Summary.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 128,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "#importing a standard plotting python package\n",
    "import matplotlib.pyplot as plt\n",
    "#this makes the plots appear here instead of opening in a new window\n",
    "%matplotlib inline\n",
    "\n",
    "#keras were kind enough to wrap fashion_mnist for us, so we don't need to download the images ourselves\n",
    "from keras.datasets import fashion_mnist\n",
    "\n",
    "#importing numpy since we will need to for arithmetic operations\n",
    "import numpy as np"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 129,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "#Load the training and testing data, X is images, y are labels\n",
    "(X_train, y_train), (X_test, y_test) = fashion_mnist.load_data()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.text.Text at 0x7f7aeb95b0f0>"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP4AAAEKCAYAAAAy4ujqAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nEI34DQ1TiEb8hoYpRCN+Q8MU4v8DqyKzDdvaF34AAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f7af0e8d908>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP4AAAEKCAYAAAAy4ujqAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f7aeba1a400>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP4AAAEKCAYAAAAy4ujqAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n43SQ/w/EJYBbF5ECLgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f7aeb9745f8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP4AAAEKCAYAAAAy4ujqAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n04ns8/kAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f7aeb984860>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#lets look at some images\n",
    "plt.figure()\n",
    "plt.imshow(X_train[0], cmap='gray')\n",
    "plt.title('sample of class number {}'.format(y_train[0]))\n",
    "\n",
    "plt.figure()\n",
    "plt.imshow(X_train[1], cmap='gray')\n",
    "plt.title('sample of class number {}'.format(y_train[1]))\n",
    "\n",
    "plt.figure()\n",
    "plt.imshow(X_train[2], cmap='gray')\n",
    "plt.title('sample of class number {}'.format(y_train[2]))\n",
    "\n",
    "plt.figure()\n",
    "plt.imshow(X_train[3], cmap='gray')\n",
    "plt.title('sample of class number {}'.format(y_train[3]))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Normalizing the images and converting to float"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 130,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "X_train = X_train.astype(np.float)\n",
    "X_test = X_test.astype(np.float)\n",
    "\n",
    "X_train /= 255.0\n",
    "X_test /= 255.0"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "### taking part of the train set for validation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "there are 60000 training samples, we will take 50K for train and 10K for validation\n"
     ]
    }
   ],
   "source": [
    "X_train_orig = X_train\n",
    "y_train_orig = y_train\n",
    "\n",
    "print('there are {} training samples, we will take 50K for train and 10K for validation'.format(X_train.shape[0]))\n",
    "X_train = X_train_orig[:50000]\n",
    "y_train = y_train_orig[:50000]\n",
    "\n",
    "X_val = X_train_orig[50000:]\n",
    "y_val = y_train_orig[50000:]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### let's try classification with KNN\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "NearestNeighbors(algorithm='auto', leaf_size=30, metric='minkowski',\n",
       "         metric_params=None, n_jobs=1, n_neighbors=10, p=2, radius=1.0)"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#Importing the NN class from sklearn  \n",
    "from sklearn.neighbors import NearestNeighbors\n",
    "\n",
    "#KNeighborsClassifier wants us to roll each image into one long vector \n",
    "X_train_rolled = X_train.reshape(-1,28*28)\n",
    "X_val_rolled = X_val.reshape(-1,28*28)\n",
    "\n",
    "#initiating a new KNN classifier \n",
    "max_K = 10\n",
    "neigh = NearestNeighbors(n_neighbors=max_K)\n",
    "#\"training\" the classifier \n",
    "neigh.fit(X_train_rolled,y_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "# getting the closest neigherst neigbors. Here we get the distance between each sample in the validtion set to\n",
    "# the five closest neighbors and also the neighbors indices in the validation set \n",
    "distances, indices = neigh.kneighbors(X_val_rolled)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[42052,  9801,  6756, ..., 41602, 13893,  9356],\n",
       "       [41387, 30078, 27732, ..., 46595, 12616, 46418],\n",
       "       [15801,  2237, 37885, ..., 49010, 23827, 12027],\n",
       "       ..., \n",
       "       [45348, 22272, 43297, ...,  2717, 21782, 17233],\n",
       "       [17378,  8495, 24332, ..., 11955, 11848, 13448],\n",
       "       [11912, 40600, 49655, ...,  2227, 45354, 29249]])"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "indices"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "#For each validation sample, we know the indices of it's 5 nn in the training set, let's see what are the labels \n",
    "#of those 5 nn\n",
    "#I'm doing this with 2 loops, not the most elegant code..\n",
    "y_val_pred = []\n",
    "for row_ind in range(indices.shape[0]):\n",
    "    cur_sample_row = []\n",
    "    for col_ind in range(indices.shape[1]):\n",
    "        cur_sample_row.append(y_train[indices[row_ind,col_ind]])\n",
    "        \n",
    "    y_val_pred.append(cur_sample_row)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "#converting this to a numpy matrix\n",
    "y_val_pred = np.array(y_val_pred)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[9, 9, 9, ..., 9, 9, 7],\n",
       "       [2, 2, 2, ..., 2, 2, 2],\n",
       "       [1, 1, 1, ..., 1, 1, 1],\n",
       "       ..., \n",
       "       [3, 3, 3, ..., 3, 3, 3],\n",
       "       [0, 0, 0, ..., 6, 0, 0],\n",
       "       [5, 5, 5, ..., 5, 5, 5]], dtype=uint8)"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "#let's look at the predictions\n",
    "y_val_pred"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "with k = 1 our acc on the validtion is 0.8458\n",
      "with k = 2 our acc on the validtion is 0.8409\n",
      "with k = 3 our acc on the validtion is 0.8519\n",
      "with k = 4 our acc on the validtion is 0.8506\n",
      "with k = 5 our acc on the validtion is 0.8523\n",
      "with k = 6 our acc on the validtion is 0.8516\n",
      "with k = 7 our acc on the validtion is 0.8532\n",
      "with k = 8 our acc on the validtion is 0.8518\n",
      "with k = 9 our acc on the validtion is 0.8511\n",
      "with k = 10 our acc on the validtion is 0.8476\n"
     ]
    }
   ],
   "source": [
    "#the number of neighbors is a hyper param. we'll loop over K=1,2,3,4,5\n",
    "#we need to most common value in each prediction vector, for that we will use the counter class\n",
    "from collections import Counter\n",
    "\n",
    "for k in range(max_K):\n",
    "    y_val_pred_for_k = y_val_pred[:,:k+1]\n",
    "    \n",
    "    final_prediction_vec = []\n",
    "    for sample_ind in range(y_val_pred_for_k.shape[0]):\n",
    "        final_prediction_vec.append(Counter(y_val_pred_for_k[sample_ind]).most_common()[0][0])\n",
    "    \n",
    "    final_prediction_vec = np.array(final_prediction_vec)\n",
    "\n",
    "    #we have a final prediction for our validation set for this k, let's see what is the accuracy\n",
    "    cur_acc = np.sum(final_prediction_vec==y_val)/float(len(y_val))\n",
    "    print ('with k = {} our acc on the validtion is {}'.format(k+1, cur_acc))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Our best k is 7, so let's try that on our test set\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "X_test_rolled = X_test.reshape(-1,28*28)\n",
    "distances, indices = neigh.kneighbors(X_test_rolled)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "y_test_pred = []\n",
    "for row_ind in range(indices.shape[0]):\n",
    "    cur_sample_row = []\n",
    "    for col_ind in range(indices.shape[1]):\n",
    "        cur_sample_row.append(y_train[indices[row_ind,col_ind]])\n",
    "        \n",
    "    y_test_pred.append(cur_sample_row)\n",
    "\n",
    "#converting this to a numpy matrix\n",
    "y_test_pred = np.array(y_test_pred)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "with k = 10 our acc on the test is 0.8474\n"
     ]
    }
   ],
   "source": [
    "y_test_pred_for_k = y_test_pred[:,:7+1]\n",
    "\n",
    "\n",
    "final_prediction_vec = []\n",
    "for sample_ind in range(y_test_pred_for_k.shape[0]):\n",
    "    final_prediction_vec.append(Counter(y_test_pred_for_k[sample_ind]).most_common()[0][0])\n",
    "\n",
    "final_prediction_vec = np.array(final_prediction_vec)\n",
    "\n",
    "#we have a final prediction for our validation set for this k, let's see what is the accuracy\n",
    "cur_acc = np.sum(final_prediction_vec==y_test)/float(len(y_test))\n",
    "print ('with k = {} our acc on the test is {}'.format(7, cur_acc))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Let's try Linear classification "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "acc of LinearSVC with C = 0.0001 is 0.8083\n",
      "acc of LinearSVC with C = 0.0003 is 0.8271\n",
      "acc of LinearSVC with C = 0.001 is 0.8417\n",
      "acc of LinearSVC with C = 0.003 is 0.8484\n",
      "acc of LinearSVC with C = 0.01 is 0.8528\n",
      "acc of LinearSVC with C = 0.03 is 0.852\n",
      "acc of LinearSVC with C = 0.1 is 0.8507\n",
      "acc of LinearSVC with C = 0.3 is 0.849\n",
      "acc of LinearSVC with C = 1.0 is 0.8485\n",
      "acc of LinearSVC with C = 3.0 is 0.8286\n"
     ]
    }
   ],
   "source": [
    "#LinearSVC is one example of a linear model\n",
    "from sklearn.svm import LinearSVC\n",
    "\n",
    "#One of the hyperparameters of LR is C, how well to fit the training data or how much regularization do we want\n",
    "for cur_C in [0.0001,0.0003, 0.001, 0.003, 0.01, 0.03,0.1,0.3,1.0]:\n",
    "    log_reg_clf = LinearSVC(C=cur_C)\n",
    "    log_reg_clf.fit(X_train_rolled,y_train) #this takes care of all the SGD and all that \n",
    "    cur_acc = log_reg_clf.score(X_val_rolled,y_val)\n",
    "    print ('acc of LinearSVC with C = {} is {}'.format(cur_C, cur_acc))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " logistic regression with C=0.01 reached 0.8404 acc\n"
     ]
    }
   ],
   "source": [
    "#The best cur is 0.01 so let's try that on the test set\n",
    "log_reg_clf = LinearSVC(C=0.01)\n",
    "log_reg_clf.fit(X_train_rolled,y_train) #this takes care of all the SGD and all that \n",
    "test_acc = log_reg_clf.score(X_test_rolled,y_test)\n",
    "print(' logistic regression with C={} reached {} acc'.format(0.01,test_acc))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Let's try regular fully connected NN "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "from keras.models import Sequential\n",
    "from keras.layers.core import Dense, Dropout, Activation\n",
    "\n",
    "#Keras wants one-hot vector\n",
    "from keras.utils import np_utils\n",
    "Y_train = np_utils.to_categorical(y_train,10)\n",
    "Y_val = np_utils.to_categorical(y_val,10)\n",
    "Y_test = np_utils.to_categorical(y_test,10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "collapsed": false,
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/ubuntu/anaconda2/envs/deep-learning/lib/python3.5/site-packages/keras/models.py:939: UserWarning: The `nb_epoch` argument in `fit` has been renamed `epochs`.\n",
      "  warnings.warn('The `nb_epoch` argument in `fit` '\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 50000 samples, validate on 10000 samples\n",
      "Epoch 1/1000\n",
      "50000/50000 [==============================] - 4s 87us/step - loss: 0.8722 - acc: 0.7218 - val_loss: 0.6541 - val_acc: 0.7844\n",
      "Epoch 2/1000\n",
      "50000/50000 [==============================] - 4s 85us/step - loss: 0.6101 - acc: 0.8035 - val_loss: 0.5793 - val_acc: 0.8081\n",
      "Epoch 3/1000\n",
      "50000/50000 [==============================] - 4s 87us/step - loss: 0.5556 - acc: 0.8206 - val_loss: 0.5458 - val_acc: 0.8159\n",
      "Epoch 4/1000\n",
      "50000/50000 [==============================] - 4s 88us/step - loss: 0.5269 - acc: 0.8273 - val_loss: 0.5230 - val_acc: 0.8206\n",
      "Epoch 5/1000\n",
      "50000/50000 [==============================] - 5s 92us/step - loss: 0.5078 - acc: 0.8319 - val_loss: 0.5095 - val_acc: 0.8248\n",
      "Epoch 6/1000\n",
      "50000/50000 [==============================] - 3s 67us/step - loss: 0.4944 - acc: 0.8361 - val_loss: 0.5000 - val_acc: 0.8306\n",
      "Epoch 7/1000\n",
      "50000/50000 [==============================] - 3s 70us/step - loss: 0.4843 - acc: 0.8380 - val_loss: 0.4885 - val_acc: 0.8312\n",
      "Epoch 8/1000\n",
      "50000/50000 [==============================] - 3s 70us/step - loss: 0.4756 - acc: 0.8416 - val_loss: 0.4926 - val_acc: 0.8272\n",
      "Epoch 9/1000\n",
      "50000/50000 [==============================] - 3s 65us/step - loss: 0.4690 - acc: 0.8426 - val_loss: 0.4774 - val_acc: 0.8336\n",
      "Epoch 10/1000\n",
      "50000/50000 [==============================] - 3s 66us/step - loss: 0.4628 - acc: 0.8449 - val_loss: 0.4717 - val_acc: 0.8344\n",
      "Epoch 11/1000\n",
      "50000/50000 [==============================] - 3s 63us/step - loss: 0.4577 - acc: 0.8463 - val_loss: 0.4701 - val_acc: 0.8392\n",
      "Epoch 12/1000\n",
      "50000/50000 [==============================] - 3s 66us/step - loss: 0.4537 - acc: 0.8476 - val_loss: 0.4634 - val_acc: 0.8399\n",
      "Epoch 13/1000\n",
      "50000/50000 [==============================] - 3s 63us/step - loss: 0.4494 - acc: 0.8493 - val_loss: 0.4597 - val_acc: 0.8396\n",
      "Epoch 14/1000\n",
      "50000/50000 [==============================] - 3s 64us/step - loss: 0.4459 - acc: 0.8491 - val_loss: 0.4574 - val_acc: 0.8419\n",
