_master': '', '_evaluation_master': '', '_is_chief': True, '_num_ps_replicas': 0, '_num_worker_replicas': 1}\n"
     ]
    }
   ],
   "source": [
    "mnist_classifier = tf.estimator.Estimator(\n",
    "model_fn = cnn_model_fn, model_dir = \"/tmp/mnist_convnet_modelpr\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     }
    },
    "colab_type": "code",
    "id": "UvUtNKARLlWJ"
   },
   "outputs": [],
   "source": [
    "# Logging predictions\n",
    "tensors_to_log = {\"probabilities\" : \"softmax_tensor\"}\n",
    "logging_hook = tf.train.LoggingTensorHook(\n",
    "  tensors = tensors_to_log, every_n_iter=500)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "base_uri": "https://localhost:8080/",
     "height": 2761
    },
    "colab_type": "code",
    "executionInfo": {
     "elapsed": 57303,
     "status": "ok",
     "timestamp": 1530974075961,
     "user": {
      "displayName": "Dipani Alish Rajesh",
      "photoUrl": "https://lh3.googleusercontent.com/a/default-user=s128",
      "userId": "105760295303045246892"
     },
     "user_tz": -330
    },
    "id": "dK8NErM0Lppm",
    "outputId": "72732731-f5f9-47ea-8fc9-c057ca8cc1ba"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Calling model_fn.\n",
      "INFO:tensorflow:Done calling model_fn.\n",
      "INFO:tensorflow:Create CheckpointSaverHook.\n",
      "INFO:tensorflow:Graph was finalized.\n",
      "INFO:tensorflow:Running local_init_op.\n",
      "INFO:tensorflow:Done running local_init_op.\n",
      "INFO:tensorflow:Saving checkpoints for 1 into /tmp/mnist_convnet_modelpr/model.ckpt.\n",
      "INFO:tensorflow:probabilities = [[0.10659222 0.10483029 0.09066836 ... 0.09512464 0.11550292 0.09699661]\n",
      " [0.09612888 0.11424766 0.08816691 ... 0.10810445 0.10691498 0.0950122 ]\n",
      " [0.09678549 0.11276269 0.09009096 ... 0.10688423 0.1057261  0.09530787]\n",
      " ...\n",
      " [0.10892161 0.11128951 0.09954917 ... 0.08338356 0.10962592 0.08133693]\n",
      " [0.10788575 0.10470948 0.08668446 ... 0.09587133 0.10356233 0.09514458]\n",
      " [0.09853997 0.09652935 0.09207782 ... 0.09598948 0.12102686 0.08984394]]\n",
      "INFO:tensorflow:loss = 2.3235822, step = 1\n",
      "INFO:tensorflow:global_step/sec: 1.99846\n",
      "INFO:tensorflow:loss = 2.0272045, step = 101 (50.038 sec)\n",
      "INFO:tensorflow:global_step/sec: 1.85846\n",
      "INFO:tensorflow:loss = 0.8408804, step = 201 (53.808 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.02533\n",
      "INFO:tensorflow:loss = 0.5647222, step = 301 (49.375 sec)\n",
      "INFO:tensorflow:global_step/sec: 1.81129\n",
      "INFO:tensorflow:loss = 0.38018516, step = 401 (55.215 sec)\n",
      "INFO:tensorflow:global_step/sec: 1.81824\n",
      "INFO:tensorflow:probabilities = [[0.00004626 0.9795084  0.00025079 ... 0.00041361 0.01473488 0.00086608]\n",
      " [0.00076472 0.00000858 0.00005558 ... 0.00023242 0.00236757 0.10015359]\n",
      " [0.00308392 0.5758523  0.03263458 ... 0.05690895 0.17641234 0.01687253]\n",
      " ...\n",
      " [0.00009612 0.12549783 0.00087488 ... 0.00664037 0.45884755 0.3888503 ]\n",
      " [0.00009893 0.00007669 0.00008087 ... 0.0193487  0.00505241 0.9006757 ]\n",
      " [0.9865072  0.00000197 0.00126355 ... 0.00020783 0.00062459 0.00006144]] (263.429 sec)\n",
      "INFO:tensorflow:loss = 0.46867067, step = 501 (54.994 sec)\n",
      "INFO:tensorflow:global_step/sec: 1.74799\n",
      "INFO:tensorflow:loss = 0.33726186, step = 601 (57.207 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.04437\n",
