2- Anomaly detection.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Detección de anomalías\n",
    "\n",
    "Video:\n",
    "\n",
    "https://www.youtube.com/watch?v=086OcT-5DYI\n",
    "\n",
    "- Aprendizaje no-supervisado\n",
    "- Detección de valores con alto desvío con respecto a la distribución de probabilidad nominal\n",
    "\n",
    "<img src=\"Anomaly_detection.png\">\n",
    "\n",
    "Una opción es utilizar cualquier modelo probabilístico generativo de los datos. Si la probabilidad de que la muestra haya sido generada por el modelo generativo es menor que un valor de umbral, entonces diremos que el dato es anómalo.\n",
    "\n",
    "Ejemplo: Datos generados con una mezcla de gausianas vibariable."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from matplotlib.colors import LogNorm\n",
    "from sklearn import mixture\n",
    "\n",
    "n_samples = 300\n",
    "\n",
    "# generate random sample, two components\n",
    "np.random.seed(0)\n",
    "\n",
    "# generate spherical data centered on (20, 20)\n",
    "shifted_gaussian = np.random.randn(n_samples, 2) + np.array([20, 20])\n",
    "\n",
    "# generate zero centered stretched Gaussian data\n",
    "C = np.array([[0., -0.7], [3.5, .7]])\n",
    "stretched_gaussian = np.dot(np.random.randn(n_samples, 2), C)\n",
    "\n",
    "# concatenate the two datasets into the final training set\n",
    "X_train = np.vstack([shifted_gaussian, stretched_gaussian])\n",
    "\n",
    "# fit a Gaussian Mixture Model with two components\n",
    "clf = mixture.GaussianMixture(n_components=5, covariance_type='full')\n",
    "clf.fit(X_train)\n",
    "\n",
    "# display predicted scores by the model as a contour plot\n",
    "x = np.linspace(-20., 30.)\n",
    "y = np.linspace(-20., 40.)\n",
    "X, Y = np.meshgrid(x, y)\n",
    "XX = np.array([X.ravel(), Y.ravel()]).T\n",
    "Z = -clf.score_samples(XX)\n",
    "Z = Z.reshape(X.shape)\n",
    "CS = plt.contour(X, Y, Z, norm=LogNorm(vmin=1.0, vmax=1000.0),\n",
    "                 levels=np.logspace(0, 3, 10))\n",
    "CB = plt.colorbar(CS, shrink=0.8, extend='both')\n",
    "plt.scatter(X_train[:, 0], X_train[:, 1], .8)\n",
    "\n",
    "plt.title('Negative log-likelihood predicted by a GMM')\n",
    "plt.axis('tight')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Las curvas de nivel definen distintos valores de probabilidad para los cuales tomar el umbral de decisión.  \n",
    "¿Qué pasa si varío la cantidad de componentes de GMM?\n",
    "\n",
    "veamos el mismo problema para N=300."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "GaussianMixture(covariance_type='full', init_params='kmeans', max_iter=100,\n",
       "        means_init=None, n_components=5, n_init=1, precisions_init=None,\n",
       "        random_state=None, reg_covar=1e-06, tol=0.001, verbose=0,\n",
       "        verbose_interval=10, warm_start=False, weights_init=None)"
      ]
     },
     "execution_count": 101,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "n_samples = 5000\n",
    "\n",
    "# generate random sample, two components\n",
    "np.random.seed(0)\n",
    "\n",
    "# generate spherical data centered on (20, 20,20,...,20)\n",
    "shifted_gaussian = 20*np.random.randn(n_samples, 30) + np.array([20]*30)\n",
    "\n",
    "# generate zero centered stretched Gaussian data\n",
    "C = np.array([np.random.randn(30,30)])\n",
    "stretched_gaussian = 20*np.dot(np.random.randn(n_samples, 30), C).reshape(5000,30)\n",
    "\n",
    "# concatenate the two datasets into the final training set\n",
    "X_train = np.vstack([shifted_gaussian, stretched_gaussian])\n",
    "\n",
    "# fit a Gaussian Mixture Model with two components\n",
    "clf = mixture.GaussianMixture(n_components=5, covariance_type='full')\n",
    "clf.fit(X_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {},
   "outputs": [],
   "source": [
    "a=-clf.score_samples(X_train)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 103,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([ 688., 3026., 1169.,  107.,    9.,   15., 1475., 2841.,  631.,\n",
       "          39.]), array([122.52720993, 128.88859996, 135.24999   , 141.61138003,\n",
       "        147.97277007, 154.3341601 , 160.69555014, 167.05694018,\n",
       "        173.41833021, 179.77972025, 186.14111028]), <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 103,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(a)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(10000,)"
      ]
     },
     "execution_count": 100,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "a.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 21.76405235,  20.40015721,  20.97873798, ...,  20.58295368,\n",
       "         19.60055097,  20.37005589],\n",
       "       [ 18.69347315,  21.65813068,  19.88183595, ...,  19.78326853,\n",
       "         19.0698435 ,  19.82141091],\n",
       "       [ 18.44957065,  20.41731882,  19.05563151, ...,  18.4584126 ,\n",
       "         22.59442459,  19.59596771],\n",
       "       ...,\n",
       "       [  5.65274769,   8.96959273, -35.86870107, ...,  48.46996094,\n",
       "          2.2198951 , -16.52006971],\n",
       "       [-15.12667051,  11.37807578,   8.26263638, ...,   1.19539895,\n",
       "        -15.251113  ,   4.95605932],\n",
       "       [ -3.3255099 ,  -9.74448279,  -1.36314746, ...,  17.28478679,\n",