      "Epoch 15/1000\n",
      "50000/50000 [==============================] - 3s 65us/step - loss: 0.4429 - acc: 0.8501 - val_loss: 0.4545 - val_acc: 0.8438\n",
      "Epoch 16/1000\n",
      "50000/50000 [==============================] - 4s 76us/step - loss: 0.4398 - acc: 0.8507 - val_loss: 0.4556 - val_acc: 0.8406\n",
      "Epoch 17/1000\n",
      "50000/50000 [==============================] - 5s 94us/step - loss: 0.4373 - acc: 0.8519 - val_loss: 0.4499 - val_acc: 0.8450\n",
      "Epoch 18/1000\n",
      "50000/50000 [==============================] - 5s 96us/step - loss: 0.4348 - acc: 0.8522 - val_loss: 0.4475 - val_acc: 0.8464\n",
      "Epoch 19/1000\n",
      "50000/50000 [==============================] - 5s 102us/step - loss: 0.4326 - acc: 0.8520 - val_loss: 0.4459 - val_acc: 0.8461\n",
      "Epoch 20/1000\n",
      "50000/50000 [==============================] - 5s 109us/step - loss: 0.4306 - acc: 0.8537 - val_loss: 0.4452 - val_acc: 0.8485\n",
      "Epoch 21/1000\n",
      "50000/50000 [==============================] - 6s 110us/step - loss: 0.4290 - acc: 0.8542 - val_loss: 0.4442 - val_acc: 0.8464\n",
      "Epoch 22/1000\n",
      "50000/50000 [==============================] - 6s 115us/step - loss: 0.4268 - acc: 0.8546 - val_loss: 0.4452 - val_acc: 0.8499\n",
      "Epoch 23/1000\n",
      "50000/50000 [==============================] - 6s 113us/step - loss: 0.4256 - acc: 0.8543 - val_loss: 0.4420 - val_acc: 0.8461\n",
      "Epoch 24/1000\n",
      "50000/50000 [==============================] - 4s 80us/step - loss: 0.4239 - acc: 0.8561 - val_loss: 0.4412 - val_acc: 0.8483\n",
      "Epoch 25/1000\n",
      "50000/50000 [==============================] - 3s 53us/step - loss: 0.4221 - acc: 0.8562 - val_loss: 0.4405 - val_acc: 0.8487\n",
      "Epoch 26/1000\n",
      "50000/50000 [==============================] - 3s 53us/step - loss: 0.4207 - acc: 0.8567 - val_loss: 0.4417 - val_acc: 0.8461\n",
      "Epoch 27/1000\n",
      "50000/50000 [==============================] - 3s 55us/step - loss: 0.4192 - acc: 0.8571 - val_loss: 0.4372 - val_acc: 0.8484\n",
      "Epoch 28/1000\n",
      "50000/50000 [==============================] - 3s 54us/step - loss: 0.4181 - acc: 0.8580 - val_loss: 0.4369 - val_acc: 0.8495\n",
      "Epoch 29/1000\n",
      "50000/50000 [==============================] - 3s 54us/step - loss: 0.4172 - acc: 0.8578 - val_loss: 0.4363 - val_acc: 0.8491\n",
      "Epoch 30/1000\n",
      "50000/50000 [==============================] - 3s 54us/step - loss: 0.4162 - acc: 0.8575 - val_loss: 0.4322 - val_acc: 0.8529\n",
      "Epoch 31/1000\n",
      "50000/50000 [==============================] - 3s 54us/step - loss: 0.4149 - acc: 0.8582 - val_loss: 0.4339 - val_acc: 0.8511\n",
      "Epoch 32/1000\n",
      "50000/50000 [==============================] - 3s 52us/step - loss: 0.4140 - acc: 0.8586 - val_loss: 0.4365 - val_acc: 0.8496\n",
      "Epoch 33/1000\n",
      "50000/50000 [==============================] - 3s 65us/step - loss: 0.4127 - acc: 0.8593 - val_loss: 0.4318 - val_acc: 0.8510\n",
      "Epoch 34/1000\n",
      "50000/50000 [==============================] - 4s 73us/step - loss: 0.4119 - acc: 0.8583 - val_loss: 0.4336 - val_acc: 0.8514\n",
      "Epoch 35/1000\n",
      "50000/50000 [==============================] - 4s 78us/step - loss: 0.4109 - acc: 0.8591 - val_loss: 0.4284 - val_acc: 0.8540\n",
      "Epoch 36/1000\n",
      "50000/50000 [==============================] - 4s 73us/step - loss: 0.4100 - acc: 0.8605 - val_loss: 0.4370 - val_acc: 0.8486\n",
      "Epoch 37/1000\n",
      "50000/50000 [==============================] - 4s 76us/step - loss: 0.4091 - acc: 0.8604 - val_loss: 0.4286 - val_acc: 0.8540\n",
      "Epoch 38/1000\n",
      "50000/50000 [==============================] - 4s 74us/step - loss: 0.4084 - acc: 0.8600 - val_loss: 0.4286 - val_acc: 0.8522\n",
      "Epoch 39/1000\n",
      "50000/50000 [==============================] - 4s 78us/step - loss: 0.4077 - acc: 0.8604 - val_loss: 0.4341 - val_acc: 0.8505\n",
      "Epoch 00039: early stopping\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7f7abca71eb8>"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "## let's start with a pretty weak model, no hidden layer, just input and output with softmax\n",
    "from keras.callbacks import EarlyStopping\n",
    "#let's monitor the validation loss, if the validation loss does not decrease during 4 epochs we will stop training\n",
    "early_stop = EarlyStopping(monitor='val_loss', patience=4, verbose=1) \n",
    "\n",
    "\n",
    "nn_model1 = Sequential()\n",
    "nn_model1.add(Dense(10, input_shape = (X_train_rolled.shape[1],)))\n",
    "nn_model1.add(Activation('softmax'))\n",
    "\n",
    "nn_model1.compile(loss='categorical_crossentropy',\n",
    "              optimizer='sgd',\n",
    "              metrics=['accuracy'])\n",
    "\n",
    "history = nn_model1.fit(X_train_rolled, Y_train, batch_size=32, \n",
    "          nb_epoch=1000,verbose=1,\n",
    "          validation_data=(X_val_rolled, Y_val),\n",
    "          callbacks=[early_stop])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "plt.plot(history.history['loss'])\n",
    "plt.plot(history.history['val_loss'])\n",
    "plt.title('model accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'test'], loc='upper left')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "acc is 0.8505\n"
     ]
    }
   ],
   "source": [
    "#we stopped training, let's calculate accuracy on the validation set\n",
    "Y_pred = nn_model1.predict(X_val_rolled)\n",
    "#how many correct predictions\n",
    "correct_num = np.sum(Y_pred.argmax(axis=1)==y_val)\n",
    "print ('acc is {}'.format(correct_num/float(len(y_val))))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "the correct class for image 1 is 9\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP4AAAD8CAYAAABXXhlaAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n4v8DIfBJzvkS9soAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f7abc671c88>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAEACAYAAACj0I2EAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f7abc25aa20>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#prediction for image 0\n",
    "plt.figure()\n",
    "plt.imshow(X_val[0],cmap='gray')\n",
    "plt.figure()\n",
    "plt.plot(Y_pred[0],'bo')\n",
    "print ( 'the correct class for image 0 is {}'.format(y_val[0]) )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 71,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "the correct class for image 1 is 2\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAP4AAAD8CAYAAABXXhlaAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f7abc336e80>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAEACAYAAACj0I2EAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAE9FJREFUeJzt3W+wXPV93/H3R8g4Aeo4qdu6uoAsi9ikPCF2RmESx12C\nKXJD0SRNW6liAqnrZsYVZJy4IyYzsq5Gj0jdpqlJmnhKwYlJ5Bg3CfXYieSSTZtpMKoBQ2QJKUKW\n9Ye4tXHjgUxTBX37YFdiuazu7oWVdvXT+zVz556z+9s9nzn33s89+9s/J1WFJOn8t2zaASRJk2Gh\nS1IjLHRJaoSFLkmNsNAlqREWuiQ1YqxCT7I2yb4k+5NsHnL9FUkeTvJYkieSvHfyUSVJi8mo16En\nWQbsB24AjgO7gfVVtW9gzK8Cj1XVryb5HuCzVbXq7MWWJC00zhH6GuBAVR2uqhPADmDdgjEngTf0\nl98IHJtcREnSOJaPMWYOODKwfpReyQ/aBuxMcidwCfCeycSTJI1rnCP0DLls4TzNBuC+qroC+BHg\nE681mCRpacY5Qj8KXDmwfjm9ufRB7wNuAqiqR5J8W5I3VdXXBwcl8YNjJOlVqKphB9cvM84R+m7g\nqiQrk1wMrAceWjDmMP1plv6Toq9fWOYDoWbqa+vWrVPPUFV0Oh+m98CngK2nl6+//sNTzzZr+8pM\nZrrQco1rZKFX1YvAJmAnsAfYUVV7k2xLcnN/2IeA9yd5AngAuG3sBAJgbm4Z8MKCS19gxQrfKiBp\nPONMuVBVvwe8fcFlWweW9wLvmmy0C8v27bfzyCNbOXhwW/+SF1i9eivbt98xzViSziMX/OFfp9OZ\ndgQAVq1aya5dd7Bx40e49tpDbNz4EXbtuoNVq1ZOO9pps7KvBplpPGYa36zmGsfINxZNdGNJncvt\nSVILklATelJUknQesNAlqREWuiQ1wkKXpEZY6JLUCAtdkhphoUtSIyx0SWqEhS5JjbDQJakRFrok\nNcJCl6RGWOiS1AgLXZIaYaFLUiMsdElqhIUuSY0Yq9CTrE2yL8n+JJuHXP9vkzye5LEkTyd5bvJR\nJUmLGXkKuiTLgP3ADcBxYDewvqr2nWH8JuDaqvrnQ67zFHSStESTPAXdGuBAVR2uqhPADmDdIuM3\nAL85XkxJ0qSMU+hzwJGB9aP9y14hyZXAW4CHX3MySdKSjFPoww7zzzRvsh540HkVSTr3lo8x5ihw\n5cD65fTm0odZD3xgsTubn58/vdzpdOh0OmNEkKQLR7fbpdvtLvl24zwpehHwNL0nRZ8FHgU2VNXe\nBePeDnyuqt66yH158C5JSzSxJ0Wr6kVgE7AT2APsqKq9SbYluXlg6Hp6T5hKkqZg5BH6RDfmEbok\nLdkkX7YoSToPWOiS1AgLXZIaYaFLUiMsdElqhIUuSY2w0CWpERa6JDXCQpekRljoktQIC12SGmGh\nS1IjLHRJaoSFLkmNsNAlqREWuiQ1wkKXpEZY6JLUCAtdkhoxVqEnWZtkX5L9STafYcw/TrInyVNJ\nPjHZmJKkUUaeJDrJMmA/cANwHNgNrK+qfQNjrgI+CVxfVd9K8qaq+vqQ+/Ik0ZK0RJM8SfQa4EBV\nHa6qE8AOYN2CMe8HfqmqvgUwrMwlSWfXOIU+BxwZWD/av2zQ24C3J/mjJP8jyU2TCihJGs/yMcYM\nO8xfOG+yHLgKeDdwJfDfk1xz6oh90Pz8/OnlTqdDp9MZN6skXRC63S7dbnfJtxtnDv06YL6q1vbX\n7wKqqu4eGPMfgD+uql/rr38e2FxVX1xwX86hS9ISTXIOfTdwVZKVSS4G1gMPLRjzO8AP9zf8JuC7\ngWeWFlmS9FqMLPSqehHYBOwE9gA7qmpvkm1Jbu6P+X3gG0n2AP8V+FBVffMs5pYkLTByymWiG3PK\nRZKWbJJTLpKk84CFLkmNsNAlqREWuiQ1wkKXpEZY6JLUCAtdkhphoUtSIyx0SWqEhS5JjbDQJakR\nFrokNcJCl6RGWOiS1AgLXZIaYaFLUiMsdElqhIUuSY2w0CWpEWMVepK1SfYl2Z9k85Drb0vyv5I8\n1v/6Z5OPKklazPJRA5IsA+4BbgCOA7uT/G5V7VswdEdV3XkWMkqSxjDOEfoa4EBVHa6qE8AOYN2Q\ncSPPSC1JOnvGKfQ54MjA+tH+ZQv9WJInkvxWkssnkk6SNLaRUy4MP/KuBesPAb9RVSeS/BTwcXpT\nNK8wPz9/ernT6dDpdMYKKkkXim63S7fbXfLtUrWwmxcMSK4D5qtqbX/9LqCq6u4zjF8GPFdVbxxy\nXY3aniTp5ZJQVSOntceZctkNXJVkZZKLgfX0jsgHN/bmgdV1wJeXElaS9NqNnHKpqheTbAJ20vsH\ncG9V7U2yDdhdVZ8B7kxyC3ACeA64/SxmliQNMXLKZaIbc8pFkpZsklMukqTzgIUuSY2w0CWpERa6\nJDXCQpekRljoktQIC12SGmGhS1IjLHRJaoSFLkmNsNAlqREWuiQ1wkKXpEZY6JLUCAtdkhphoUtS\nIyx0SWqEhS5JjbDQJakRYxV6krVJ9iXZn2TzIuN+PMnJJO+YXERJ0jhGFnqSZcA9wE3ANcCGJFcP\nGXcZcAfwyKRDSpJGG+cIfQ1woKoOV9UJYAewbsi47cDdwF9OMJ8kaUzjFPoccGRg/Wj/stOSXAtc\nXlWfnWA2SdISLB9jTIZcVqevTAL8AnDbiNsAMD8/f3q50+nQ6XTGiCBJF45ut0u3213y7VJViw9I\nrgPmq2ptf/0uoKrq7v76G4A/BZ6nV+RvBr4B3FJVjy24rxq1PUnSyyWhqs54oHx63BiFfhHwNHAD\n8CzwKLChqvaeYfwfAD9TVY8Puc5Cl6QlGrfQR86hV9WLwCZgJ7AH2FFVe5NsS3LzsJuwyJSLJOns\nGHmEPtGNeYQuSUs2sSN0SdL5wUKXpEZY6JLUCAtdkhphoUtSIyx0SWqEhS5JjbDQJakRFrokNcJC\nl6RGWOiS1AgLXZIaYaFLUiMsdElqhIUuSY2w0CWpERa6JDXCQpekRljoktSIsQo9ydok+5LsT7J5\nyPU/leTJJI8n+W9Jrp58VEnSYkaeJDrJMmA/cANwHNgNrK+qfQNjLquq5/vL/wD4QFW9d8h9eZJo\nSVqiSZ4keg1woKoOV9UJYAewbnDAqTLvuww4uZSwkqTXbvkYY+aAIwPrR+mV/Msk+QDwM8DrgB+e\nSDpJ0tjGKfRhh/mvmDepql8GfjnJemALcPuwO5ufnz+93Ol06HQ6Y0SQpAtHt9ul2+0u+XbjzKFf\nB8xX1dr++l1AVdXdZxgf4JtV9cYh1zmHLklLNMk59N3AVUlWJrkYWA88tGBjVw2s3kzvSVRJ0jk0\ncsqlql5MsgnYSe8fwL1VtTfJNmB3VX0G2JTkPcD/A74J3HY2Q0uSXmnklMtEN+aUiyQt2SSnXCRJ\n5wELXZIaYaFLUiMsdElqhIUuSY2w0CWpERa6JDXCQpekRljoktQIC12SGmGhS1IjLHRJaoSFLkmN\nsNAlqREWuiQ1wkKXpEZY6JLUCAtdkhphoUtSI8Yq9CRrk+xLsj/J5iHXfzDJniRPJNmV5IrJR5Uk\nLWZkoSdZBtwD3ARcA2xIcvWCYY8B76yqa4FPA/960kElSYsb5wh9DXCgqg5X1QlgB7BucEBV/WFV\n/d/+6iPA3GRjSpJGGafQ54AjA+tHWbyw3wd87rWEkiQt3fIxxmTIZTV0YHIr8E7g757pzubn508v\ndzodOp3OGBEk6cLR7XbpdrtLvl2qhnbzSwOS64D5qlrbX78LqKq6e8G49wC/CLy7qr5xhvuqUduT\nJL1cEqpq2MH1y4wz5bIbuCrJyiQXA+uBhxZs7HuBXwFuOVOZS5LOrpGFXlUvApuAncAeYEdV7U2y\nLcnN/WE/D1wKfCrJ40l+56wlliQNNXLKZaIbc8pFkpZsklMukqTzwDivcpF0njp06DBbttzPsWMn\nmZtbxvbtt7Nq1cppx9JZ4pSL1KhDhw5z440f5eDBbfSe4nqB1au3smvXHZb6ecYpF+kCt2XL/QNl\nDnApBw9uY8uW+6eYSmeThS416tixk7xU5qdcyvHjJ6cRR+eAhS41am5uGfDCgktfYMUK/+xb5U9W\natT27bezevVWXir13hz69u23Ty2Tzi6fFJUadupVLsePn2TFCl/lcr4a90lRC12SZpyvcpGkC4yF\nLkmNsNAlqREWuiQ1wkKXpEZY6JLUCAtdkhphoUtSIyx0SWrEWIWeZG2SfUn2J9k85PofSvLFJCeS\n/NjkY0qSRhlZ6EmWAfcANwHXABuSXL1g2GHgNuCBiSeUJI1lnFPQrQEOVNVhgCQ7gHXAvlMDquqr\n/ev8oBZJmpJxplzmgCMD60f7l0mSZsg4hT7sE748EpekGTPOlMtR4MqB9cuB4692g/Pz86eXO50O\nnU7n1d6VJDWp2+3S7XaXfLuRn4ee5CLgaeAG4FngUWBDVe0dMvY+4DNV9ekz3Jefhy5JSzSxz0Ov\nqheBTcBOYA+wo6r2JtmW5Ob+xr4vyRHgx4FfSfLUa4svSVoqz1gkSTPOMxZJ0gXGQpekRljoktQI\nC12SGmGhS1IjLHRJaoSFLkmNsNAlqREWuiQ1wkKXpEZY6JLUCAtdkhphoUtSIyx0SWqEhS5JjbDQ\nJakRFrokNWKck0TrAnfo0GG2bLmfY8dOMje3jO3bb2fVqpVmkmaMp6DTog4dOsyNN36Ugwe3AZcC\nL7B69VZ27bpjagU6i5mks2mip6BLsjbJviT7k2wecv3FSXYkOZDkj5Nc+WpCa/Zs2XL/QHECXMrB\ng9vYsuV+M6kphw4d5tZbt3H99Vu59dZtHDp0eNqRlmxkoSdZBtwD3ARcA2xIcvWCYe8Dnquq7wb+\nHfDzZ7q/WdlRp354115728xkOqXb7U47wmnHjp3kpeLs9r9fyvHjJ6cTiNnMNGiWfn6nmGlxpx71\nPfDAh+h2r+eBBz7EjTd+dKZ6YRzjHKGvAQ5U1eGqOgHsANYtGLMO+Hh/+UHghjPd2SzsqMEf3pe+\ntGomMg2apV/0ubllwAv9tW7/+wusWDG959NnMRO8dJBw++2zd4Q3S79Ts7ifXv6or8usPOo7ta/G\nVlWLfgH/EPjYwPqtwL9fMOYpYMXA+gHgu4bcV0EVPF8bN87XtGzcOF/wfD/L1pnINGjr1q3TjnDa\nM898pVav/tn+/tpa8HytXv2z9cwzXzHTjGcaNCu/U7O6nzqdD/d7YLATqq6//sNTy/TyfUXViK6u\nqrGO0IdNxC98ZnPhmAwZM2CWHrKfMjsP2WfJqlUr2bXrDjZu/AhvecsfsHHjR6b+5OMsZnJefzyz\nup9e/qjvlOk+6nvlvhpt5KtcklwHzFfV2v76XfT+W9w9MOZz/TFfSHIR8GxV/c0h9+VLXCTpVagx\nXuUyzuvQdwNXJVkJPAusBzYsGPNfgNuALwD/CHj41QaSJL06Iwu9ql5MsgnYSe9J1Huram+SbcDu\nqvoMcC/w60kOAN+gV/qSpHPonL6xSJJ09pyzGf9Rb04615Lcm+RrSZ6cdpZTklye5OEkX07yVJI7\nZyDT65N8Icnj/Uxbp53plCTLkjyW5KFpZzklyVeSfKm/vx6ddh6AJN+R5FNJ9ibZk+T7p5znbf39\n81j/+5/PyO/6B5P8SZInkzyQ5OIZyPTT/b+78fpgnJfCvNYvev84/hRYCbwOeAK4+lxse5FM7wKu\nBZ6cZo4Fmd4MXNtfvgx4etr7qZ/lkv73i4BHgDXTztTP80HgE8BD084ykOkZ4DunnWNBpvuBn+wv\nLwfeMO1MA9mWAceBK6acY0X/Z3dxf/2TwE9MOdM1wJPA6/t/e7uA1Yvd5lwdoY/z5qRzqqr+CPjm\nNDMsVFV/VlVP9JefB/YCc9NNBVX1F/3F19MrhKnP0yW5HPj7wH+cdpYFwgx9immSvwb8UFXdB1BV\nf1VV35pyrEHvAQ5W1ZFpB6FXmpcmWQ5cQu8fzTR9D/BIVf1lVb0I/CHwo4vd4Fz94s0Bgz+wo8xA\nUc2yJG+h9wjiC9NNcnpq43Hgz4BdVbV72pmAXwD+FTPwz2WBAn4/ye4k7592GOCtwNeT3Nef4vhY\nkm+fdqgB/wT4zWmHqKrjwL8BvgocA/5PVX1+uqn4E+DdSb4zySX0DmCuWOwG56rQx3lzkvqSXEbv\nIxR+un+kPlVVdbKqvhe4HPj+JH9nmnmS/Ajwtf6jmTD892tafqCqvo/eH9+/TPKuKedZDrwD+KWq\negfwF8Bd043Uk+R1wC3Ap2YgyxvpzRqspDf9clmSfzrNTFW1D7gb+DzwWXpT1X+12G3OVaEfBQY/\ngfFypv9wZib1H+49CPx6Vf3utPMM6j9U7wJrpxzlB4FbkjxD7+ju+iS/NuVMQG/arP/9fwO/TW+6\ncZqOAkeq6n/21x+kV/Cz4L3AF/v7atreAzxTVc/1pzf+M/ADU85EVd1XVe+sqg69KeIDi40/V4V+\n+s1J/WeO1wOz8MqEWTu6A/hPwJer6henHQQgyZuSfEd/+dvp/eLvm2amqvq5qrqyqt5K73fp4ar6\niWlmAkhySf/RFUkuBf4evYfNU1NVXwOOJHlb/6IbgC9PMdKgDczAdEvfV4HrknxbktDbT3unnIkk\nf6P//Up68+eL7q9zcsaiOsObk87Fts8kyW8AHeCvJ/kqsPXUE0dTzPSDwEbgqf6cdQE/V1W/N8VY\nfxv4eP9jlJcBn6yqz04xzyz7W8Bv9z/iYjnwQFXtnHImgDuBB/pTHM8APznlPIMHB/9i2lkAqurR\nJA8CjwMn+t8/Nt1UAHw6yXfRy/SBqvrzxQb7xiJJasTMvLxKkvTaWOiS1AgLXZIaYaFLUiMsdElq\nhIUuSY2w0CWpERa6JDXi/wNZkfzwCuFFvgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f7abc229cf8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "#prediction for image 1\n",
    "plt.figure()\n",