      "INFO:tensorflow:loss = 0.2537545, step = 701 (48.915 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.03194\n",
      "INFO:tensorflow:loss = 0.21686605, step = 801 (49.214 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.02581\n",
      "INFO:tensorflow:loss = 0.29313082, step = 901 (49.363 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.06868\n",
      "INFO:tensorflow:probabilities = [[0.8928992  0.00001678 0.03090751 ... 0.00040774 0.00150171 0.00052631]\n",
      " [0.8109052  0.00058608 0.01324294 ... 0.00498541 0.01887195 0.01388375]\n",
      " [0.00003287 0.979832   0.00940939 ... 0.00117848 0.00594752 0.00048883]\n",
      " ...\n",
      " [0.00334382 0.00288185 0.0709128  ... 0.00001955 0.70093095 0.00012849]\n",
      " [0.00219018 0.00013512 0.02328229 ... 0.00249038 0.68072164 0.27963215]\n",
      " [0.00011481 0.99502987 0.00056    ... 0.00043991 0.00158001 0.00042356]] (253.041 sec)\n",
      "INFO:tensorflow:loss = 0.15834256, step = 1001 (48.341 sec)\n",
      "INFO:tensorflow:global_step/sec: 1.69514\n",
      "INFO:tensorflow:loss = 0.17976737, step = 1101 (58.992 sec)\n",
      "INFO:tensorflow:Saving checkpoints for 1134 into /tmp/mnist_convnet_modelpr/model.ckpt.\n",
      "INFO:tensorflow:global_step/sec: 1.669\n",
      "INFO:tensorflow:loss = 0.26541263, step = 1201 (59.916 sec)\n",
      "INFO:tensorflow:global_step/sec: 1.83145\n",
      "INFO:tensorflow:loss = 0.24987194, step = 1301 (54.601 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.02329\n",
      "INFO:tensorflow:loss = 0.15893228, step = 1401 (49.425 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.05947\n",
      "INFO:tensorflow:probabilities = [[0.00002481 0.00014554 0.00004496 ... 0.00018032 0.00067228 0.0012192 ]\n",
      " [0.00001406 0.00001412 0.00012403 ... 0.00103174 0.00200182 0.9856695 ]\n",
      " [0.00001464 0.00000213 0.00002039 ... 0.01966186 0.00034586 0.92376834]\n",
      " ...\n",
      " [0.00001036 0.00000021 0.00012324 ... 0.00000001 0.99953544 0.00000931]\n",
      " [0.00000268 0.99362284 0.00351589 ... 0.00055347 0.00121206 0.00000995]\n",
      " [0.0000007  0.00000182 0.00004801 ... 0.9970493  0.00000415 0.00282203]] (271.491 sec)\n",
      "INFO:tensorflow:loss = 0.13995415, step = 1501 (48.557 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.1093\n",
      "INFO:tensorflow:loss = 0.1473303, step = 1601 (47.408 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.10318\n",
      "INFO:tensorflow:loss = 0.25883085, step = 1701 (47.548 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.11374\n",
      "INFO:tensorflow:loss = 0.12423104, step = 1801 (47.308 sec)\n",
      "INFO:tensorflow:global_step/sec: 1.74176\n",
      "INFO:tensorflow:loss = 0.10194868, step = 1901 (57.413 sec)\n",
      "INFO:tensorflow:global_step/sec: 1.89998\n",
      "INFO:tensorflow:probabilities = [[0.0010292  0.00001382 0.9334862  ... 0.02332089 0.02261804 0.00044146]\n",
      " [0.14599544 0.00001374 0.00068623 ... 0.00000002 0.00132985 0.00001305]\n",
      " [0.00014976 0.8823304  0.01673553 ... 0.00499883 0.06228948 0.00023072]\n",
      " ...\n",
      " [0.00001373 0.00008616 0.9988833  ... 0.00020867 0.00009648 0.00001079]\n",
      " [0.9979825  0.00000032 0.00016837 ... 0.00000237 0.00031837 0.00003833]\n",
      " [0.00003906 0.00000017 0.00008625 ... 0.00000158 0.9992729  0.00005996]] (252.311 sec)\n",
      "INFO:tensorflow:loss = 0.21569568, step = 2001 (52.634 sec)\n",