    "plt.imshow(X_val[1],cmap='gray')\n",
    "plt.figure()\n",
    "plt.plot(Y_pred[1],'bo')\n",
    "print ( 'the correct class for image 1 is {}'.format(y_val[1]) )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {
    "collapsed": false,
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/ubuntu/anaconda2/envs/deep-learning/lib/python3.5/site-packages/keras/models.py:939: UserWarning: The `nb_epoch` argument in `fit` has been renamed `epochs`.\n",
      "  warnings.warn('The `nb_epoch` argument in `fit` '\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 50000 samples, validate on 10000 samples\n",
      "Epoch 1/1000\n",
      "50000/50000 [==============================] - 4s 85us/step - loss: 0.7807 - acc: 0.7505 - val_loss: 0.5617 - val_acc: 0.8115\n",
      "Epoch 2/1000\n",
      "50000/50000 [==============================] - 4s 78us/step - loss: 0.5248 - acc: 0.8226 - val_loss: 0.5122 - val_acc: 0.8233\n",
      "Epoch 3/1000\n",
      "50000/50000 [==============================] - 4s 85us/step - loss: 0.4771 - acc: 0.8357 - val_loss: 0.4722 - val_acc: 0.8353\n",
      "Epoch 4/1000\n",
      "50000/50000 [==============================] - 5s 91us/step - loss: 0.4518 - acc: 0.8446 - val_loss: 0.4515 - val_acc: 0.8422\n",
      "Epoch 5/1000\n",
      "50000/50000 [==============================] - 5s 104us/step - loss: 0.4340 - acc: 0.8506 - val_loss: 0.4436 - val_acc: 0.8430\n",
      "Epoch 6/1000\n",
      "50000/50000 [==============================] - 5s 98us/step - loss: 0.4200 - acc: 0.8545 - val_loss: 0.4250 - val_acc: 0.8516\n",
      "Epoch 7/1000\n",
      "50000/50000 [==============================] - 6s 118us/step - loss: 0.4076 - acc: 0.8588 - val_loss: 0.4150 - val_acc: 0.8569\n",
      "Epoch 8/1000\n",
      "50000/50000 [==============================] - 5s 105us/step - loss: 0.3970 - acc: 0.8627 - val_loss: 0.4191 - val_acc: 0.8542\n",
      "Epoch 9/1000\n",
      "50000/50000 [==============================] - 6s 129us/step - loss: 0.3902 - acc: 0.8647 - val_loss: 0.4048 - val_acc: 0.8599\n",
      "Epoch 10/1000\n",
      "50000/50000 [==============================] - 7s 143us/step - loss: 0.3813 - acc: 0.8676 - val_loss: 0.3971 - val_acc: 0.8625\n",
      "Epoch 11/1000\n",
      "50000/50000 [==============================] - 7s 135us/step - loss: 0.3747 - acc: 0.8700 - val_loss: 0.3983 - val_acc: 0.8588\n",
      "Epoch 12/1000\n",
      "50000/50000 [==============================] - 5s 104us/step - loss: 0.3688 - acc: 0.8713 - val_loss: 0.4025 - val_acc: 0.8579\n",
      "Epoch 13/1000\n",
      "50000/50000 [==============================] - 5s 110us/step - loss: 0.3628 - acc: 0.8739 - val_loss: 0.4032 - val_acc: 0.8597\n",
      "Epoch 14/1000\n",
      "50000/50000 [==============================] - 5s 99us/step - loss: 0.3571 - acc: 0.8753 - val_loss: 0.3913 - val_acc: 0.8642\n",
      "Epoch 15/1000\n",
      "50000/50000 [==============================] - 5s 96us/step - loss: 0.3526 - acc: 0.8772 - val_loss: 0.3795 - val_acc: 0.8681\n",
      "Epoch 16/1000\n",
      "50000/50000 [==============================] - 5s 97us/step - loss: 0.3477 - acc: 0.8784 - val_loss: 0.3924 - val_acc: 0.8605\n",
      "Epoch 17/1000\n",
      "50000/50000 [==============================] - 5s 100us/step - loss: 0.3428 - acc: 0.8796 - val_loss: 0.3766 - val_acc: 0.8647\n",
      "Epoch 18/1000\n",
      "50000/50000 [==============================] - 5s 97us/step - loss: 0.3392 - acc: 0.8811 - val_loss: 0.3719 - val_acc: 0.8671\n",
      "Epoch 19/1000\n",
      "50000/50000 [==============================] - 5s 96us/step - loss: 0.3345 - acc: 0.8823 - val_loss: 0.3961 - val_acc: 0.8569\n",
      "Epoch 20/1000\n",
      "50000/50000 [==============================] - 5s 96us/step - loss: 0.3303 - acc: 0.8830 - val_loss: 0.3709 - val_acc: 0.8688\n",
      "Epoch 21/1000\n",
      "50000/50000 [==============================] - 5s 99us/step - loss: 0.3265 - acc: 0.8847 - val_loss: 0.3580 - val_acc: 0.8742\n",
      "Epoch 22/1000\n",
      "50000/50000 [==============================] - 5s 96us/step - loss: 0.3221 - acc: 0.8862 - val_loss: 0.3631 - val_acc: 0.8724\n",
      "Epoch 23/1000\n",
      "50000/50000 [==============================] - 5s 96us/step - loss: 0.3188 - acc: 0.8877 - val_loss: 0.3537 - val_acc: 0.8741\n",
      "Epoch 24/1000\n",
      "50000/50000 [==============================] - 5s 100us/step - loss: 0.3156 - acc: 0.8888 - val_loss: 0.3573 - val_acc: 0.8756\n",
      "Epoch 25/1000\n",
      "50000/50000 [==============================] - 5s 104us/step - loss: 0.3119 - acc: 0.8903 - val_loss: 0.3681 - val_acc: 0.8737\n",
      "Epoch 26/1000\n",
      "50000/50000 [==============================] - 5s 102us/step - loss: 0.3089 - acc: 0.8911 - val_loss: 0.3528 - val_acc: 0.8747\n",
      "Epoch 27/1000\n",
      "50000/50000 [==============================] - 5s 99us/step - loss: 0.3062 - acc: 0.8930 - val_loss: 0.3501 - val_acc: 0.8764\n",
      "Epoch 28/1000\n",
      "50000/50000 [==============================] - 5s 106us/step - loss: 0.3030 - acc: 0.8926 - val_loss: 0.3568 - val_acc: 0.8733\n",
      "Epoch 29/1000\n",
      "50000/50000 [==============================] - 5s 101us/step - loss: 0.3006 - acc: 0.8941 - val_loss: 0.3529 - val_acc: 0.8776\n",
      "Epoch 30/1000\n",
      "50000/50000 [==============================] - 5s 97us/step - loss: 0.2973 - acc: 0.8958 - val_loss: 0.3590 - val_acc: 0.8727\n",
      "Epoch 31/1000\n",
      "50000/50000 [==============================] - 5s 98us/step - loss: 0.2949 - acc: 0.8953 - val_loss: 0.3479 - val_acc: 0.8785\n",
      "Epoch 32/1000\n",
      "50000/50000 [==============================] - 5s 96us/step - loss: 0.2929 - acc: 0.8966 - val_loss: 0.3464 - val_acc: 0.8757\n",
      "Epoch 33/1000\n",
      "50000/50000 [==============================] - 5s 101us/step - loss: 0.2892 - acc: 0.8983 - val_loss: 0.3481 - val_acc: 0.8770\n",
      "Epoch 34/1000\n",
      "50000/50000 [==============================] - 5s 97us/step - loss: 0.2873 - acc: 0.8980 - val_loss: 0.3398 - val_acc: 0.8796\n",
      "Epoch 35/1000\n",
      "50000/50000 [==============================] - 5s 95us/step - loss: 0.2843 - acc: 0.9003 - val_loss: 0.3392 - val_acc: 0.8814\n",
      "Epoch 36/1000\n",
      "50000/50000 [==============================] - 5s 96us/step - loss: 0.2821 - acc: 0.9003 - val_loss: 0.3403 - val_acc: 0.8816\n",
      "Epoch 37/1000\n",
      "50000/50000 [==============================] - 5s 96us/step - loss: 0.2798 - acc: 0.9016 - val_loss: 0.3496 - val_acc: 0.8751\n",
      "Epoch 38/1000\n",
      "50000/50000 [==============================] - 5s 99us/step - loss: 0.2776 - acc: 0.9027 - val_loss: 0.3397 - val_acc: 0.8798\n",
      "Epoch 39/1000\n",
      "50000/50000 [==============================] - 5s 98us/step - loss: 0.2753 - acc: 0.9022 - val_loss: 0.3349 - val_acc: 0.8834\n",
      "Epoch 40/1000\n",
      "50000/50000 [==============================] - 5s 99us/step - loss: 0.2729 - acc: 0.9043 - val_loss: 0.3364 - val_acc: 0.8793\n",
      "Epoch 41/1000\n",
      "50000/50000 [==============================] - 5s 99us/step - loss: 0.2708 - acc: 0.9037 - val_loss: 0.3574 - val_acc: 0.8752\n",
      "Epoch 42/1000\n",
      "50000/50000 [==============================] - 5s 98us/step - loss: 0.2687 - acc: 0.9059 - val_loss: 0.3392 - val_acc: 0.8799\n",
      "Epoch 43/1000\n",
      "50000/50000 [==============================] - 5s 102us/step - loss: 0.2667 - acc: 0.9060 - val_loss: 0.3325 - val_acc: 0.8825\n",
      "Epoch 44/1000\n",
      "50000/50000 [==============================] - 5s 103us/step - loss: 0.2650 - acc: 0.9057 - val_loss: 0.3343 - val_acc: 0.8828\n",
      "Epoch 45/1000\n",
      "50000/50000 [==============================] - 5s 97us/step - loss: 0.2631 - acc: 0.9076 - val_loss: 0.3340 - val_acc: 0.8818\n",
      "Epoch 46/1000\n",
      "50000/50000 [==============================] - 5s 103us/step - loss: 0.2611 - acc: 0.9077 - val_loss: 0.3388 - val_acc: 0.8801\n",
      "Epoch 47/1000\n",
      "50000/50000 [==============================] - 4s 75us/step - loss: 0.2592 - acc: 0.9090 - val_loss: 0.3379 - val_acc: 0.8810\n",
      "Epoch 00047: early stopping\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7f7abc0aee48>"
      ]
     },
     "execution_count": 74,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "## Let's try a more complicated model  - adding a hidden layer with 100 neurons and relu activation\n",
    "\n",
    "#let's monitor the validation loss, if the validation loss does not decrease during 4 epochs we will stop training\n",
    "early_stop = EarlyStopping(monitor='val_loss', patience=4, verbose=1) \n",
    "\n",
    "\n",
    "nn_model2 = Sequential()\n",
    "nn_model2.add(Dense(100, input_shape = (X_train_rolled.shape[1],)))\n",
    "nn_model2.add(Activation('relu'))\n",
    "\n",
    "nn_model2.add(Dense(10))\n",
    "nn_model2.add(Activation('softmax'))\n",
    "\n",
    "nn_model2.compile(loss='categorical_crossentropy',\n",
    "              optimizer='sgd',\n",
    "              metrics=['accuracy'])\n",
    "\n",
    "history = nn_model2.fit(X_train_rolled, Y_train, batch_size=32, \n",
    "          nb_epoch=1000,verbose=1,\n",
    "          validation_data=(X_val_rolled, Y_val),\n",
    "          callbacks=[early_stop])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "plt.plot(history.history['loss'])\n",
    "plt.plot(history.history['val_loss'])\n",
    "plt.title('model accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'test'], loc='upper left')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "acc is 0.881\n"
     ]
    }
   ],
   "source": [
    "#we stopped training, let's calculate accuracy on the validation set\n",
    "Y_pred = nn_model2.predict(X_val_rolled)\n",
    "#how many correct predictions\n",
    "correct_num = np.sum(Y_pred.argmax(axis=1)==y_val)\n",
    "print ('acc is {}'.format(correct_num/float(len(y_val))))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {
    "collapsed": false,
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/ubuntu/anaconda2/envs/deep-learning/lib/python3.5/site-packages/keras/models.py:939: UserWarning: The `nb_epoch` argument in `fit` has been renamed `epochs`.\n",
      "  warnings.warn('The `nb_epoch` argument in `fit` '\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 50000 samples, validate on 10000 samples\n",
      "Epoch 1/1000\n",
      "50000/50000 [==============================] - 16s 311us/step - loss: 0.7039 - acc: 0.7665 - val_loss: 0.5182 - val_acc: 0.8138\n",
      "Epoch 2/1000\n",
      "50000/50000 [==============================] - 16s 321us/step - loss: 0.4699 - acc: 0.8341 - val_loss: 0.4587 - val_acc: 0.8353\n",
      "Epoch 3/1000\n",
      "50000/50000 [==============================] - 16s 320us/step - loss: 0.4216 - acc: 0.8509 - val_loss: 0.4141 - val_acc: 0.8535\n",
      "Epoch 4/1000\n",
      "50000/50000 [==============================] - 16s 329us/step - loss: 0.3916 - acc: 0.8619 - val_loss: 0.4443 - val_acc: 0.8437\n",
      "Epoch 5/1000\n",
      "50000/50000 [==============================] - 17s 333us/step - loss: 0.3686 - acc: 0.8684 - val_loss: 0.4166 - val_acc: 0.8476\n",
      "Epoch 6/1000\n",
      "50000/50000 [==============================] - 12s 246us/step - loss: 0.3520 - acc: 0.8737 - val_loss: 0.4086 - val_acc: 0.8469\n",
      "Epoch 7/1000\n",
      "50000/50000 [==============================] - 16s 326us/step - loss: 0.3362 - acc: 0.8790 - val_loss: 0.3917 - val_acc: 0.8587\n",
      "Epoch 8/1000\n",
      "50000/50000 [==============================] - 15s 299us/step - loss: 0.3241 - acc: 0.8830 - val_loss: 0.3605 - val_acc: 0.8674\n",
      "Epoch 9/1000\n",
      "50000/50000 [==============================] - 16s 324us/step - loss: 0.3128 - acc: 0.8872 - val_loss: 0.3932 - val_acc: 0.8501\n",
      "Epoch 10/1000\n",
      "50000/50000 [==============================] - 22s 436us/step - loss: 0.3020 - acc: 0.8904 - val_loss: 0.3528 - val_acc: 0.8745\n",
      "Epoch 11/1000\n",
      "50000/50000 [==============================] - 23s 450us/step - loss: 0.2933 - acc: 0.8935 - val_loss: 0.3448 - val_acc: 0.8739\n",
      "Epoch 12/1000\n",
      "50000/50000 [==============================] - 21s 423us/step - loss: 0.2835 - acc: 0.8979 - val_loss: 0.3274 - val_acc: 0.8797\n",
      "Epoch 13/1000\n",
      "50000/50000 [==============================] - 25s 491us/step - loss: 0.2774 - acc: 0.8977 - val_loss: 0.3361 - val_acc: 0.8791\n",
      "Epoch 14/1000\n",
      "50000/50000 [==============================] - 25s 501us/step - loss: 0.2689 - acc: 0.9022 - val_loss: 0.3372 - val_acc: 0.8751\n",
      "Epoch 15/1000\n",
      "50000/50000 [==============================] - 26s 525us/step - loss: 0.2619 - acc: 0.9038 - val_loss: 0.3115 - val_acc: 0.8874\n",
      "Epoch 16/1000\n",
      "50000/50000 [==============================] - 25s 505us/step - loss: 0.2557 - acc: 0.9072 - val_loss: 0.3193 - val_acc: 0.8823\n",
      "Epoch 17/1000\n",
      "50000/50000 [==============================] - 22s 444us/step - loss: 0.2481 - acc: 0.9096 - val_loss: 0.3228 - val_acc: 0.8803\n",
      "Epoch 18/1000\n",
      "50000/50000 [==============================] - 29s 582us/step - loss: 0.2420 - acc: 0.9119 - val_loss: 0.3391 - val_acc: 0.8779\n",
      "Epoch 19/1000\n",
      "50000/50000 [==============================] - 30s 591us/step - loss: 0.2358 - acc: 0.9139 - val_loss: 0.3403 - val_acc: 0.8784\n",
      "Epoch 00019: early stopping\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7f7abc671c88>"
      ]
     },
     "execution_count": 76,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "## Let's try an even more complicated mode, 2 hidden layers with 512 neurons and another hidden layer with 64 neurons\n",
    "\n",
    "#let's monitor the validation loss, if the validation loss does not decrease during 4 epochs we will stop training\n",
    "early_stop = EarlyStopping(monitor='val_loss', patience=4, verbose=1) \n",
    "\n",
    "\n",
    "nn_model3 = Sequential()\n",
    "nn_model3.add(Dense(512, input_shape = (X_train_rolled.shape[1],)))\n",
    "nn_model3.add(Activation('relu'))\n",
    "nn_model3.add(Dense(512, input_shape = (X_train_rolled.shape[1],)))\n",
    "nn_model3.add(Activation('relu'))\n",
    "nn_model3.add(Dense(64, input_shape = (X_train_rolled.shape[1],)))\n",
    "nn_model3.add(Activation('relu'))\n",
    "\n",
    "\n",
    "nn_model3.add(Dense(10))\n",
    "nn_model3.add(Activation('softmax'))\n",
    "\n",
    "nn_model3.compile(loss='categorical_crossentropy',\n",
    "              optimizer='sgd',\n",
    "              metrics=['accuracy'])\n",
    "\n",
    "history = nn_model3.fit(X_train_rolled, Y_train, batch_size=32, \n",
    "          nb_epoch=1000,verbose=1,\n",
    "          validation_data=(X_val_rolled, Y_val),\n",
    "          callbacks=[early_stop])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7f7a9fa73f28>"
      ]
     },
     "execution_count": 103,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYYAAAEZCAYAAACTsIJzAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nPP4/tsfOUMPBq4oAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f7abc3dfb00>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(history.history['loss'])\n",
    "plt.plot(history.history['val_loss'])\n",
    "plt.title('model accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'test'], loc='upper left')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "acc is 0.8784\n"
     ]
    }
   ],
   "source": [
    "#we stopped training, let's calculate accuracy on the validation set\n",
    "Y_pred = nn_model3.predict(X_val_rolled)\n",
    "#how many correct predictions\n",
    "correct_num = np.sum(Y_pred.argmax(axis=1)==y_val)\n",
    "print ('acc is {}'.format(correct_num/float(len(y_val))))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#the accuracy is worse, perhpas I overfitted"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Ensemble"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 118,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "Y_pred1 = nn_model1.predict(X_val_rolled)\n",
    "Y_pred2 = nn_model2.predict(X_val_rolled)\n",
    "Y_pred3 = nn_model3.predict(X_val_rolled)\n",
    "\n",
    "\n",
    "Y_ensemble = (Y_pred1 + Y_pred2 + Y_pred3)/3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 119,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "acc is 0.8831\n"
     ]
    }
   ],
   "source": [
    "correct_num = np.sum(Y_ensemble.argmax(axis=1)==y_val)\n",
    "print ('acc is {}'.format(correct_num/float(len(y_val))))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 120,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# got a slight improvment "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Let's try regular ConvNets\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 95,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "out data sahpe is (50000, 28, 28, 1)\n",
      "we need to add a dummy dimension as the last dimension\n",
      "now our data shape is (50000, 28, 28, 1)\n"
     ]
    }
   ],
   "source": [
    "from keras.layers.convolutional import Convolution2D\n",
    "from keras.layers.pooling import MaxPooling2D\n",
    "from keras.layers.core import  Flatten\n",
    "\n",
    "print ('out data sahpe is {}'.format(X_train.shape))\n",
    "print ('we need to add a dummy dimension as the last dimension')\n",
    "X_train_orig = X_train.copy()\n",
    "X_val_orig = X_val.copy()\n",
    "\n",
    "X_train = X_train.reshape(X_train.shape[0],X_train.shape[1],X_train.shape[2],1)\n",
    "X_val = X_val.reshape(X_val.shape[0],X_val.shape[1],X_val.shape[2],1)\n",
    "\n",
    "print ('now our data shape is {}'.format(X_train.shape))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "metadata": {
    "collapsed": false,
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/ubuntu/anaconda2/envs/deep-learning/lib/python3.5/site-packages/ipykernel/__main__.py:5: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(20, (3, 3), padding=\"valid\", input_shape=(28, 28, 1...)`\n",
      "/home/ubuntu/anaconda2/envs/deep-learning/lib/python3.5/site-packages/keras/models.py:939: UserWarning: The `nb_epoch` argument in `fit` has been renamed `epochs`.\n",
      "  warnings.warn('The `nb_epoch` argument in `fit` '\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 50000 samples, validate on 10000 samples\n",
      "Epoch 1/1000\n",
      "50000/50000 [==============================] - 17s 336us/step - loss: 0.7538 - acc: 0.7421 - val_loss: 0.6260 - val_acc: 0.7720\n",
      "Epoch 2/1000\n",
      "50000/50000 [==============================] - 17s 347us/step - loss: 0.5225 - acc: 0.8148 - val_loss: 0.5169 - val_acc: 0.8150\n",
      "Epoch 3/1000\n",
      "50000/50000 [==============================] - 17s 347us/step - loss: 0.4795 - acc: 0.8323 - val_loss: 0.4712 - val_acc: 0.8269\n",
      "Epoch 4/1000\n",
      "50000/50000 [==============================] - 18s 358us/step - loss: 0.4479 - acc: 0.8426 - val_loss: 0.4710 - val_acc: 0.8355\n",
      "Epoch 5/1000\n",
      "50000/50000 [==============================] - 18s 358us/step - loss: 0.4247 - acc: 0.8537 - val_loss: 0.4461 - val_acc: 0.8352\n",
      "Epoch 6/1000\n",
      "50000/50000 [==============================] - 18s 362us/step - loss: 0.4052 - acc: 0.8604 - val_loss: 0.4801 - val_acc: 0.8250\n",
      "Epoch 7/1000\n",
      "50000/50000 [==============================] - 20s 396us/step - loss: 0.3893 - acc: 0.8667 - val_loss: 0.3871 - val_acc: 0.8634\n",
      "Epoch 8/1000\n",
      "50000/50000 [==============================] - 21s 426us/step - loss: 0.3769 - acc: 0.8705 - val_loss: 0.3804 - val_acc: 0.8657\n",
      "Epoch 9/1000\n",
      "50000/50000 [==============================] - 21s 416us/step - loss: 0.3651 - acc: 0.8740 - val_loss: 0.3833 - val_acc: 0.8642\n",
      "Epoch 10/1000\n",
      "50000/50000 [==============================] - 23s 461us/step - loss: 0.3545 - acc: 0.8770 - val_loss: 0.3577 - val_acc: 0.8769\n",
      "Epoch 11/1000\n",
      "50000/50000 [==============================] - 24s 478us/step - loss: 0.3469 - acc: 0.8796 - val_loss: 0.3551 - val_acc: 0.8744\n",
      "Epoch 12/1000\n",
      "50000/50000 [==============================] - 24s 482us/step - loss: 0.3402 - acc: 0.8823 - val_loss: 0.3453 - val_acc: 0.8805\n",
      "Epoch 13/1000\n",
      "50000/50000 [==============================] - 25s 502us/step - loss: 0.3339 - acc: 0.8842 - val_loss: 0.3673 - val_acc: 0.8703\n",
      "Epoch 14/1000\n",
      "50000/50000 [==============================] - 25s 509us/step - loss: 0.3281 - acc: 0.8869 - val_loss: 0.3343 - val_acc: 0.8843\n",
      "Epoch 15/1000\n",
      "50000/50000 [==============================] - 26s 516us/step - loss: 0.3229 - acc: 0.8879 - val_loss: 0.3307 - val_acc: 0.8876\n",
      "Epoch 16/1000\n",
      "50000/50000 [==============================] - 27s 533us/step - loss: 0.3176 - acc: 0.8888 - val_loss: 0.3426 - val_acc: 0.8810\n",
      "Epoch 17/1000\n",
      "50000/50000 [==============================] - 28s 553us/step - loss: 0.3122 - acc: 0.8916 - val_loss: 0.3271 - val_acc: 0.8852\n",
      "Epoch 18/1000\n",
      "50000/50000 [==============================] - 28s 556us/step - loss: 0.3087 - acc: 0.8929 - val_loss: 0.3317 - val_acc: 0.8828\n",
      "Epoch 19/1000\n",
      "50000/50000 [==============================] - 22s 442us/step - loss: 0.3049 - acc: 0.8939 - val_loss: 0.3195 - val_acc: 0.8904\n",
      "Epoch 20/1000\n",
      "50000/50000 [==============================] - 20s 407us/step - loss: 0.3009 - acc: 0.8951 - val_loss: 0.3138 - val_acc: 0.8899\n",
      "Epoch 21/1000\n",
      "50000/50000 [==============================] - 21s 418us/step - loss: 0.2971 - acc: 0.8966 - val_loss: 0.3290 - val_acc: 0.8842\n",
      "Epoch 22/1000\n",
      "50000/50000 [==============================] - 21s 416us/step - loss: 0.2929 - acc: 0.8994 - val_loss: 0.3531 - val_acc: 0.8727\n",
      "Epoch 23/1000\n",
      "50000/50000 [==============================] - 21s 416us/step - loss: 0.2894 - acc: 0.8993 - val_loss: 0.3124 - val_acc: 0.8902\n",
      "Epoch 24/1000\n",
      "50000/50000 [==============================] - 21s 420us/step - loss: 0.2869 - acc: 0.9015 - val_loss: 0.3060 - val_acc: 0.8932\n",
      "Epoch 25/1000\n",
      "50000/50000 [==============================] - 21s 418us/step - loss: 0.2835 - acc: 0.9018 - val_loss: 0.3149 - val_acc: 0.8896\n",
      "Epoch 26/1000\n",
      "50000/50000 [==============================] - 21s 416us/step - loss: 0.2800 - acc: 0.9031 - val_loss: 0.3027 - val_acc: 0.8937\n",
      "Epoch 27/1000\n",
      "50000/50000 [==============================] - 21s 420us/step - loss: 0.2772 - acc: 0.9037 - val_loss: 0.3003 - val_acc: 0.8933\n",
      "Epoch 28/1000\n",
      "50000/50000 [==============================] - 21s 415us/step - loss: 0.2750 - acc: 0.9040 - val_loss: 0.2982 - val_acc: 0.8968\n",
      "Epoch 29/1000\n",
      "50000/50000 [==============================] - 20s 401us/step - loss: 0.2720 - acc: 0.9052 - val_loss: 0.2992 - val_acc: 0.8946\n",
      "Epoch 30/1000\n",
      "50000/50000 [==============================] - 20s 407us/step - loss: 0.2692 - acc: 0.9067 - val_loss: 0.3196 - val_acc: 0.8878\n",
      "Epoch 31/1000\n",
      "50000/50000 [==============================] - 20s 407us/step - loss: 0.2669 - acc: 0.9080 - val_loss: 0.2911 - val_acc: 0.8970\n",
      "Epoch 32/1000\n",
      "50000/50000 [==============================] - 21s 418us/step - loss: 0.2643 - acc: 0.9084 - val_loss: 0.2933 - val_acc: 0.8955\n",
      "Epoch 33/1000\n",
      "50000/50000 [==============================] - 21s 412us/step - loss: 0.2618 - acc: 0.9092 - val_loss: 0.2969 - val_acc: 0.8931\n",
      "Epoch 34/1000\n",
      "50000/50000 [==============================] - 20s 410us/step - loss: 0.2600 - acc: 0.9096 - val_loss: 0.2952 - val_acc: 0.8946\n",
      "Epoch 35/1000\n",
      "50000/50000 [==============================] - 21s 428us/step - loss: 0.2576 - acc: 0.9110 - val_loss: 0.2915 - val_acc: 0.8955\n",
      "Epoch 00035: early stopping\n"
     ]
    }
   ],
   "source": [
    "#let's start with a simple convnet - one conv layer, followed by  relu operation, and then fully connected layer\n",
    "conv_model1 = Sequential()\n",
    "conv_model1.add(Convolution2D(20, 3, 3,\n",
    "                        border_mode='valid',\n",
    "                        input_shape=(28, 28,1)))\n",
    "conv_model1.add(Activation('relu'))\n",
    "conv_model1.add(MaxPooling2D( pool_size = (2,2)))\n",
    "conv_model1.add(Flatten())\n",
    "conv_model1.add(Dense(10))\n",
    "conv_model1.add(Activation('softmax'))\n",
    "\n",
    "conv_model1.compile(loss='categorical_crossentropy',\n",
    "              optimizer='sgd',\n",
    "              metrics=['accuracy'])\n",
    "\n",
    "history = conv_model1.fit(X_train, Y_train, batch_size=32, \n",
    "          nb_epoch=1000,verbose=1,\n",
    "          validation_data=(X_val, Y_val),\n",
    "          callbacks=[early_stop])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 104,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7f7a9f985f98>"
      ]
     },
     "execution_count": 104,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYYAAAEZCAYAAACTsIJzAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nPP4/tsfOUMPBq4oAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f7a9fa3ccc0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(history.history['loss'])\n",
    "plt.plot(history.history['val_loss'])\n",
    "plt.title('model accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'test'], loc='upper left')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "acc is 0.8955\n"
     ]
    }
   ],
   "source": [
    "Y_pred = conv_model1.predict(X_val)\n",
    "#how many correct predictions\n",
    "correct_num = np.sum(Y_pred.argmax(axis=1)==y_val)\n",
    "print ('acc is {}'.format(correct_num/float(len(y_val))))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 105,
   "metadata": {
    "collapsed": false,
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/ubuntu/anaconda2/envs/deep-learning/lib/python3.5/site-packages/ipykernel/__main__.py:6: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(20, (3, 3), padding=\"valid\", input_shape=(28, 28, 1...)`\n",
      "/home/ubuntu/anaconda2/envs/deep-learning/lib/python3.5/site-packages/ipykernel/__main__.py:12: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(20, (3, 3), padding=\"valid\", input_shape=(28, 28, 1...)`\n",
      "/home/ubuntu/anaconda2/envs/deep-learning/lib/python3.5/site-packages/keras/models.py:939: UserWarning: The `nb_epoch` argument in `fit` has been renamed `epochs`.\n",
      "  warnings.warn('The `nb_epoch` argument in `fit` '\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 50000 samples, validate on 10000 samples\n",
      "Epoch 1/1000\n",
      "50000/50000 [==============================] - 26s 511us/step - loss: 1.0382 - acc: 0.6334 - val_loss: 0.6725 - val_acc: 0.7359\n",
      "Epoch 2/1000\n",
      "50000/50000 [==============================] - 23s 455us/step - loss: 0.6053 - acc: 0.7783 - val_loss: 0.5666 - val_acc: 0.7880\n",
      "Epoch 3/1000\n",
      "50000/50000 [==============================] - 20s 397us/step - loss: 0.5259 - acc: 0.8099 - val_loss: 0.5254 - val_acc: 0.8089\n",
      "Epoch 4/1000\n",
      "50000/50000 [==============================] - 20s 399us/step - loss: 0.4860 - acc: 0.8275 - val_loss: 0.4914 - val_acc: 0.8213\n",
      "Epoch 5/1000\n",
      "50000/50000 [==============================] - 20s 397us/step - loss: 0.4612 - acc: 0.8346 - val_loss: 0.5481 - val_acc: 0.7856\n",
      "Epoch 6/1000\n",
      "50000/50000 [==============================] - 20s 398us/step - loss: 0.4420 - acc: 0.8417 - val_loss: 0.4552 - val_acc: 0.8333\n",