      "INFO:tensorflow:global_step/sec: 1.92397\n",
      "INFO:tensorflow:loss = 0.13367024, step = 2101 (51.975 sec)\n",
      "INFO:tensorflow:global_step/sec: 1.96194\n",
      "INFO:tensorflow:loss = 0.18302341, step = 2201 (50.970 sec)\n",
      "INFO:tensorflow:global_step/sec: 1.91126\n",
      "INFO:tensorflow:loss = 0.07999713, step = 2301 (52.321 sec)\n",
      "INFO:tensorflow:Saving checkpoints for 2311 into /tmp/mnist_convnet_modelpr/model.ckpt.\n",
      "INFO:tensorflow:global_step/sec: 1.94301\n",
      "INFO:tensorflow:loss = 0.11136244, step = 2401 (51.467 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.02918\n",
      "INFO:tensorflow:probabilities = [[0.00071574 0.00029744 0.00058002 ... 0.00023731 0.9450165  0.00133512]\n",
      " [0.00048615 0.00003662 0.0004674  ... 0.00000161 0.996447   0.00025156]\n",
      " [0.01262538 0.0002054  0.00185008 ... 0.00001417 0.00229643 0.00004486]\n",
      " ...\n",
      " [0.0000031  0.0000003  0.00370023 ... 0.99361444 0.00227035 0.00032416]\n",
      " [0.00001588 0.9928697  0.0030909  ... 0.00048195 0.00172796 0.00010448]\n",
      " [0.99993443 0.00000002 0.00002936 ... 0.00000003 0.00000472 0.00000457]] (256.015 sec)\n",
      "INFO:tensorflow:loss = 0.13349113, step = 2501 (49.283 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.10559\n",
      "INFO:tensorflow:loss = 0.07522559, step = 2601 (47.491 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.10704\n",
      "INFO:tensorflow:loss = 0.10997838, step = 2701 (47.460 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.05395\n",
      "INFO:tensorflow:loss = 0.03059917, step = 2801 (48.686 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.06169\n",
      "INFO:tensorflow:loss = 0.10485183, step = 2901 (48.504 sec)\n",
      "INFO:tensorflow:global_step/sec: 1.89363\n",
      "INFO:tensorflow:probabilities = [[0.0000024  0.00003382 0.00000201 ... 0.0000032  0.00053368 0.00000275]\n",
      " [0.00000592 0.00002785 0.00039481 ... 0.00007012 0.00002079 0.00011037]\n",
      " [0.00001555 0.00001017 0.9649204  ... 0.00549417 0.00073434 0.00000165]\n",
      " ...\n",
      " [0.9998211  0.00000001 0.00007838 ... 0.00000527 0.00001726 0.0000027 ]\n",
      " [0.23259398 0.173102   0.16418526 ... 0.00752369 0.00093075 0.01288975]\n",
      " [0.00000039 0.00000781 0.00011895 ... 0.99973327 0.00000183 0.00011906]] (244.952 sec)\n",
      "INFO:tensorflow:loss = 0.122532494, step = 3001 (52.810 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.0905\n",
      "INFO:tensorflow:loss = 0.083618015, step = 3101 (47.834 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.04693\n",
      "INFO:tensorflow:loss = 0.14325829, step = 3201 (48.853 sec)\n",
      "INFO:tensorflow:global_step/sec: 1.905\n",
      "INFO:tensorflow:loss = 0.021349119, step = 3301 (52.493 sec)\n",
      "INFO:tensorflow:global_step/sec: 1.93769\n",
      "INFO:tensorflow:loss = 0.056211285, step = 3401 (51.608 sec)\n",
      "INFO:tensorflow:global_step/sec: 1.97587\n",
      "INFO:tensorflow:probabilities = [[0.0002207  0.96161276 0.00150984 ... 0.02518726 0.00427658 0.00082153]\n",
      " [0.00009182 0.00001066 0.00004643 ... 0.00605436 0.00133981 0.9905184 ]\n",
      " [0.01140292 0.00003139 0.00997936 ... 0.00068594 0.1265757  0.13715725]\n",
      " ...\n",
      " [0.00000359 0.         0.00000279 ... 0.98373175 0.00000009 0.01624323]\n",
      " [0.00000093 0.         0.0000184  ... 0.99854076 0.00000041 0.00142243]\n",
      " [0.99238443 0.00000182 0.00180899 ... 0.000016   0.00339061 0.00006058]] (251.404 sec)\n",
      "INFO:tensorflow:loss = 0.06677065, step = 3501 (50.615 sec)\n",
      "INFO:tensorflow:Saving checkpoints for 3518 into /tmp/mnist_convnet_modelpr/model.ckpt.\n",
      "INFO:tensorflow:global_step/sec: 1.90027\n",
      "INFO:tensorflow:loss = 0.108734116, step = 3601 (52.619 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.03358\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:loss = 0.0674668, step = 3701 (49.174 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.00122\n",
      "INFO:tensorflow:loss = 0.07927959, step = 3801 (49.970 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.01801\n",
      "INFO:tensorflow:loss = 0.07494091, step = 3901 (49.554 sec)\n",
      "INFO:tensorflow:global_step/sec: 1.93374\n",
      "INFO:tensorflow:probabilities = [[0.00060723 0.00000083 0.9911618  ... 0.00001834 0.00779005 0.00024694]\n",
      " [0.00508307 0.00008368 0.00008998 ... 0.30250007 0.00021835 0.6882529 ]\n",
      " [0.0000018  0.00000576 0.00032815 ... 0.05877894 0.00030658 0.12432382]\n",
      " ...\n",
      " [0.         0.00000028 0.00000001 ... 0.00000211 0.00000402 0.00053092]\n",
      " [0.00000698 0.00001025 0.00000095 ... 0.00066589 0.00025255 0.00149377]\n",
      " [0.00001095 0.00004838 0.00018946 ... 0.00012632 0.00081564 0.00382439]] (253.032 sec)\n",
      "INFO:tensorflow:loss = 0.09973985, step = 4001 (51.715 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.01088\n",
      "INFO:tensorflow:loss = 0.093766615, step = 4101 (49.727 sec)\n",
      "INFO:tensorflow:global_step/sec: 2.00492\n",
      "INFO:tensorflow:loss = 0.07450803, step = 4201 (49.879 sec)\n",
      "INFO:tensorflow:Saving checkpoints for 4297 into /tmp/mnist_convnet_modelpr/model.ckpt.\n",
      "INFO:tensorflow:Loss for final step: 0.07060338.\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<tensorflow.python.estimator.estimator.Estimator at 0x7fad06a7b828>"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Training\n",
    "train_input_fn = tf.estimator.inputs.numpy_input_fn(\n",
    "  x = {\"x\": train_data},\n",
    "  y=train_labels,\n",
    "\tbatch_size=128,\n",
    "\tnum_epochs = 10,\n",
    "\tshuffle=True)\n",
    "mnist_classifier.train(input_fn = train_input_fn, hooks = [logging_hook])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "base_uri": "https://localhost:8080/",
     "height": 243
    },
    "colab_type": "code",
    "executionInfo": {
     "elapsed": 2110,
     "status": "ok",
     "timestamp": 1530974078141,
     "user": {
      "displayName": "Dipani Alish Rajesh",
      "photoUrl": "https://lh3.googleusercontent.com/a/default-user=s128",
      "userId": "105760295303045246892"
     },
     "user_tz": -330
    },
    "id": "HpKUCrbyLuSN",
    "outputId": "cbba1eb9-2ee6-432a-db81-6af90b0b47d7"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO:tensorflow:Calling model_fn.\n",
      "INFO:tensorflow:Done calling model_fn.\n",
      "INFO:tensorflow:Starting evaluation at 2018-07-14-14:06:37\n",
      "INFO:tensorflow:Graph was finalized.\n",
      "INFO:tensorflow:Restoring parameters from /tmp/mnist_convnet_modelpr/model.ckpt-4297\n",
      "INFO:tensorflow:Running local_init_op.\n",
      "INFO:tensorflow:Done running local_init_op.\n",
      "INFO:tensorflow:Finished evaluation at 2018-07-14-14:06:51\n",
      "INFO:tensorflow:Saving dict for global step 4297: accuracy = 0.9799, global_step = 4297, loss = 0.06461509\n",