      "Epoch 7/1000\n",
      "50000/50000 [==============================] - 20s 398us/step - loss: 0.4254 - acc: 0.8464 - val_loss: 0.4370 - val_acc: 0.8407\n",
      "Epoch 8/1000\n",
      "50000/50000 [==============================] - 20s 404us/step - loss: 0.4107 - acc: 0.8527 - val_loss: 0.4060 - val_acc: 0.8511\n",
      "Epoch 9/1000\n",
      "50000/50000 [==============================] - 20s 396us/step - loss: 0.3989 - acc: 0.8567 - val_loss: 0.4036 - val_acc: 0.8536\n",
      "Epoch 10/1000\n",
      "50000/50000 [==============================] - 24s 477us/step - loss: 0.3878 - acc: 0.8597 - val_loss: 0.4432 - val_acc: 0.8316\n",
      "Epoch 11/1000\n",
      "50000/50000 [==============================] - 28s 565us/step - loss: 0.3793 - acc: 0.8639 - val_loss: 0.4054 - val_acc: 0.8532\n",
      "Epoch 12/1000\n",
      "50000/50000 [==============================] - 28s 565us/step - loss: 0.3715 - acc: 0.8677 - val_loss: 0.4013 - val_acc: 0.8525\n",
      "Epoch 13/1000\n",
      "50000/50000 [==============================] - 29s 579us/step - loss: 0.3635 - acc: 0.8710 - val_loss: 0.3787 - val_acc: 0.8630\n",
      "Epoch 14/1000\n",
      "50000/50000 [==============================] - 32s 631us/step - loss: 0.3571 - acc: 0.8723 - val_loss: 0.3688 - val_acc: 0.8672\n",
      "Epoch 15/1000\n",
      "50000/50000 [==============================] - 32s 639us/step - loss: 0.3504 - acc: 0.8748 - val_loss: 0.3933 - val_acc: 0.8527\n",
      "Epoch 16/1000\n",
      "50000/50000 [==============================] - 23s 462us/step - loss: 0.3447 - acc: 0.8775 - val_loss: 0.3668 - val_acc: 0.8665\n",
      "Epoch 17/1000\n",
      "50000/50000 [==============================] - 26s 511us/step - loss: 0.3404 - acc: 0.8785 - val_loss: 0.3629 - val_acc: 0.8694\n",
      "Epoch 18/1000\n",
      "50000/50000 [==============================] - 32s 634us/step - loss: 0.3352 - acc: 0.8804 - val_loss: 0.3550 - val_acc: 0.8706\n",
      "Epoch 19/1000\n",
      "50000/50000 [==============================] - 30s 604us/step - loss: 0.3308 - acc: 0.8812 - val_loss: 0.3847 - val_acc: 0.8632\n",
      "Epoch 20/1000\n",
      "50000/50000 [==============================] - 20s 406us/step - loss: 0.3260 - acc: 0.8840 - val_loss: 0.3474 - val_acc: 0.8711\n",
      "Epoch 21/1000\n",
      "50000/50000 [==============================] - 22s 437us/step - loss: 0.3232 - acc: 0.8848 - val_loss: 0.3949 - val_acc: 0.8555\n",
      "Epoch 22/1000\n",
      "50000/50000 [==============================] - 31s 624us/step - loss: 0.3203 - acc: 0.8851 - val_loss: 0.3746 - val_acc: 0.8653\n",
      "Epoch 23/1000\n",
      "50000/50000 [==============================] - 31s 625us/step - loss: 0.3156 - acc: 0.8872 - val_loss: 0.3319 - val_acc: 0.8813\n",
      "Epoch 24/1000\n",
      "50000/50000 [==============================] - 31s 627us/step - loss: 0.3131 - acc: 0.8878 - val_loss: 0.3457 - val_acc: 0.8786\n",
      "Epoch 25/1000\n",
      "50000/50000 [==============================] - 31s 628us/step - loss: 0.3100 - acc: 0.8892 - val_loss: 0.3818 - val_acc: 0.8587\n",
      "Epoch 26/1000\n",
      "50000/50000 [==============================] - 32s 631us/step - loss: 0.3071 - acc: 0.8895 - val_loss: 0.3235 - val_acc: 0.8828\n",
      "Epoch 27/1000\n",
      "50000/50000 [==============================] - 31s 630us/step - loss: 0.3041 - acc: 0.8904 - val_loss: 0.3250 - val_acc: 0.8831\n",
      "Epoch 28/1000\n",
      "50000/50000 [==============================] - 31s 629us/step - loss: 0.3023 - acc: 0.8914 - val_loss: 0.3267 - val_acc: 0.8826\n",
      "Epoch 29/1000\n",
      "50000/50000 [==============================] - 27s 532us/step - loss: 0.2990 - acc: 0.8935 - val_loss: 0.3218 - val_acc: 0.8849\n",
      "Epoch 30/1000\n",
      "50000/50000 [==============================] - 26s 515us/step - loss: 0.2969 - acc: 0.8936 - val_loss: 0.3265 - val_acc: 0.8837\n",
      "Epoch 31/1000\n",
      "50000/50000 [==============================] - 28s 565us/step - loss: 0.2946 - acc: 0.8953 - val_loss: 0.3493 - val_acc: 0.8736\n",
      "Epoch 32/1000\n",
      "50000/50000 [==============================] - 25s 496us/step - loss: 0.2921 - acc: 0.8958 - val_loss: 0.3177 - val_acc: 0.8849\n",
      "Epoch 33/1000\n",
      "50000/50000 [==============================] - 24s 482us/step - loss: 0.2899 - acc: 0.8967 - val_loss: 0.3193 - val_acc: 0.8861\n",
      "Epoch 34/1000\n",
      "50000/50000 [==============================] - 36s 715us/step - loss: 0.2879 - acc: 0.8974 - val_loss: 0.3137 - val_acc: 0.8848\n",
      "Epoch 35/1000\n",
      "50000/50000 [==============================] - 35s 699us/step - loss: 0.2861 - acc: 0.8986 - val_loss: 0.3210 - val_acc: 0.8849\n",
      "Epoch 36/1000\n",
      "50000/50000 [==============================] - 24s 477us/step - loss: 0.2841 - acc: 0.8994 - val_loss: 0.3105 - val_acc: 0.8864\n",
      "Epoch 37/1000\n",
      "50000/50000 [==============================] - 25s 501us/step - loss: 0.2821 - acc: 0.8994 - val_loss: 0.3125 - val_acc: 0.8863\n",
      "Epoch 38/1000\n",
      "50000/50000 [==============================] - 26s 513us/step - loss: 0.2805 - acc: 0.9000 - val_loss: 0.3076 - val_acc: 0.8905\n",
      "Epoch 39/1000\n",
      "50000/50000 [==============================] - 25s 509us/step - loss: 0.2782 - acc: 0.9007 - val_loss: 0.3159 - val_acc: 0.8877\n",
      "Epoch 40/1000\n",
      "50000/50000 [==============================] - 27s 549us/step - loss: 0.2768 - acc: 0.9009 - val_loss: 0.3170 - val_acc: 0.8876\n",
      "Epoch 41/1000\n",
      "50000/50000 [==============================] - 26s 522us/step - loss: 0.2742 - acc: 0.9021 - val_loss: 0.3186 - val_acc: 0.8862\n",
      "Epoch 42/1000\n",
      "50000/50000 [==============================] - 25s 491us/step - loss: 0.2727 - acc: 0.9017 - val_loss: 0.3655 - val_acc: 0.8663\n",
      "Epoch 00042: early stopping\n"
     ]
    }
   ],
   "source": [
    "#a more complicated - one conv layer, followed by  relu operation, followed by another conv layer,\n",
    "# and then fully connected layer\n",
    "conv_model2 = Sequential()\n",
    "conv_model2 .add(Convolution2D(20, 3, 3,\n",
    "                        border_mode='valid',\n",
    "                        input_shape=(28, 28,1)))\n",
    "conv_model2.add(Activation('relu'))\n",
    "conv_model2.add(MaxPooling2D( pool_size = (2,2)))\n",
    "\n",
    "conv_model2 .add(Convolution2D(20, 3, 3,\n",
    "                        border_mode='valid',\n",
    "                        input_shape=(28, 28,1)))\n",
    "conv_model2.add(Activation('relu'))\n",
    "conv_model2.add(MaxPooling2D( pool_size = (2,2)))\n",
    "\n",
    "\n",
    "conv_model2.add(Flatten())\n",
    "conv_model2.add(Dense(10))\n",
    "conv_model2.add(Activation('softmax'))\n",
    "\n",
    "conv_model2.compile(loss='categorical_crossentropy',\n",
    "              optimizer='sgd',\n",
    "              metrics=['accuracy'])\n",
    "\n",
    "history = conv_model2.fit(X_train, Y_train, batch_size=32, \n",
    "          nb_epoch=1000,verbose=1,\n",
    "          validation_data=(X_val, Y_val),\n",
    "          callbacks=[early_stop])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 106,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7f7a9f6a7780>"
      ]
     },
     "execution_count": 106,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYYAAAEZCAYAAACTsIJzAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzt3Xl4VOX1wPHvSSBhS0KAyBIgIFEWsQLuokIBCwgKIqig\nIlSxtbW2Vav2pxXs5oa7orYgKIqAggguLAqRIgi4gCAEkH2HACEhhBCS8/vj3gkzIcsEZjKZ5Hye\n5z5z13fOXPGevO9773tFVTHGGGM8IkIdgDHGmIrFEoMxxhgflhiMMcb4sMRgjDHGhyUGY4wxPiwx\nGGOM8WGJwVQJIjJeRP7u576bRaRbsGMypqKyxGCMMcaHJQZjwoiIRIY6BlP5WWIwFYbbhPOgiKwU\nkUwR+a+InCUin4lIhojMFZE4r/2vF5HVInJQROaLSBuvbR1F5DsROSwik4Eahb6rr4j8ICKHRGSR\niJzvZ4zXisj3brlbRWRkoe1XisjXbrlbRWSou76GiDwnIlvcbQtFJFpEuojI9iLOQzd3fqSIfCAi\nE0UkHbhDRC4WkcVuOTtF5BURqeZ1/HnuuTogIrtF5BERaSgiWSIS77XfhSKyz5KNKcwSg6loBgDd\ngXOB64HPgEeA+kAkcB+AiJwLTHKXE4DPgVkiUk1EqgMfAW8D9YAPgBs9XyAinYBxwAh3+5vATPe4\n0hwBblfVOKAP8FsRud4tt7kb70tAA6ADsMI97jmgI3CZ+50PAfnuttLGpbkemKqqdYH3gBPAn9xy\nLge6Ab9zY6gDzHPjaAwkA1+q6l5gAXCTV7m3Au+rap4fv9tUIZYYTEXziqqmqepu4H/AUlX9UVVz\ncS72Hd39bgI+UdX57oVtNE6t4Aqci281VX1ZVfNUdRqw3Os77gLeUNVv1TERyHGPK5GqLlTVn9z5\n1cBkoIu7eQgwT1Wnut97SFV/FBEBhgP3qeoe9zu/cX+TP5ao6iz3O3NU9QdVXeaWsw34j1cMfYHd\nqvqiqh5X1SxV9fz2d4DbAUQkAhgMTPQzBlOFWGIwFc1er/nsIpbruPNNgK2eDeqMBrkDSHS37SxU\n7lav+STgAbcJ6qCIHAKauseVSEQucZut9rlNO7/BqR0ANAM2FnFYAyAa2FRa+cUo3NR0jojMcpuJ\n0oF/+REDwMdAWxFpAfwKSFfVb08zJlOJWWIw4WoXzgXeWzOchLAb50LvrbnX/HbgX6paz53iVbWO\nqk7x43snATOARLdp501AvMpNLuKYNOAY0KqIbVlALc+C296fUGifwk1NrwNrgVZuDI/6EQOqmgNM\nBW5zJ6stmCJZYjDhairQR0R+6fYrPIhz8V0MLAFyReQPIhIpIgOAS7yO/S9O38AlACJS2+1Uru3H\n99YBDqlqrnv8EK9t7wHdRWSg+731ROQCtzYzHnheRBqLSISIXOb2aawHaohIb7cD+TEgqpQYYoAM\nVT3qdrjf47XtE6ChiNwnIlEiUsfzO10TgWHAdcC7fvxeUwVZYjAVSeG/jIvtlFXV9Th/9b4K7Mfp\nCL5OVU+4bfcDcNr1DwKDgGlex36H0/H8qogcxLk43+HP9+J08v5DRA7jXMQLahmquh24FnjQ/d4f\ngF+4mx8EVuH0dRwAngIiVDXDLXMcTlNYpvtZkgeBW0UkA6fGMtkrhiPANTgd1nvc39bVa/tinE7v\n793+CWNOIfaiHmOqFhH5EnhPVd8KdSymYrLEYEwVIiIXA3OAZqqaFep4TMVkTUnGVBEiMgGYC/zR\nkoIpidUYjDHG+LAagzHGGB/VSt+lYhARq9oYY8xpUFUpfa+TwqrGoKoVaho5cmTIYwiHmCpqXBaT\nxVQV4jodYZUYjDHGBJ8lBmOMMT4sMZyBrl27hjqEU1TEmKBixmUx+cdi8l9FjauswuZ2VRHRcInV\nGGMqChFBy9j5HDZ3JRWnRYsWbN26tfQdq6ikpCS2bNkS6jCMMWEk7GsMbjYMQUThwc6PMVXb6dQY\nrI/BGGOMD0sMxhhjfFhiMMYY48MSQwV3zz338K9//SvUYRhjqhDrfA6yli1bMm7cOLp16xaS76/o\n58cYE1zW+Rxm8vLyQh2CMcacwhJDEA0dOpRt27bRt29fYmNjefbZZ4mIiOCtt94iKSmJ7t27A3DT\nTTfRuHFj4uPj6dq1K2vWrCkoY/jw4Tz++OMAfPXVVzRr1oznn3+ehg0bkpiYyIQJE0Lx04wxlZgl\nhiB65513aN68OZ9++ikZGRncdNNNACxcuJDU1FTmzJkDwLXXXsvGjRvZt28fnTp14tZbby22zD17\n9pCZmcmuXbsYO3Ysv//97zl8+HC5/B5jTNVQ6RODSGCmM+Hdxi8iPPHEE9SsWZPo6GgAhg0bRq1a\ntahevTqPP/44K1euJDMzs8iyoqKi+Nvf/kZkZCS9e/emTp06rFu37swCNMYYL5U+MagGZgqkpk2b\nFszn5+fzyCOPkJycTN26dWnZsiUiQlpaWpHH1q9fn4iIk//ZatWqxZEjRwIboDGmSqv0iSHUpIjq\nhve6SZMmMWvWLObPn096ejpbtmw5oxdsGGPMmbLEEGSNGjVi06ZNAEVe8DMzM4mOjiY+Pp6srCz+\n+te/FplMjDGmvAQ1MYjIOBHZKyI/lrDPyyKyQURWiEiHYMYTCo888gj/+Mc/qFevHtOmTTvloj90\n6FCaN29OYmIi7du354orrihT+ZZEjDGBFtQH3ETkSuAI8I6q/qKI7b2Be1W1j4hcCrykqpcVU5Zm\nZys1apyy3ppdSmDnx5iqrcI94Kaqi4BDJezSD3jH3XcpECciDYvb+VBJJRljjAmIUPcxJALbvZZ3\nuuuKdPBg0OMxxpgqL9RvcCuqelNsu8fo0aNISnLmu3btWmner2qMMYGSkpJCSkrKGZUR9EH0RCQJ\nmFVMH8MbwAJVneIupwJdVHVvEfvqxx8r119/ynprQy+BnR9jqrYK18fgEoquGQDMBIYCiMhlQHpR\nScHDmpKMMSb4gtqUJCKTgK5AfRHZBowEogBV1f+o6mcicq2I/AxkAcNLKs8SgzHGBF9QE4OqDvFj\nn3v9Lc8SgzHGBF+o70oqE7td1Rhjgi+sEoPVGIwxJvgsMQRZy5YtmT9//hmV8fbbb3PVVVcFKCJj\njCmZJYYwoKo2JpIxptyEVWIItz4Gz6s9r7vuOmJjYxk9ejRLly6lc+fOxMfH07FjR7766quC/SdM\nmECrVq2IjY2lVatWvP/++6SmpnLPPfewZMkSYmJiqFevXgh/kTGmSvAMBV3RJ0Dj4/UUzk+ouFq0\naKHz589XVdWdO3dq/fr1dfbs2aqq+sUXX2j9+vU1LS1Ns7KyNDY2Vjds2KCqqnv27NE1a9aoquqE\nCRP0qquuOq3vr+jnxxgTXO41oEzX21APiVEmGRmQlweRkf4fI08EpglGR57+08PqPnn87rvv0qdP\nH3r27AlA9+7dueiii/jss8+48cYbiYyMZNWqVTRt2pSGDRvSsGGx4wkaY0zQhFViiImBw4ehLK0p\nZ3JBD7StW7cydepUZs2aBTgJ48SJE3Tr1o1atWoxZcoUnn32WX79619z5ZVXMnr0aFq3bh3iqI0x\nVU1Y9THUqxd+/QzencbNmjVj6NChHDx4kIMHD3Lo0CEyMzN56KGHALjmmmuYO3cue/bsoXXr1tx9\n992nlGGMMcEWVokhPj787kzyfrXnbbfdxqxZs5g7dy75+fkcO3aMr776il27drFv3z5mzZrF0aNH\nqV69OnXq1CHSbTNr2LAhO3bsIDc3N5Q/xRhTRYRVYqhXL/wSg/erPadOncrHH3/Mv//9bxISEkhK\nSmL06NHk5+eTn5/Pc889R2JiIg0aNGDhwoWMGTMGgG7dunHeeefRqFEjzjrrrBD/ImNMZRf0YbcD\nRUT05puVfv1g8GCf9TasdAns/BhTtVXUYbcDJhz7GIwxJtyEVWIIxz4GY4wJN2GVGMKxj8EYY8KN\nJQZjjDE+wioxxMdbH4MxxgRbWCUGqzEYY0zwhdWQGEUlhqSkJHsyuARJSUmhDsEYE2bCLjEUbkra\nsmVLSGIxxpjKKqyakjy3q9rzWsYYEzxhlRhq1gQRyM4OdSTGGFN5hVViAOuANsaYYAvLxGC3rBpj\nTPCEXWKwYTGMMSa4wi4xWFOSMcYElyUGY4wxPsIyMVgfgzHGBE/YJQbrYzDGmOAKu8RgTUnGGBNc\nlhiMMcb4CMvEYH0MxhgTPGGXGKyPwRhjgivsEoM1JRljTHBZYjDGGONDNEzGsBYRVVXy8yEqCo4d\ng2ph9TYJY4wpfyKCqpbpbWZhV2OIiIC4OEhPD3UkxhhTOYVdYgBrTjLGmGCyxGCMMcZHWCaG+Hh7\nlsEYY4IlLBOD1RiMMSZ4gp4YRKSXiKSKyHoRebiI7c1EZL6IfC8iK0Skd2llWmIwxpjgCWpiEJEI\n4FWgJ3AeMFhE2hTa7TFgiqp2AgYDY0or14bFMMaY4Al2jeESYIOqblXVXGAy0K/QPvlArDtfF9hZ\nWqE2LIYxxgRPsBNDIrDda3mHu87bE8DtIrId+AT4Q2mFWlOSMcYET7CfHS7qabvCj1oPBsar6gsi\nchnwLk6z0ylGjRoFwLp1sGVLV6BroOI0xphKISUlhZSUlDMqI6hDYrgX+lGq2stdfgRQVX3aa5/V\nQE9V3ekubwQuVdW0QmWpJ9avv4aHHnI+jTHGFK8iDomxHEgWkSQRiQJuAWYW2mcr0ANARNoC0YWT\nQmHWx2CMMcET1MSgqnnAvcBc4CdgsqquFZEnRKSvu9uDwAgRWQG8B9xRWrnWx2CMMcETdqOrAuTk\nQJ06cPw4SJkqSMYYU7VUxKakoIiOdqasrFBHYowxlU9YJgawfgZjjAmWsE0M1s9gjDHBYYnBGGOM\nj7BODDZekjHGBF7YJgbrYzDGmOAI28RgTUnGGBMclhiMMcb4COvEYH0MxhgTeGGbGKyPwRhjgiNs\nE4M1JRljTHBYYjDGGOPDr8QgItNEpI/7DucKIT7e+hiMMSYY/L3Qvw4MATaIyFMi0iaIMfnFagzG\nGBMcZRp2W0TicF7F+SjOu5z/C7yrqrnBCc/nu9U7VlWIinJGWI2KCva3G2NMeArqsNsiUh8YBtwF\n/AC8BHQC5pXlCwNFxJqTjDEmGKr5s5OITAfaABOB61R1t7tpioh8G6zgSuNJDA0bhioCY4ypfPxK\nDMCrqjq/qA2qelEA4ykT62cwxpjA87cpqa2I1PUsiEi8iPwuSDH5zRKDMcYEnr+JYYSqpnsWVPUQ\nMCI4IfnPhsUwxpjA8zcxRIhIQa+2iEQCIb8XyIbFMMaYwPO3j2EOMFVE3gAU+C0wO2hR+cmakowx\nJvD8TQwPA78B7gEEmAuMDVZQ/qpXDzZsCHUUxhhTufiVGFQ1H+fp59eDG07ZWB+DMcYEnr/PMZwD\nPAm0A2p41qvq2UGKyy/Wx2CMMYHnb+fzeJzawgngl8A7wLvBCspf1sdgjDGB529iqKmqX+KMrbRV\nVUcBfYIXln8sMRhjTOD52/l8zB1ye4OI3AvsBOoELyz/WB+DMcYEnr81hj8BtYD7gAuB24A7ghVU\ncbakb/FZrlvXSQz5+eUdiTHGVF6lJgb3YbabVfWIqu5Q1eGqeqOqflMO8fmY/bPvoxPVq0OtWpCZ\nWd6RGGNM5VVqYlDVPODKcoilVJ///Pkp66yfwRhjAsvfPoYfRGQm8AGQ5VmpqtODElUxUrakkHMi\nh+hq0QXrPP0MLVuWZyTGGFN5+dvHUAM4AHQDrnOnvsEKqjhtG7Tl6+1f+6yzZxmMMSaw/H3yeXiw\nA/FHr+RefL7hc7q17FawzpqSjDEmsPx98nk8zuB5PlT11wGPqAS9k3tz16y7eJZnC9ZZYjDGmMDy\nt4/hE6/5GsANwK7Ah1Oyi5pcxO7M3Ww/vJ1mcc0Ae5bBGGMCza8+BlWd5jW9B9wElPsrPSMjIvlV\nq18xZ+OcgnXWx2CMMYHlb+dzYecAZwUyEH/1Su7lc9uqNSUZY0xg+ZUYRCRTRDI8EzAL5x0N5a5n\nq57M3zyf3LxcwBKDMcYEmr93JcUEOxB/NazTkFbxrViyYwlXJ11NfLz1MRhjTCD5W2O4QUTivJbr\nikh/P4/tJSKpIrJeRIqsZYjITSLyk4isEpFSh/P23LYKVmMwxphA87ePYaSqHvYsqGo6MLK0g9wR\nWV8FegLnAYNFpE2hfZJxmqUuV9XzcQbsK1Hv5N7M3uiMm2SJwRhjAsvfxFDUfv40Q10CbHDf4ZAL\nTAb6FdpnBPCaqmYAqGpaaYVe2vRStqZvZXfmbksMxhgTYP4mhm9F5HkRaSUiZ4vIC8B3fhyXCGz3\nWt7hrvN2LtBaRBaJyGIR6VlaodUiqtHj7B7M2TiHWrWgdm3YtMnPX2KMMaZE/iaGPwDHgSnAVCAb\n+L0fx0kR6wo/QV0NSAauBoYAY0UktrSCPbetisCQIfD2235EY4wxplT+3pWUBTxyGuXvAJp7LTfl\n1CemdwBLVDUf2CIi63CekzilRjJq1KiC+faXtGfexnmcyD/B8OHV6N8fRo6EiNN9MsMYYyqBlJQU\nUlJSzqgMUT1lCKRTdxKZBwxyO50RkXhgsqqW2OzjvuRnHdAd2A0sAwar6lqvfXq664aJSAOchNBB\nVQ8VKksLx3rBGxfwep/XuaLZFXTsCKNHQ/fupf9oY4ypKkQEVS2q9aZY/v593cCTFADci3apTz67\nL/m5F5gL/ISTTNaKyBMi0tfdZw5wQER+Ar4EHiycFIrTO7l3wVvdhg2D8eP9/DXGGGOK5W+N4Tvg\nBlXd5i63AKaraqegRucbwyk1hpQtKfxl3l9YPmI5aWmQnAxbt0JcXDGFGGNMFRPMGsOjwCIRmSgi\nE4GvgL+WNcBAu6LZFWw4sIF9Wfto0MBpRpo6NdRRGWNMePN3dNXZOKOprsO5M+kBnDuTQioqMopf\ntvwlczfOBWD4cGtOMsaYM+XvkBh34bT/P+BOE4FRwQvLf72TexeMttqrF2zeDKmpIQ7KGGPCmL9N\nSX8ELga2quovgY5AesmHlI9eyb2Y8/McsnOzqVYNbrsNJkwIdVTGGBO+/E0Mx1T1GICIRKtqKtA6\neGH5r3lcc65pdQ0Pzn0QcJqT3nkHTpwIcWDGGBOm/E0MO0SkLjADmCciHwNbgxdW2bze53U++/kz\nZqTOoF07aNYM5s0LdVTGGBOe/Lpd1ecAkS5AHDBbVY8HJaqiv/eU21W9Ldm+hP5T+vPtiG/59P1m\nzJ9vdygZY8zp3K5a5sQQKqUlBoAn//ckszfOZvr182l1diSbNjnDcp8uVUWkTOfTGGMqlGA+xxAW\nHur8ENUiqvHaqn9x7bUwadLpl6WqXPf+dXy45sPABWiMMWGgUiWGyIhIJt4wkTHLx3DRgEVn9EzD\nvE3zmLtxLhN/nBi4AI0xJgxUqsQA0CSmCeOuH8dL229lT/ohfvyx7GXkaz5//fKvjOkzhgWbF5CZ\nkxn4QI0xpoKqdIkBoM+5fbih7Q3E3HYX4yeUvQ9l2pppCMKdHe+kc/POfLbhsyBEaYwxFVOlTAwA\nT/d4moj6mxj3w3/IzfX/uNy8XB6d/yhPdn8SEWFAmwFMWzsteIEaY0wFU2kTQ3S1aD4aMpnsKx7j\nlSmr/T5uwooJNItrRo+zewDQv01/5mx0nqw2xpiqoNImBoDWDVrzQLsXeWjFdazavLvU/bNzs/n7\nwr8X1BYAEmon0Klxp4KB+owxprKr1IkB4Klbb6VzzbvoPKYPh7NL7kR+bflrXNzkYi5JvMRn/Y1t\nb7TmJGNMlVGpHnArTm6u0vz3vyWm6VZ+enQW1SOrn7LP4WOHOeeVc0gZlkK7hHY+23Zm7OT8189n\nz4N7iIqMOq0YjDEmFKr8A27FqV5d+OZvr7FtSzWuH3s3RSWY0YtH0+fcPqckBYDE2ERaN2jNgs0L\nyiNcY4wJqSqRGACSmlVjysApfLlqNQ9+Ospn294jexnz7RhGdRlV5LFgzUnGmKqjyiQGgH7X1uZ3\ndT/h9UXv8p9vxxas/9f//sXtv7idpLpJxR47oO0AZqTOIC8/rzxCNcaYkKkW6gDK2+hRDVl83ec8\nUPNqmsY1oV1CO95b9R5rf7+2xOPOjj+bxNhEFm1bRJcWXcopWmOMKX9VLjFUqwYzxp3L+dd+xJDq\n19MxsT33XnwvZ9U+q9RjPc1JlhiMMZVZlWpK8mjSBCY/ezkRn/yXTQe28sAVD/h13I1tb2T62unk\na36QIzTGmNCpkokB4Jpr4N4e/Wk5ayMRubF+HdM2oS2x0bEs27ksyNEZY0zoVNnEADByJJyTLHTu\nDFv9fFHpgLYDmL52enADM8aYEKrSiSEyEv7zH7jzTrj8cli0qPRjPP0M4fJgoDHGlFWVTgwAInDf\nfTBhAtx4I7z1Vsn7d2jUAVVl5d6V5RKfMcaUtyqfGDx+9StYuBCeegruvx9OnCh6PxEp6IQ2xpjK\nyBKDl9atYelSWLUKrrsODh8uer8Bbe0dDcaYyssSQyHx8fD553DOOXDppfDTT6fuc2nTS0k/lk5q\nWmr5B2iMMUFmiaEI1arByy/Dww9D164wahTk5JzcHiERzpvd1litwRhT+VhiKMHw4bBihTN17AiL\nF5/cNvj8wby09CUmr55sdygZYyqVKvE+hjOlCtOmOXcvDRgA//43xMbC4u2LuefTe0iolcBr175G\n6watQxKfMcYUx97HECQiMHCg099w7Bi0bw+ffAJXNLuC7+7+jr7n9qXzW515bP5jHM09GvR4wiWZ\nG2PCk9UYTsP8+XD33dCpEzz9NLRs6bzl7YG5D7Bs5zJe7v0yfc/tG5TvvvezezmQfYBJAyYVvJfa\nGGOKYzWGctKtG/z4o1NzuOgiePBBqJWXyOSBk3mz75vcP+d++k/uz4GjBwL6ve/++C7zNs1j06FN\nPLXoqYCWbYwxHpYYTlOtWvD4407z0pEjzjMQzz8PVze9hlX3rKJF3RZcM/Ea0o+lB+T71uxfw5/n\n/JkPB33I9Jum88qyV5j98+yAlG2MMd4sMZyhRo3gjTcgJQUWLIC2bWHGtGie/9ULdEnqQs93e5KR\nk3FG35F1PItBHwzimR7PcH7D80mMTWTKwCncMeMONh3aFJgfYowxLutjCLCUFKdpKTIS/v53ZcaJ\n37Nq74/Mvm02daLqlLk8VeWOGXcQGRHJ+H7jfba9svQVxv4wlsW/XkztqNoB+gXGmMrkdPoYLDEE\nQX4+TJ0Kf/87xMTmE3f7CE7U2cynQz6lZvWaZSpr3PfjeHHpiyy9aym1qtfy2aaqDPt4GMfzjltn\ntDGmSJYYKpj8fOf5h7//I49dl95BUts0vv7DDGpWr+HX8Sv3rKTHxB78b/j/aNOgTZH7ZOdmc+X4\nK7n1/Fu5//L7Axm+MaYSqJB3JYlILxFJFZH1IvJwCfsNFJF8EekU7JjKS0QEDBoEK1dE8mbvCWxO\njaHRH25i8gfHyS/l7aAZORkM+mAQL/V6qdikAFCzek2m3zSdZ75+hvmb5wf4FxhjqqKg1hhEJAJY\nD3QHdgHLgVtUNbXQfnWAT4HqwL2q+n0RZYVdjaGw4ydyufrVgaxPrU6DlMmMuLMaQ4dCw4a++6kq\nN394M/Vr1uf1vq/7Vfb8zfMZMm0IS+9aSlLdpCBEb4wJRxWuKUlELgNGqmpvd/kRQFX16UL7vQDM\nA/4CPFBZEwNAzokc+k/uz7o92zmx/2z2bEqg5VkJXH1hAp07JtCwTgLf7f6O6Wuns/jOxdSo5l+z\nE8DoxaMZ98M4BrYdSKt6rWgV34pW9VrRuE5j638wpoqqiInhRqCnqt7tLt8GXKKq93nt0wF4VFUH\nicgCKnliADied5zlO5ez/+h+th3Yz4KlaSxZuZ/MvP00Tt5Pg0bHeO+mcbSq16pM5aoq09ZOY/W+\n1Ww8tJGNBzey8dBGjhw/Qsu6LWlVrxV3dryT61tfH6RfZoypaE4nMVQLVjCuooIpuLqL82fsC8Ad\npRwDwKhRowrmu3btSteuXc84wFCIioyic/POBcv3ubMrV8LYsTDpFbj7A7jjDmfQvjp+3uUqIgxs\nN5CB7Qb6rM/MyWTToU2s2b+GP87+IylbUniqx1NERUYF6ieFHVW1WpSplFJSUkhJSTmjMsqjKWmU\nqvZyl32akkQkFvgZOIKTEBoBB4DrC9caKlONoTTHjsGsWfD227BoEfTr5ySJrl2dDu0zcTD7IMM/\nHs7eI3uZMnBKQPojlu5Yypebv+Thzg8TGRF5xuUFm6rS671eREgEY68bS2JsYqhDMiZoKuJdScuB\nZBFJEpEo4BZgpmejqmao6lmqeraqtgS+Aa4rqimpKqlRw7mb6ZNPIDUVOnRw3kPdsiU89pjz6tHT\nzZH1atZjxs0zGNRuEJeMvYRZ62adUaxvr3ib696/jpnrZjLog0Fk52afUXnl4e2Vb5N2NI3Lm15O\nxzc7MnHlRBux1hhvqhrUCegFrAM2AI+4654A+hax73ygUzHlaFW3YoXq/fertmyp2rSp6l13qU6b\nppqefnrlLd62WJu/0FwfnPOgHj9xvEzH5ubl6v2z79fkl5P1p30/ac6JHB0ybYhePvZyTctKO72A\nymjJ9iW6fOfyMh2z78g+PevZs/S7Xd+pqur3u77X9mPaa7/3++mezD3BCNOYkHKvnWW6btsDbmFI\nFdavh9mznfdTf/21MwR4r17OdMEF/jc5HTh6gKEzhnIo+xBTBk6hWVyzUo85lH2IW6bdQr7mM2Xg\nFOrVrAdAvubzf1/+Hx+lfsTsW2fTMr7lmfzMEn2x6QsGTxtMtYhqLLlzCS3qtvDruNs/up2GtRsy\n+lejC9blnMjhia+e4K0f3uKV3q8w6LxBQYk563gWv/nkNzzZ/Um/zrMxgXA6TUlBrzEEasJqDMXK\nylL97DPVP/xB9ZxzVBs0UB00SPWNN1Q3bFDNzy/5+Lz8PH160dMa92Sc3jjlRp28arJm5mQWue+a\nfWv0nJfP0T9+/kfNzcstcp9Xl76qTZ5rot/u/PZMf1qR5vw8RxOeSdCFWxbqC0te0Atev0CzjmeV\netzcn+fPT4O5AAAUd0lEQVRq0gtJeiTnSJHbv9n+jbZ+pbXe/MHNAa/15OXn6YApA7Tx6MY6YuaI\ngJZtTEk4jRpDyC/4fgdqicFv27apTpigevvtqk2aqDZvrjpsmOrEiaq7dhV/XFpWmo79bqz2nNhT\nY5+M1QFTBuj7q94vSBKfrv9UE55J0HHfjys1hulrpmvCMwn6+YbPA/WzVFX18w2fa8IzCbpo6yJV\nVc3Pz9fbp9+ut3x4i+aXkAGzjmfp2S+drZ+u/7TE8o8eP6p/nv1nPfulszU9+zTb6Irw2JePaedx\nnXV35m6t/3R93XBgQ8DKNqYklhjMKfLzVdeuVX31VdUbblCNj1c97zzVP/1J9ZNPVDMyij7uwNED\nOu77cdrr3V4a+2Ssdn+7uzYe3Vi/3va139/99bavteGzDXXc9+NKvGj765N1n2jCMwm6eNtin/VH\njx/VC9+8UJ9Z9Eyxxz4872G95cNb/P6uu2ferUM/GnrasXqb9OMkbfFiC917ZK+qqj6R8oTePv32\ngJRtTGlOJzFYH0MVk5cH330HX3zhTMuXQ8eO0KOH82a6Cy+EmoUGgD2YfZC5G+dyZfMraRrbtEzf\nty5tHf0m92NHxg5axrekZV13cufPjj+b5HrJpY46O2vdLO6ceSczB8/ksqaXnbJ9++HtXDr2Usb3\nG0/P5J4+21buWck1E50XKDWs0/CUY4uSdTyLjm925J/d/slN593k/w8uZNnOZfSZ1Icvh37JLxr+\nAnDGwUp+OZmvhn1F24S2p122CR8fp37Mmv1r+OtVfy33765wTz4HkiWG4MjKcp6V+OIL50VDa9Y4\nLxu69NKT07nnnvnzE4ePHWZz+mY2H9rs87np0Ca2Hd7GRU0uonvL7nQ/uzsXN7mY6pHVC479OPVj\n7v7kbmYNnsUliZcU+x0Lty5k0AeD+PrXX5NcLxmAvPw8Lh93OXdfeDd3dbqrTDEv37mcvu/35bu7\nvytzQgTnPeCXjr2U1659jX5t+vlse3rR03y/53umDJxS5nJNeDl87DDnjTmPSTdO4uqkq8v9+y0x\nmDOWnQ0//ABLl8KyZc7nwYNw8cVw2WVw+eVOsqhfP3DfmZmTyf+2/Y8vN33JF5u/YEv6Fq5qfhXd\nW3andlRt/rbgb3w65FMuanJRqWWNWT6GMcvH8M1d31Anqg6vLH2FD9d+yII7FhAhZc9u/1z4T1K2\npDD39rllOv5o7lGuHn81A9sN5JErHzlle9bxLJJfSWbObXMKahKmcvrtJ78F4I2+b4Tk+y0xmKDY\nt89JEt9840zLlkHjxicTxWWXQfv2UC1AA6zsy9rHgs0L+HLzl6zet5pXr32VTo39G41dVRkxawSH\njh3ihZ4v0OnNTiz69aIShy4vyYn8E3SZ0IWBbQfy58v/7HcMt0y7hajIKN7p/06xQ2+8+M2LpGxJ\nYcYtM04rNlPxLdy6kCHThvDT734irkZcSGKwxGDKRV6e0+S0ZImTKJYsge3bneTQsePJ6fzznae4\ny1vOiRy6vt2Vrelb+c2Fv2Fk15FnVN6mQ5u4dOylzB86n/Mbnl/q/k+kPMHsjbNZcMeCEkfHPXbi\nGMkvJ/PRzR9xceLFZxRjsMzbOI/ZP8/m6WueplpEsIdWq1yOnTjGBW9cwNM9nqZ/m/4hi8MSgwmZ\nw4edQQB/+OHktH49JCc7Q3q0aQPnnHNy8ndgwNO1K3MXo1JG8UrvV4iuFn3G5U1YMYHnlzzPshHL\nir3Ybz60meeWPMes9bNYetdSGtVpVGq5ry9/nZnrZ/L5rZ+fcYyBlJefxz8W/oP/fv9fWsW3okXd\nFkzoP+G0muOqqsfmP0ZqWiof3vRhSOOwxGAqlJwc+OknWLEC1q2DDRucaeNGiIs7mSRat3ZqG+3b\nQ7NmUBEHPVVVBn0wiKS4JJ7r+ZzPtu92fcezi5/li01fMKLTCP542R/9SgrgDMHe+tXWvHvDuz4j\n7obS3iN7uXX6reRpHu/f+D6x0bH0fq837Rq0Y0yfMTYqrR9+3PsjPd7pwcrfrqRxTOOQxmKJwYSF\n/HzYuRN+/tlJFGvXOglk9WrnLqnzzjuZKM47z0kcTZqc+Z1RZ+rA0QNc8MYFTOg/ge4tuzN341ye\nWfwMGw5s4E+X/YkRnUYQEx1T5nLf+uEt3v3xXebfEfpXs3raxId3GM7IriMLmo8ycjLo8U4PuiR1\n4ZlrnrHkUIIzuRMuGCwxmLB34MDJJLF6tTO/fj1kZDjNUuec49w+6/ls3RoaNCi/+OZtnMewj4fR\noFYD8jWfh654iFva3+Jze21Zncg/QdvX2vJm3zfp1rJbAKP1X77m8+zXz/LCNy8wof8EeiX3OmWf\ng9kH6TqhKwPbDeTxLo+HIMrw8OI3L/Lxuo+ZP3R+hUiglhhMpZWR4dQw1q93ahnr1zvTunUQFeU8\ne9GunTN55hs3Dk6z1PgfxtM4pjE9W/UM2P/47/34Hq8tf42vf/11mcpUVbJys8jIySAmKua0aixp\nR9MYNmMYB7MPljqQ4t4je7lq/FXcc9E9ft+lVZVsSd/CRf+5iCV3LuGc+ueEOhzAEoOpglRh926n\nOWrNmpOfa9Y4Lzxq0gQaNSp+atzYqXFEhvj9Qnn5efzijV8wpP0QEmonkJmTSUZOBpnHM5354xkF\n6w7nHCYjJ8PZnpNJjWo1iImOISMng/o169MuoR1tG7SlXUK7gqlujbrszdpLaloq69LWkZqWSuqB\nVFLTUtlzZA+/u+h3PNXjKb9qPtsOb+Pq8Vfz6FWPMuLCEeVwdsKDqtL7vd50bdG1yGdXQsUSgzFe\n0tOdpLFnz6nT7t2wd6/zmZ7uJIfGjU8mi8REp5mqbVunyap27eDHu2jbIl5d9mrBX/4xUTHERscW\nzMdExxAXHUdcjThio2OJi44jJjqmoB8gX/PZmr6VNfvXOFPamoL5vPw8akfVpk2DNrSu35o2DdoU\nTC3qtijzrag/H/yZLhO68Ow1zzLk/CGoKjl5OQUJyzNFR0bT/qz2ftdkdmTsYOHWhSzcupCoyCgG\ntx/MZU0vqxBNMsXJy88j83gmH/z0Aa8tf43lI5afUdNioFliMOY05OY6D/Ht3n0ykWzf7rw9LzXV\nabo66yznlts2bZyEkZgICQknp7i4ink3FTh/yWYezyQ2Ojag5a7et5ru73TnRP4JMnIyiJRIYqNj\nfaas3CzW7F9Dk5gmdGjUgQsaXkCHRh3o0KgDiTGJbE7fzFdbvmLhNicZHD52mKuTrqZLUhcycjKY\ntHoSx04cY3D7wQw5fwjtz2of0N/gbUfGDr7Y9AW7MneRnZtN9olsjuYePfmZm01WblZBzc0zZZ/I\nJiYqhvq16jN14FQubHJh0GI8HZYYjAmCvDzYutVJEmvXOp+7d0NaGuzf70zHjjm1joQEJ4kkJjpT\n06Yn5xMTnW2hvrsqkDJzMsnJyyEmKqbY50VO5J9gw4ENrNizwpn2Op+ZOZnUrVGXLi26cHXzq+nS\nogttGrTxeVZCVVm5dyWTVk3i/dXvE18jniHnD+H61tejqhzMPsihY4c4lH3I57NujboFTWqtG7Qu\n8tmT7NxsFm5dyJyNc5izcQ57j+ylx9k9aFm3JbWq16Jm9ZrOZ7WaPsuempon+dWuXrtC12gsMRgT\nIseOOYkiLc2pcezcWfSUnu4kB+9mK+9Pz3zjxhB95s/lVWjpx9KJi47z+6Kar/ks2raISasmMWfj\nHGpUq0F8jXjq1axHfM144mu4U814DmYfZM3+NaxNW8vGgxtpFtesoN+lbo26LNiygMXbF9OhUQd6\ntupJz1Y96dS4E5ERIe5sCgJLDMZUcDk5Tt+Gp5/Du8/D8+np/6hT52SS8CSOevWgbt2ip/r1oXrF\nadquMHLzctl4aGNBf0va0TS6JHWhW8tuIRu/qDxZYjCmksjPd57p8CQKT+I4dMipdRSePOsbNHCa\nrzyTpzmrSRNnW716zlT4nRum8rLEYEwVduKEkzx27HCarXbsODnt2uUkmoMHnc/IyJNJon79k1OD\nBs7kmfd8JiRATEzF7WA3xbPEYIwplarz3g3vRHHw4Mk+kgMHfOf373fmjx8/mSQSEk7OexKIdzLx\nzNeqFepfaywxGGOCxtPB7rkTyzMdOHBy8iQTz7wqxMc7U716J+c9U1yc0z8SF3fqVLeu81S7OTOW\nGIwxFcrRo07/R+Hp4EHn8/BhZ0pPPznvvS462kkgdeue+ulpCvMkHe/luLjKf1eXvywxGGMqDVVn\ntF3vDnfPvHeC8Uzey4cPO2XExJw6xcb69qd496U0aOAklTp1Kk9txRKDMca4cnIgM/PU6fBh32Yv\nT3+KZ8rIcPaLjHQSSZ06Jz/j4pznUM466+TDjJ55zxPwMTEV67ZhSwzGGBMAqr6J5cgR5zM9/WTf\nyr59p356kkr16qfWVPr3hwceKP/fcjqJwV7iaowxhYg47yuvUcOpCZSF564v71pKRobTXBUurMZg\njDGV2OnUGCrRcF7GGGMCwRKDMcYYH5YYjDHG+LDEYIwxxoclBmOMMT4sMRhjjPFhicEYY4wPSwzG\nGGN8WGIwxhjjwxKDMcYYH5YYjDHG+Ah6YhCRXiKSKiLrReThIrb/WUR+EpEVIjJPRJoFOyZjjDHF\nC2piEJEI4FWgJ3AeMFhE2hTa7XvgQlXtAEwDng1mTIGUkpIS6hBOURFjgooZl8XkH4vJfxU1rrIK\ndo3hEmCDqm5V1VxgMtDPewdV/UpVj7mL3wCJQY4pYCriP4KKGBNUzLgsJv9YTP6rqHGVVbATQyKw\n3Wt5ByVf+O8EPg9qRMYYY0oU7Bf1FDUGeJEvVRCR24ALgS5BjcgYY0yJgvqiHhG5DBilqr3c5UcA\nVdWnC+3XA3gJuFpVDxRTlr2lxxhjTkOFeueziEQC64DuwG5gGTBYVdd67dMR+ADoqaobgxaMMcYY\nvwS1j0FV84B7gbnAT8BkVV0rIk+ISF93t2eA2sAHIvKDiMwIZkzGGGNKFjbvfDbGGFM+wuLJ59Ie\nkgsFEdkiIivdWs6yEMUwTkT2isiPXuviRWSuiKwTkTkiElcBYhopIjtE5Ht36lXOMTUVkfkiskZE\nVonIfe76kJ2rImL6g7s+1OcqWkSWuv+uV4nISHd9CxH5xj1X74tIsG9c8Sem8SKyyV3/vYj8orxi\n8ootwv3ume5yyM5ToZh+8IppQpnPk6pW6Aknef0MJAHVgRVAmwoQ1yYgPsQxXAl0AH70Wvc08JA7\n/zDwVAWIaSRwfwjPUyOggztfB6ffq00oz1UJMYX0XLnx1HI/I3GeLboUmAIMcte/DvymAsQ0HhgQ\n4nP1Z+BdYKa7HNLzVExM44EbylJGONQYSn1ILkSEENe4VHURcKjQ6n7A2+7820D/ChATFH3rcrlQ\n1T2qusKdPwKsBZoSwnNVTEyeZ3xCdq4AVPWoOxuNc0u7Ar/EGZkAnHN1Q4hjyneXQ3auRKQpcC0w\n1mt1N0J4noqJCcp4rQqHxFDWh+TKiwJzRGS5iIwIdTBezlLVveBcfICEEMfj8Xt3PKyx5d285U1E\nWuDUaL4BGlaEc+UV01J3VUjPlacpAtgDzAM2Aumq6rkY7wCahDImVV3ubvqne66eE5Hq5RkT8ALw\nF9xns0SkPnAolOepcExeynSewiEx+P2QXDm7QlUvwsnOvxeRK0MdUAU2BmilznhYe4DnQxGEiNQB\nPgT+6P6VHvJ/R0XEFPJzpar5qtoRp1Z1CdC2qN1CGZOItAMeUdW2wMVAfZzmwHIhIn2AvW6tz3ON\nEk69XpXbeSomJjiN8xQOiWEH0NxruSmwK0SxFHD/wkRV9wMf4fwPVBHsFZGGACLSCNgX4nhQ1f3q\nNnYC/8X5B1qu3E7AD4GJqvqxuzqk56qomCrCufJQ1QzgK+AyoK44g2JCCP8f9Iqpl1dtLxenHb08\n/x/sDFwvIpuA93GakF4E4kJ4nk6JSUTeOZ3zFA6JYTmQLCJJIhIF3ALMDGVAIlLL/UsPEakN/ApY\nHapw8P3rYCYwzJ2/A/i48AHlwCcm96LrMYDQnKu3gDWq+pLXulCfq1NiCvW5EpEGnuYrEakJ9ADW\nAAuAQe5u5Xquiokp1XOuRERw+ofK7Vyp6v+panNVPRvnmjRfVW8jhOepmJiGns55KvdbqcpKVfNE\nxPOQXAQwTr2enA6RhsBH4gzTUQ14T1XnlncQIjIJ6ArUF5FtOHe0PIXzsOCvgW2c/Ecayph+KSId\ncDoMtwC/KeeYOgO3AqvcdmoF/g/nrqSpoThXJcQ0JJTnCmgMvO3+1RsBTFHVz0RkLTBZRP4B/ACM\nqwAxfSkiDXD+CFkB/LYcYyrOI4TuPBXnvbKeJ3vAzRhjjI9waEoyxhhTjiwxGGOM8WGJwRhjjA9L\nDMYYY3xYYjDGGOPDEoMxxhgflhiMKQci0kVEZoU6DmP8YYnBmPJjDw2ZsGCJwRgvInKr+1KY70Xk\ndXdUz0wReV5EVovIPHcUTUSkg4gscUetnOY1bEMrd78VIvKtiLR0i48RkQ9EZK2ITAzZjzSmFJYY\njHGJSBvgZpyRczvhDEtxK1ALWKaq7YGFOMN8gDPe/l/ckVBXe61/D3jFXX8FsNtd3wG4D2gHtBKR\nK4L/q4wpuwo/VpIx5ag70AlY7g44VgPYi5Mgprr7vAtME5FYIM59MRE4SWKqO7hioqrOBFDV4wBO\ncSxT1d3u8gqgBbC4HH6XMWViicGYkwR4W1Uf9Vkp8rdC+6nX/kWVUZwcr/k87P8/U0FZU5IxJ30J\nDBSRBAARiReR5jjvGR7o7nMrsMh9L8BBd5RUgNuBr1Q1E9guIv3cMqLcoaKNCRv2F4sxLlVdKyKP\nAXPdIZ6PA/cCWThvDfsbTtPSze4hdwBvuhf+TcBwd/3twH9E5O9uGUUN5213KJkKy4bdNqYUIpKp\nqjGhjsOY8mJNScaUzv56MlWK1RiMMcb4sBqDMcYYH5YYjDHG+LDEYIwxxoclBmOMMT4sMRhjjPFh\nicEYY4yP/wekUJqvZOSGdwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f7a9f9e5080>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(history.history['loss'])\n",
    "plt.plot(history.history['val_loss'])\n",
    "plt.title('model accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'test'], loc='upper left')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 107,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "acc is 0.8663\n"
     ]
    }
   ],
   "source": [
    "Y_pred = conv_model2.predict(X_val)\n",
    "#how many correct predictions\n",
    "correct_num = np.sum(Y_pred.argmax(axis=1)==y_val)\n",
    "print ('acc is {}'.format(correct_num/float(len(y_val))))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 108,
   "metadata": {
    "collapsed": false,
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/ubuntu/anaconda2/envs/deep-learning/lib/python3.5/site-packages/ipykernel/__main__.py:5: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(20, (3, 3), padding=\"valid\", input_shape=(28, 28, 1...)`\n",
      "/home/ubuntu/anaconda2/envs/deep-learning/lib/python3.5/site-packages/keras/models.py:939: UserWarning: The `nb_epoch` argument in `fit` has been renamed `epochs`.\n",
      "  warnings.warn('The `nb_epoch` argument in `fit` '\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 50000 samples, validate on 10000 samples\n",
      "Epoch 1/1000\n",
      "50000/50000 [==============================] - 25s 501us/step - loss: 1.0610 - acc: 0.6082 - val_loss: 0.6214 - val_acc: 0.7671\n",
      "Epoch 2/1000\n",
      "50000/50000 [==============================] - 22s 434us/step - loss: 0.7234 - acc: 0.7297 - val_loss: 0.5574 - val_acc: 0.7790\n",
      "Epoch 3/1000\n",
      "50000/50000 [==============================] - 25s 493us/step - loss: 0.6480 - acc: 0.7602 - val_loss: 0.4965 - val_acc: 0.8164\n",
      "Epoch 4/1000\n",
      "50000/50000 [==============================] - 21s 417us/step - loss: 0.5943 - acc: 0.7800 - val_loss: 0.4690 - val_acc: 0.8299\n",
      "Epoch 5/1000\n",
      "50000/50000 [==============================] - 21s 412us/step - loss: 0.5544 - acc: 0.7979 - val_loss: 0.4388 - val_acc: 0.8381\n",