      "{'accuracy': 0.9799, 'loss': 0.06461509, 'global_step': 4297}\n",
      "\n",
      "The accuracy on Test Dataset is : 97.99000024795532% and the loss is : 0.06461509\n"
     ]
    }
   ],
   "source": [
    "#Eval\n",
    "eval_input_fn = tf.estimator.inputs.numpy_input_fn(\n",
    "\tx={\"x\":eval_data},\n",
    "\ty=eval_labels,\n",
    "\tnum_epochs=1,\n",
    "\tshuffle=False)\n",
    "eval_results = mnist_classifier.evaluate(input_fn=eval_input_fn)\n",
    "print(eval_results)\n",
    "print()\n",
    "print(\"The accuracy on Test Dataset is : \"+str(eval_results[\"accuracy\"]*100) + \"% and the loss is : \" + str(eval_results[\"loss\"]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "colab": {
     "autoexec": {
      "startup": false,
      "wait_interval": 0
     },
     "base_uri": "https://localhost:8080/",
     "height": 779
    },
    "colab_type": "code",
    "executionInfo": {
     "elapsed": 3378,
     "status": "ok",
     "timestamp": 1530974081604,
     "user": {
      "displayName": "Dipani Alish Rajesh",
      "photoUrl": "https://lh3.googleusercontent.com/a/default-user=s128",
      "userId": "105760295303045246892"
     },
     "user_tz": -330
    },
    "id": "pzMGL6nsYqxB",
    "outputId": "e5ff079a-dff8-4133-82d3-30ab1e121e00"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "16 Random images and their labels and predicted class\n",
      "INFO:tensorflow:Calling model_fn.\n",
      "INFO:tensorflow:Done calling model_fn.\n",
      "INFO:tensorflow:Graph was finalized.\n",
      "INFO:tensorflow:Restoring parameters from /tmp/mnist_convnet_modelpr/model.ckpt-4297\n",
      "INFO:tensorflow:Running local_init_op.\n",
      "INFO:tensorflow:Done running local_init_op.\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 16 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#Preview random image\n",
    "fig = plt.figure(figsize=(8,8))\n",
    "plt.subplots_adjust(top=1.2)\n",
    "col = 4\n",
    "row = 4\n",
    "print(str(col*row)+\" Random images and their labels and predicted class\")\n",
    "#Generating random numbers, these numbers will be the indices of the images chosen\n",
    "ri = random.sample(range(0,len(eval_data)),col*row)\n",
    "\n",
    "#predict_x stores images chosen\n",
    "predict_x = []\n",
    "for i in ri:\n",
    "  predict_x.append(np.array(eval_data[i],dtype=np.float32))\n",
    "\n",
    "#Estimator to predict classes\n",
    "predict_input_fn = tf.estimator.inputs.numpy_input_fn(x={\"x\":np.array(predict_x,dtype=np.float32)},shuffle=False)\n",
    "predictions = mnist_classifier.predict(input_fn = predict_input_fn)\n",
    "#Storing the classes predicted\n",
    "predicted_class = []\n",
    "for pred_dict in predictions:\n",
    "  predicted_class.append(pred_dict['classes'])\n",
    "\n",
    "#Plotting\n",
    "rii = 0\n",
    "for i in range(1, col*row+1):\n",
    "  img = np.copy(eval_data[ri[rii]])\n",
    "  img = img.reshape((28,28))\n",
    "  img = np.asarray(img, dtype=np.float32)\n",
    "  fig.add_subplot(row,col,i)\n",
    "  x_label = 'Label : ' + str(eval_labels[ri[rii]]) + \"\\n\" + \"Predicted Class : \" + str(predicted_class[rii])\n",
    "  plt.xlabel(x_label)\n",
    "  plt.imshow(img, interpolation = 'nearest')\n",
    "  rii = rii + 1\n",
    "_ = plt.show()"
   ]
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "default_view": {},
   "name": "MNIST.ipynb",
   "provenance": [],
   "version": "0.3.2",
   "views": {}
  },
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}