      "Epoch 6/1000\n",
      "50000/50000 [==============================] - 27s 549us/step - loss: 0.5303 - acc: 0.8087 - val_loss: 0.4254 - val_acc: 0.8438\n",
      "Epoch 7/1000\n",
      "50000/50000 [==============================] - 21s 422us/step - loss: 0.5081 - acc: 0.8139 - val_loss: 0.4065 - val_acc: 0.8523\n",
      "Epoch 8/1000\n",
      "50000/50000 [==============================] - 28s 560us/step - loss: 0.4927 - acc: 0.8241 - val_loss: 0.3919 - val_acc: 0.8600\n",
      "Epoch 9/1000\n",
      "50000/50000 [==============================] - 31s 610us/step - loss: 0.4780 - acc: 0.8277 - val_loss: 0.3845 - val_acc: 0.8646\n",
      "Epoch 10/1000\n",
      "50000/50000 [==============================] - 32s 632us/step - loss: 0.4679 - acc: 0.8323 - val_loss: 0.3741 - val_acc: 0.8623\n",
      "Epoch 11/1000\n",
      "50000/50000 [==============================] - 30s 590us/step - loss: 0.4520 - acc: 0.8380 - val_loss: 0.3612 - val_acc: 0.8686\n",
      "Epoch 12/1000\n",
      "50000/50000 [==============================] - 24s 473us/step - loss: 0.4465 - acc: 0.8409 - val_loss: 0.3573 - val_acc: 0.8724\n",
      "Epoch 13/1000\n",
      "50000/50000 [==============================] - 29s 589us/step - loss: 0.4339 - acc: 0.8452 - val_loss: 0.3463 - val_acc: 0.8724\n",
      "Epoch 14/1000\n",
      "50000/50000 [==============================] - 24s 488us/step - loss: 0.4275 - acc: 0.8463 - val_loss: 0.3437 - val_acc: 0.8764\n",
      "Epoch 15/1000\n",
      "50000/50000 [==============================] - 29s 583us/step - loss: 0.4223 - acc: 0.8490 - val_loss: 0.3402 - val_acc: 0.8764\n",
      "Epoch 16/1000\n",
      "50000/50000 [==============================] - 23s 457us/step - loss: 0.4169 - acc: 0.8538 - val_loss: 0.3364 - val_acc: 0.8793\n",
      "Epoch 17/1000\n",
      "50000/50000 [==============================] - 22s 439us/step - loss: 0.4118 - acc: 0.8513 - val_loss: 0.3289 - val_acc: 0.8804\n",
      "Epoch 18/1000\n",
      "50000/50000 [==============================] - 31s 625us/step - loss: 0.4025 - acc: 0.8559 - val_loss: 0.3304 - val_acc: 0.8785\n",
      "Epoch 19/1000\n",
      "50000/50000 [==============================] - 33s 656us/step - loss: 0.3941 - acc: 0.8572 - val_loss: 0.3201 - val_acc: 0.8833\n",
      "Epoch 20/1000\n",
      "50000/50000 [==============================] - 32s 633us/step - loss: 0.3913 - acc: 0.8586 - val_loss: 0.3235 - val_acc: 0.8820\n",
      "Epoch 21/1000\n",
      "50000/50000 [==============================] - 32s 646us/step - loss: 0.3888 - acc: 0.8606 - val_loss: 0.3140 - val_acc: 0.8860\n",
      "Epoch 22/1000\n",
      "50000/50000 [==============================] - 29s 580us/step - loss: 0.3813 - acc: 0.8622 - val_loss: 0.3121 - val_acc: 0.8847\n",
      "Epoch 23/1000\n",
      "50000/50000 [==============================] - 22s 445us/step - loss: 0.3800 - acc: 0.8632 - val_loss: 0.3097 - val_acc: 0.8873\n",
      "Epoch 24/1000\n",
      "50000/50000 [==============================] - 27s 539us/step - loss: 0.3772 - acc: 0.8646 - val_loss: 0.3052 - val_acc: 0.8885\n",
      "Epoch 25/1000\n",
      "50000/50000 [==============================] - 25s 502us/step - loss: 0.3730 - acc: 0.8657 - val_loss: 0.3028 - val_acc: 0.8920\n",
      "Epoch 26/1000\n",
      "50000/50000 [==============================] - 32s 637us/step - loss: 0.3638 - acc: 0.8685 - val_loss: 0.2996 - val_acc: 0.8915\n",
      "Epoch 27/1000\n",
      "50000/50000 [==============================] - 25s 508us/step - loss: 0.3646 - acc: 0.8681 - val_loss: 0.3000 - val_acc: 0.8902\n",
      "Epoch 28/1000\n",
      "50000/50000 [==============================] - 23s 463us/step - loss: 0.3600 - acc: 0.8724 - val_loss: 0.2985 - val_acc: 0.8898\n",
      "Epoch 29/1000\n",
      "50000/50000 [==============================] - 24s 475us/step - loss: 0.3610 - acc: 0.8685 - val_loss: 0.2965 - val_acc: 0.8917\n",
      "Epoch 30/1000\n",
      "50000/50000 [==============================] - 23s 456us/step - loss: 0.3565 - acc: 0.8719 - val_loss: 0.2921 - val_acc: 0.8936\n",
      "Epoch 31/1000\n",
      "50000/50000 [==============================] - 21s 418us/step - loss: 0.3533 - acc: 0.8723 - val_loss: 0.2873 - val_acc: 0.8944\n",
      "Epoch 32/1000\n",
      "50000/50000 [==============================] - 24s 477us/step - loss: 0.3520 - acc: 0.8748 - val_loss: 0.2915 - val_acc: 0.8923\n",
      "Epoch 33/1000\n",
      "50000/50000 [==============================] - 23s 468us/step - loss: 0.3532 - acc: 0.8741 - val_loss: 0.2873 - val_acc: 0.8957\n",
      "Epoch 34/1000\n",
      "50000/50000 [==============================] - 30s 606us/step - loss: 0.3472 - acc: 0.8743 - val_loss: 0.2836 - val_acc: 0.8962\n",
      "Epoch 35/1000\n",
      "50000/50000 [==============================] - 31s 616us/step - loss: 0.3466 - acc: 0.8756 - val_loss: 0.2877 - val_acc: 0.8935\n",
      "Epoch 36/1000\n",
      "50000/50000 [==============================] - 24s 478us/step - loss: 0.3433 - acc: 0.8766 - val_loss: 0.2829 - val_acc: 0.8962\n",
      "Epoch 37/1000\n",
      "50000/50000 [==============================] - 23s 460us/step - loss: 0.3378 - acc: 0.8766 - val_loss: 0.2811 - val_acc: 0.8974\n",
      "Epoch 38/1000\n",
      "50000/50000 [==============================] - 25s 507us/step - loss: 0.3383 - acc: 0.8776 - val_loss: 0.2790 - val_acc: 0.8974\n",
      "Epoch 39/1000\n",
      "50000/50000 [==============================] - 31s 611us/step - loss: 0.3390 - acc: 0.8752 - val_loss: 0.2816 - val_acc: 0.8949\n",
      "Epoch 40/1000\n",
      "50000/50000 [==============================] - 25s 505us/step - loss: 0.3347 - acc: 0.8772 - val_loss: 0.2782 - val_acc: 0.8959\n",
      "Epoch 41/1000\n",
      "50000/50000 [==============================] - 23s 466us/step - loss: 0.3320 - acc: 0.8793 - val_loss: 0.2749 - val_acc: 0.8981\n",
      "Epoch 42/1000\n",
      "50000/50000 [==============================] - 24s 473us/step - loss: 0.3314 - acc: 0.8797 - val_loss: 0.2723 - val_acc: 0.8997\n",
      "Epoch 43/1000\n",
      "50000/50000 [==============================] - 23s 454us/step - loss: 0.3289 - acc: 0.8809 - val_loss: 0.2751 - val_acc: 0.8979\n",
      "Epoch 44/1000\n",
      "50000/50000 [==============================] - 23s 464us/step - loss: 0.3281 - acc: 0.8812 - val_loss: 0.2780 - val_acc: 0.8969\n",
      "Epoch 45/1000\n",
      "50000/50000 [==============================] - 28s 569us/step - loss: 0.3238 - acc: 0.8826 - val_loss: 0.2698 - val_acc: 0.9009\n",
      "Epoch 46/1000\n",
      "50000/50000 [==============================] - 31s 626us/step - loss: 0.3257 - acc: 0.8817 - val_loss: 0.2721 - val_acc: 0.9012\n",
      "Epoch 47/1000\n",
      "50000/50000 [==============================] - 26s 518us/step - loss: 0.3220 - acc: 0.8835 - val_loss: 0.2685 - val_acc: 0.9021\n",
      "Epoch 48/1000\n",
      "50000/50000 [==============================] - 29s 579us/step - loss: 0.3207 - acc: 0.8836 - val_loss: 0.2696 - val_acc: 0.9024\n",
      "Epoch 49/1000\n",
      "50000/50000 [==============================] - 23s 450us/step - loss: 0.3189 - acc: 0.8836 - val_loss: 0.2679 - val_acc: 0.9043\n",
      "Epoch 50/1000\n",
      "50000/50000 [==============================] - 24s 482us/step - loss: 0.3196 - acc: 0.8825 - val_loss: 0.2642 - val_acc: 0.9037\n",
      "Epoch 51/1000\n",
      "50000/50000 [==============================] - 22s 441us/step - loss: 0.3167 - acc: 0.8835 - val_loss: 0.2682 - val_acc: 0.9017\n",
      "Epoch 52/1000\n",
      "50000/50000 [==============================] - 26s 516us/step - loss: 0.3160 - acc: 0.8832 - val_loss: 0.2687 - val_acc: 0.9010\n",
      "Epoch 53/1000\n",
      "50000/50000 [==============================] - 24s 482us/step - loss: 0.3141 - acc: 0.8863 - val_loss: 0.2645 - val_acc: 0.9002\n",
      "Epoch 54/1000\n",
      "50000/50000 [==============================] - 23s 452us/step - loss: 0.3098 - acc: 0.8874 - val_loss: 0.2639 - val_acc: 0.9033\n",
      "Epoch 55/1000\n",
      "50000/50000 [==============================] - 27s 546us/step - loss: 0.3125 - acc: 0.8864 - val_loss: 0.2634 - val_acc: 0.9039\n",
      "Epoch 56/1000\n",
      "50000/50000 [==============================] - 25s 509us/step - loss: 0.3103 - acc: 0.8855 - val_loss: 0.2647 - val_acc: 0.9041\n",
      "Epoch 57/1000\n",
      "50000/50000 [==============================] - 29s 573us/step - loss: 0.3083 - acc: 0.8885 - val_loss: 0.2631 - val_acc: 0.9050\n",
      "Epoch 58/1000\n",
      "50000/50000 [==============================] - 24s 483us/step - loss: 0.3081 - acc: 0.8880 - val_loss: 0.2608 - val_acc: 0.9062\n",
      "Epoch 59/1000\n",
      "50000/50000 [==============================] - 32s 642us/step - loss: 0.3050 - acc: 0.8893 - val_loss: 0.2611 - val_acc: 0.9027\n",
      "Epoch 60/1000\n",
      "50000/50000 [==============================] - 24s 487us/step - loss: 0.3038 - acc: 0.8884 - val_loss: 0.2582 - val_acc: 0.9056\n",
      "Epoch 61/1000\n",
      "50000/50000 [==============================] - 24s 486us/step - loss: 0.3048 - acc: 0.8882 - val_loss: 0.2571 - val_acc: 0.9058\n",
      "Epoch 62/1000\n",
      "50000/50000 [==============================] - 21s 428us/step - loss: 0.3014 - acc: 0.8900 - val_loss: 0.2575 - val_acc: 0.9057\n",
      "Epoch 63/1000\n",
      "50000/50000 [==============================] - 23s 455us/step - loss: 0.2983 - acc: 0.8909 - val_loss: 0.2603 - val_acc: 0.9044\n",
      "Epoch 64/1000\n",
      "50000/50000 [==============================] - 26s 510us/step - loss: 0.2998 - acc: 0.8903 - val_loss: 0.2571 - val_acc: 0.9067\n",
      "Epoch 65/1000\n",
      "50000/50000 [==============================] - 32s 638us/step - loss: 0.2982 - acc: 0.8907 - val_loss: 0.2534 - val_acc: 0.9081\n",
      "Epoch 66/1000\n",
      "50000/50000 [==============================] - 25s 495us/step - loss: 0.2970 - acc: 0.8913 - val_loss: 0.2524 - val_acc: 0.9065\n",
      "Epoch 67/1000\n",
      "50000/50000 [==============================] - 23s 454us/step - loss: 0.2996 - acc: 0.8904 - val_loss: 0.2570 - val_acc: 0.9059\n",
      "Epoch 68/1000\n",
      "50000/50000 [==============================] - 24s 487us/step - loss: 0.2963 - acc: 0.8920 - val_loss: 0.2543 - val_acc: 0.9067\n",
      "Epoch 69/1000\n",
      "50000/50000 [==============================] - 25s 507us/step - loss: 0.2936 - acc: 0.8927 - val_loss: 0.2514 - val_acc: 0.9089\n",
      "Epoch 70/1000\n",
      "50000/50000 [==============================] - 29s 587us/step - loss: 0.2930 - acc: 0.8917 - val_loss: 0.2521 - val_acc: 0.9059\n",
      "Epoch 71/1000\n",
      "50000/50000 [==============================] - 26s 522us/step - loss: 0.2905 - acc: 0.8939 - val_loss: 0.2502 - val_acc: 0.9080\n",
      "Epoch 72/1000\n",
      "50000/50000 [==============================] - 35s 703us/step - loss: 0.2917 - acc: 0.8925 - val_loss: 0.2493 - val_acc: 0.9089\n",
      "Epoch 73/1000\n",
      "50000/50000 [==============================] - 32s 647us/step - loss: 0.2929 - acc: 0.8933 - val_loss: 0.2529 - val_acc: 0.9067\n",
      "Epoch 74/1000\n",
      "50000/50000 [==============================] - 32s 636us/step - loss: 0.2906 - acc: 0.8927 - val_loss: 0.2508 - val_acc: 0.9082\n",
      "Epoch 75/1000\n",
      "50000/50000 [==============================] - 32s 632us/step - loss: 0.2883 - acc: 0.8940 - val_loss: 0.2511 - val_acc: 0.9049\n",
      "Epoch 76/1000\n",
      "50000/50000 [==============================] - 30s 609us/step - loss: 0.2870 - acc: 0.8945 - val_loss: 0.2516 - val_acc: 0.9056\n",
      "Epoch 00076: early stopping\n"
     ]
    }
   ],
   "source": [
    "#lower acc! :( let's try to add another fc layer\n",
    "conv_model3 = Sequential()\n",
    "conv_model3.add(Convolution2D(20, 3, 3,\n",
    "                        border_mode='valid',\n",
    "                        input_shape=(28, 28,1)))\n",
    "conv_model3.add(Activation('relu'))\n",
    "conv_model3.add(MaxPooling2D( pool_size = (2,2)))\n",
    "conv_model3.add(Flatten())\n",
    "conv_model3.add(Dropout(0.5))\n",
    "conv_model3.add(Dense(100))\n",
    "conv_model3.add(Activation('relu'))\n",
    "conv_model3.add(Dropout(0.5))\n",
    "conv_model3.add(Dense(10))\n",
    "conv_model3.add(Activation('softmax'))\n",
    "\n",
    "conv_model3.compile(loss='categorical_crossentropy',\n",
    "              optimizer='sgd',\n",
    "              metrics=['accuracy'])\n",
    "\n",
    "history = conv_model3.fit(X_train, Y_train, batch_size=32, \n",
    "          nb_epoch=1000,verbose=1,\n",
    "          validation_data=(X_val, Y_val),\n",
    "          callbacks=[early_stop])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 109,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.legend.Legend at 0x7f7a9f2d4f60>"
      ]
     },
     "execution_count": 109,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYYAAAEZCAYAAACTsIJzAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nwxUyAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f7a9f65d080>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(history.history['loss'])\n",
    "plt.plot(history.history['val_loss'])\n",
    "plt.title('model accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'test'], loc='upper left')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 110,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "acc is 0.9056\n"
     ]
    }
   ],
   "source": [
    "Y_pred = conv_model3.predict(X_val)\n",
    "#how many correct predictions\n",
    "correct_num = np.sum(Y_pred.argmax(axis=1)==y_val)\n",
    "print ('acc is {}'.format(correct_num/float(len(y_val))))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 113,
   "metadata": {
    "collapsed": false,
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/ubuntu/anaconda2/envs/deep-learning/lib/python3.5/site-packages/ipykernel/__main__.py:14: UserWarning: Update your `Conv2D` call to the Keras 2 API: `Conv2D(20, (3, 3), padding=\"valid\", input_shape=(28, 28, 1...)`\n",
      "/home/ubuntu/anaconda2/envs/deep-learning/lib/python3.5/site-packages/ipykernel/__main__.py:32: UserWarning: The semantics of the Keras 2 argument `steps_per_epoch` is not the same as the Keras 1 argument `samples_per_epoch`. `steps_per_epoch` is the number of batches to draw from the generator at each epoch. Basically steps_per_epoch = samples_per_epoch/batch_size. Similarly `nb_val_samples`->`validation_steps` and `val_samples`->`steps` arguments have changed. Update your method calls accordingly.\n",
      "/home/ubuntu/anaconda2/envs/deep-learning/lib/python3.5/site-packages/ipykernel/__main__.py:32: UserWarning: Update your `fit_generator` call to the Keras 2 API: `fit_generator(<keras.pre..., epochs=1000, validation_data=(array([[[..., verbose=1, callbacks=[<keras.ca...)`\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/1000\n",
      "1563/1563 [==============================] - 26s 16ms/step - loss: 1.1510 - acc: 0.5737 - val_loss: 0.6620 - val_acc: 0.7529\n",
      "Epoch 2/1000\n",
      "1563/1563 [==============================] - 26s 17ms/step - loss: 0.7693 - acc: 0.7148 - val_loss: 0.5733 - val_acc: 0.7834\n",
      "Epoch 3/1000\n",
      "1563/1563 [==============================] - 22s 14ms/step - loss: 0.6798 - acc: 0.7493 - val_loss: 0.5128 - val_acc: 0.8062\n",
      "Epoch 4/1000\n",
      "1563/1563 [==============================] - 24s 15ms/step - loss: 0.6158 - acc: 0.7724 - val_loss: 0.4811 - val_acc: 0.8243\n",
      "Epoch 5/1000\n",
      "1563/1563 [==============================] - 20s 13ms/step - loss: 0.5793 - acc: 0.7908 - val_loss: 0.4544 - val_acc: 0.8339\n",
      "Epoch 6/1000\n",
      "1563/1563 [==============================] - 21s 14ms/step - loss: 0.5502 - acc: 0.8014 - val_loss: 0.4330 - val_acc: 0.8434\n",
      "Epoch 7/1000\n",
      "1563/1563 [==============================] - 21s 14ms/step - loss: 0.5275 - acc: 0.8121 - val_loss: 0.4176 - val_acc: 0.8467\n",
      "Epoch 8/1000\n",
      "1563/1563 [==============================] - 22s 14ms/step - loss: 0.5103 - acc: 0.8176 - val_loss: 0.4062 - val_acc: 0.8496\n",
      "Epoch 9/1000\n",
      "1563/1563 [==============================] - 22s 14ms/step - loss: 0.4935 - acc: 0.8238 - val_loss: 0.3910 - val_acc: 0.8587\n",
      "Epoch 10/1000\n",
      "1563/1563 [==============================] - 22s 14ms/step - loss: 0.4787 - acc: 0.8298 - val_loss: 0.3788 - val_acc: 0.8608\n",
      "Epoch 11/1000\n",
      "1563/1563 [==============================] - 24s 16ms/step - loss: 0.4686 - acc: 0.8327 - val_loss: 0.3728 - val_acc: 0.8658\n",
      "Epoch 12/1000\n",
      "1563/1563 [==============================] - 21s 13ms/step - loss: 0.4568 - acc: 0.8378 - val_loss: 0.3639 - val_acc: 0.8701\n",
      "Epoch 13/1000\n",
      "1563/1563 [==============================] - 20s 13ms/step - loss: 0.4495 - acc: 0.8387 - val_loss: 0.3564 - val_acc: 0.8695\n",
      "Epoch 14/1000\n",
      "1563/1563 [==============================] - 20s 13ms/step - loss: 0.4398 - acc: 0.8440 - val_loss: 0.3472 - val_acc: 0.8725\n",
      "Epoch 15/1000\n",
      "1563/1563 [==============================] - 20s 13ms/step - loss: 0.4342 - acc: 0.8443 - val_loss: 0.3464 - val_acc: 0.8730\n",
      "Epoch 16/1000\n",
      "1563/1563 [==============================] - 22s 14ms/step - loss: 0.4299 - acc: 0.8470 - val_loss: 0.3405 - val_acc: 0.8764\n",
      "Epoch 17/1000\n",
      "1563/1563 [==============================] - 22s 14ms/step - loss: 0.4249 - acc: 0.8492 - val_loss: 0.3376 - val_acc: 0.8741\n",
      "Epoch 18/1000\n",
      "1563/1563 [==============================] - 22s 14ms/step - loss: 0.4195 - acc: 0.8500 - val_loss: 0.3339 - val_acc: 0.8776\n",
      "Epoch 19/1000\n",
      "1563/1563 [==============================] - 22s 14ms/step - loss: 0.4166 - acc: 0.8510 - val_loss: 0.3323 - val_acc: 0.8784\n",
      "Epoch 20/1000\n",
      "1563/1563 [==============================] - 22s 14ms/step - loss: 0.4098 - acc: 0.8522 - val_loss: 0.3262 - val_acc: 0.8796\n",
      "Epoch 21/1000\n",
      "1563/1563 [==============================] - 23s 15ms/step - loss: 0.4057 - acc: 0.8542 - val_loss: 0.3254 - val_acc: 0.8787\n",
      "Epoch 22/1000\n",
      "1563/1563 [==============================] - 21s 13ms/step - loss: 0.4014 - acc: 0.8562 - val_loss: 0.3207 - val_acc: 0.8799\n",
      "Epoch 23/1000\n",
      "1563/1563 [==============================] - 21s 14ms/step - loss: 0.4001 - acc: 0.8585 - val_loss: 0.3199 - val_acc: 0.8812\n",
      "Epoch 24/1000\n",
      "1563/1563 [==============================] - 25s 16ms/step - loss: 0.3943 - acc: 0.8584 - val_loss: 0.3147 - val_acc: 0.8834\n",
      "Epoch 25/1000\n",
      "1563/1563 [==============================] - 47s 30ms/step - loss: 0.3922 - acc: 0.8592 - val_loss: 0.3147 - val_acc: 0.8835\n",
      "Epoch 26/1000\n",
      "1563/1563 [==============================] - 22s 14ms/step - loss: 0.3891 - acc: 0.8601 - val_loss: 0.3163 - val_acc: 0.8816\n",
      "Epoch 27/1000\n",
      "1563/1563 [==============================] - 22s 14ms/step - loss: 0.3878 - acc: 0.8612 - val_loss: 0.3121 - val_acc: 0.8833\n",
      "Epoch 28/1000\n",
      "1563/1563 [==============================] - 25s 16ms/step - loss: 0.3861 - acc: 0.8604 - val_loss: 0.3090 - val_acc: 0.8849\n",
      "Epoch 29/1000\n",
      "1563/1563 [==============================] - 54s 35ms/step - loss: 0.3824 - acc: 0.8640 - val_loss: 0.3068 - val_acc: 0.8873\n",
      "Epoch 30/1000\n",
      "1563/1563 [==============================] - 61s 39ms/step - loss: 0.3789 - acc: 0.8660 - val_loss: 0.3056 - val_acc: 0.8863\n",
      "Epoch 31/1000\n",
      "1563/1563 [==============================] - 63s 41ms/step - loss: 0.3751 - acc: 0.8662 - val_loss: 0.3036 - val_acc: 0.8872\n",
      "Epoch 32/1000\n",
      "1563/1563 [==============================] - 62s 39ms/step - loss: 0.3757 - acc: 0.8654 - val_loss: 0.3061 - val_acc: 0.8862\n",
      "Epoch 33/1000\n",
      "1563/1563 [==============================] - 61s 39ms/step - loss: 0.3707 - acc: 0.8670 - val_loss: 0.2986 - val_acc: 0.8888\n",
      "Epoch 34/1000\n",
      "1563/1563 [==============================] - 60s 38ms/step - loss: 0.3674 - acc: 0.8674 - val_loss: 0.2967 - val_acc: 0.8912\n",
      "Epoch 35/1000\n",
      "1563/1563 [==============================] - 59s 38ms/step - loss: 0.3641 - acc: 0.8691 - val_loss: 0.3010 - val_acc: 0.8856\n",
      "Epoch 36/1000\n",
      "1563/1563 [==============================] - 62s 39ms/step - loss: 0.3677 - acc: 0.8693 - val_loss: 0.2962 - val_acc: 0.8902\n",
      "Epoch 37/1000\n",
      "1563/1563 [==============================] - 60s 38ms/step - loss: 0.3666 - acc: 0.8690 - val_loss: 0.2960 - val_acc: 0.8918\n",
      "Epoch 38/1000\n",
      "1563/1563 [==============================] - 62s 39ms/step - loss: 0.3634 - acc: 0.8679 - val_loss: 0.2911 - val_acc: 0.8914\n",
      "Epoch 39/1000\n",
      "1563/1563 [==============================] - 62s 40ms/step - loss: 0.3575 - acc: 0.8723 - val_loss: 0.2885 - val_acc: 0.8928\n",
      "Epoch 40/1000\n",
      "1563/1563 [==============================] - 66s 42ms/step - loss: 0.3564 - acc: 0.8701 - val_loss: 0.2911 - val_acc: 0.8933\n",
      "Epoch 41/1000\n",
      "1563/1563 [==============================] - 43s 27ms/step - loss: 0.3572 - acc: 0.8732 - val_loss: 0.2869 - val_acc: 0.8943\n",
      "Epoch 42/1000\n",
      "1563/1563 [==============================] - 28s 18ms/step - loss: 0.3538 - acc: 0.8718 - val_loss: 0.2824 - val_acc: 0.8960\n",
      "Epoch 43/1000\n",
      "1563/1563 [==============================] - 21s 14ms/step - loss: 0.3543 - acc: 0.8725 - val_loss: 0.2830 - val_acc: 0.8938\n",
      "Epoch 44/1000\n",
      "1563/1563 [==============================] - 21s 13ms/step - loss: 0.3527 - acc: 0.8740 - val_loss: 0.2838 - val_acc: 0.8940\n",
      "Epoch 45/1000\n",
      "1563/1563 [==============================] - 21s 13ms/step - loss: 0.3495 - acc: 0.8739 - val_loss: 0.2830 - val_acc: 0.8967\n",
      "Epoch 46/1000\n",
      "1563/1563 [==============================] - 21s 13ms/step - loss: 0.3485 - acc: 0.8754 - val_loss: 0.2817 - val_acc: 0.8949\n",
      "Epoch 47/1000\n",
      "1563/1563 [==============================] - 24s 16ms/step - loss: 0.3471 - acc: 0.8741 - val_loss: 0.2818 - val_acc: 0.8952\n",
      "Epoch 48/1000\n",
      "1563/1563 [==============================] - 22s 14ms/step - loss: 0.3455 - acc: 0.8748 - val_loss: 0.2772 - val_acc: 0.8976\n",
      "Epoch 49/1000\n",
      "1563/1563 [==============================] - 23s 15ms/step - loss: 0.3439 - acc: 0.8747 - val_loss: 0.2846 - val_acc: 0.8954\n",
      "Epoch 50/1000\n",
      "1563/1563 [==============================] - 32s 20ms/step - loss: 0.3394 - acc: 0.8769 - val_loss: 0.2769 - val_acc: 0.8969\n",
      "Epoch 51/1000\n",
      "1563/1563 [==============================] - 32s 21ms/step - loss: 0.3382 - acc: 0.8779 - val_loss: 0.2746 - val_acc: 0.8998\n",
      "Epoch 52/1000\n",
      "1563/1563 [==============================] - 33s 21ms/step - loss: 0.3393 - acc: 0.8769 - val_loss: 0.2789 - val_acc: 0.8970\n",
      "Epoch 53/1000\n",
      "1563/1563 [==============================] - 33s 21ms/step - loss: 0.3369 - acc: 0.8786 - val_loss: 0.2777 - val_acc: 0.8956\n",
      "Epoch 54/1000\n",
      "1563/1563 [==============================] - 32s 20ms/step - loss: 0.3388 - acc: 0.8783 - val_loss: 0.2743 - val_acc: 0.9004\n",
      "Epoch 55/1000\n",
      "1563/1563 [==============================] - 20s 13ms/step - loss: 0.3370 - acc: 0.8788 - val_loss: 0.2717 - val_acc: 0.9018\n",
      "Epoch 56/1000\n",
      "1563/1563 [==============================] - 20s 13ms/step - loss: 0.3353 - acc: 0.8785 - val_loss: 0.2726 - val_acc: 0.8999\n",
      "Epoch 57/1000\n",
      "1563/1563 [==============================] - 22s 14ms/step - loss: 0.3323 - acc: 0.8791 - val_loss: 0.2709 - val_acc: 0.8989\n",
      "Epoch 58/1000\n",
      "1563/1563 [==============================] - 32s 20ms/step - loss: 0.3350 - acc: 0.8797 - val_loss: 0.2700 - val_acc: 0.8977\n",
      "Epoch 59/1000\n",
      "1563/1563 [==============================] - 32s 20ms/step - loss: 0.3323 - acc: 0.8783 - val_loss: 0.2711 - val_acc: 0.9015\n",
      "Epoch 60/1000\n",
      "1563/1563 [==============================] - 32s 21ms/step - loss: 0.3319 - acc: 0.8809 - val_loss: 0.2670 - val_acc: 0.9020\n",
      "Epoch 61/1000\n",
      "1563/1563 [==============================] - 33s 21ms/step - loss: 0.3287 - acc: 0.8819 - val_loss: 0.2674 - val_acc: 0.9020\n",
      "Epoch 62/1000\n",
      "1563/1563 [==============================] - 32s 21ms/step - loss: 0.3279 - acc: 0.8818 - val_loss: 0.2703 - val_acc: 0.9026\n",
      "Epoch 63/1000\n",
      "1563/1563 [==============================] - 33s 21ms/step - loss: 0.3271 - acc: 0.8818 - val_loss: 0.2682 - val_acc: 0.9027\n",
      "Epoch 64/1000\n",
      "1563/1563 [==============================] - 33s 21ms/step - loss: 0.3232 - acc: 0.8832 - val_loss: 0.2681 - val_acc: 0.9004\n",
      "Epoch 00064: early stopping\n"
     ]
    }
   ],
   "source": [
    "#Bamm! 90% acc! #maybe we can squeeze a bit more using data augmentation\n",
    "\n",
    "from keras.preprocessing.image import ImageDataGenerator\n",
    "\n",
    "datagen = ImageDataGenerator(\n",
    "    #width_shift_range=0.1,  # randomly shift images horizontally (fraction of total width)\n",
    "    #height_shift_range=0.1,  # randomly shift images vertically (fraction of total height)\n",
    "    horizontal_flip=True,  # randomly flip images\n",
    "    vertical_flip=False)  # randomly flip images\n",
    "\n",
    "conv_model3 = Sequential()\n",
    "conv_model3.add(Convolution2D(20, 3, 3,\n",
    "                        border_mode='valid',\n",
    "                        input_shape=(28, 28,1)))\n",
    "conv_model3.add(Activation('relu'))\n",
    "conv_model3.add(MaxPooling2D( pool_size = (2,2)))\n",
    "conv_model3.add(Flatten())\n",
    "conv_model3.add(Dropout(0.5))\n",
    "conv_model3.add(Dense(100))\n",
    "conv_model3.add(Activation('relu'))\n",
    "conv_model3.add(Dropout(0.5))\n",
    "conv_model3.add(Dense(10))\n",
    "conv_model3.add(Activation('softmax'))\n",
    "\n",
    "conv_model3.compile(loss='categorical_crossentropy',\n",
    "              optimizer='sgd',\n",
    "              metrics=['accuracy'])\n",
    "\n",
    "history = conv_model3.fit_generator(datagen.flow(X_train, Y_train, batch_size=32), \n",
    "          nb_epoch=1000,verbose=1,\n",
    "          validation_data=(X_val, Y_val),\n",
    "          callbacks=[early_stop])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 114,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "acc is 0.9004\n"
     ]
    }
   ],
   "source": [
    "Y_pred = conv_model3.predict(X_val)\n",
    "#how many correct predictions\n",
    "correct_num = np.sum(Y_pred.argmax(axis=1)==y_val)\n",
    "print ('acc is {}'.format(correct_num/float(len(y_val))))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 115,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# the acc is lower, strange, never mind"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Conv3 gave the best acc on val, so let's report its performance on test "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 131,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "acc is 0.897\n"
     ]
    }
   ],
   "source": [
    "X_test = X_test.reshape(X_test.shape[0],X_test.shape[1],X_test.shape[2],1)\n",
    "\n",
    "Y_pred = conv_model3.predict(X_test)\n",
    "#how many correct predictions\n",
    "correct_num = np.sum(Y_pred.argmax(axis=1)==y_test)\n",
    "print ('acc is {}'.format(correct_num/float(len(y_test))))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "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": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}