       "        -14.02059223, -23.16757352]])"
      ]
     },
     "execution_count": 66,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Detección de anomalías con autoencoders"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import keras\n",
    "import keras.backend as K\n",
    "from keras.optimizers import Adam\n",
    "from keras.callbacks import EarlyStopping, ModelCheckpoint, ReduceLROnPlateau\n",
    "from keras.layers import Input,Dense\n",
    "from keras.models import Model\n",
    "\n",
    "early_stopping = EarlyStopping(patience=10, verbose=1)\n",
    "model_checkpoint = ModelCheckpoint(\"./keras.model\", save_best_only=True, verbose=1)\n",
    "reduce_lr = ReduceLROnPlateau(factor=0.1, patience=5, min_lr=0.00001, verbose=1)\n",
    "crunch_dim=4\n",
    "\n",
    "layer0 = Input(shape=(784,))\n",
    "crunch = Dense(crunch_dim, activation='relu')(layer0)\n",
    "layer2 = Dense(784, activation='sigmoid')(crunch)\n",
    "\n",
    "autoencoder = Model(layer0, layer2)\n",
    "encoder = Model(layer0, crunch)\n",
    "\n",
    "decrunch = Input(shape=(crunch_dim,))\n",
    "decoded = autoencoder.layers[-1](decrunch)\n",
    "decoder = Model(decrunch, decoded)\n",
    "\n",
    "autoencoder.compile(optimizer=Adam(lr=0.01), loss='binary_crossentropy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'autoencoder' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-3-6b6da48a72ba>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m     25\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     26\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 27\u001b[0;31m \u001b[0mautoencoder\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfit\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mX2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mX2\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mvalidation_data\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mX2_t\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mX2_t\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mbatch_size\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m128\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mepochs\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mepochs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mverbose\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mcallbacks\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mearly_stopping\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmodel_checkpoint\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mreduce_lr\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mshuffle\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mTrue\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     28\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mNameError\u001b[0m: name 'autoencoder' is not defined"
     ]
    }
   ],
   "source": [
    "from keras.datasets import mnist\n",
    "#(train), (test)\n",
    "(Xa, ya), (Xt, yt) = mnist.load_data()\n",
    "mask=(ya==8)\n",
    "y2=ya[mask]\n",
    "X2=Xa[mask,:,:]\n",
    "mask=(ya==7)\n",
    "y1=ya[mask]\n",
    "X1=Xa[mask,:,:]\n",
    "mask=(yt==8)\n",
    "y2_t=yt[mask]\n",
    "X2_t=Xt[mask,:,:]\n",
    "mask=(yt==7)\n",
    "y1_t=yt[mask]\n",
    "X1_t=Xt[mask,:,:]\n",
    "\n",
    "#print(Xa.shape, Ya.shape)\n",
    "isize = 28 #image size (square)\n",
    "n_classes = 10\n",
    "epochs = 500\n",
    "X1 = (X1.astype('float32')/255.0).reshape((X1.shape[0], np.prod(X1.shape[1:])))\n",
    "X2 = (X2.astype('float32')/255.0).reshape((X2.shape[0], np.prod(X2.shape[1:])))\n",
    "X1_t = (X1_t.astype('float32')/255.0).reshape((X1_t.shape[0], np.prod(X1_t.shape[1:])))\n",
    "X2_t = (X2_t.astype('float32')/255.0).reshape((X2_t.shape[0], np.prod(X2_t.shape[1:])))\n",
    "\n",
    "\n",
    "autoencoder.fit(X2, X2, validation_data=[X2_t, X2_t], batch_size=128, epochs=epochs, verbose=1,callbacks=[early_stopping, model_checkpoint, reduce_lr], shuffle=True)\n",
    "                    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "out1=[autoencoder.evaluate(X1[idx:idx+1,:],X1[idx:idx+1,:],verbose=0) for idx in range(int(X1.shape[0]))]\n",
    "out2=[autoencoder.evaluate(X2[idx:idx+1,:],X2[idx:idx+1,:],verbose=0) for idx in range(int(X2.shape[0]))]\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([8.870e+02, 2.923e+03, 1.221e+03, 5.170e+02, 1.920e+02, 7.500e+01,\n",
       "        2.600e+01, 7.000e+00, 2.000e+00, 1.000e+00]),\n",
       " array([0.12937903, 0.17535574, 0.22133244, 0.26730915, 0.31328585,\n",
       "        0.35926256, 0.40523926, 0.45121596, 0.49719267, 0.54316937,\n",
       "        0.58914608]),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from matplotlib import pyplot as plt\n",
    "plt.hist(out1)\n",
    "plt.hist(out2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x288 with 20 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x288 with 20 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "decoded_imgs = autoencoder.predict(X2)\n",
    "\n",
    "def bare(ax):\n",
    "    ax.get_xaxis().set_visible(False)\n",
    "    ax.get_yaxis().set_visible(False)\n",
    "    return ax\n",
    "\n",
    "n = 10\n",
    "plt.figure(figsize=(20,4))\n",
    "for i in range(n):\n",
    "    ax = plt.subplot(2, n, i+1)\n",
    "    plt.imshow(X2[i].reshape(28,28))\n",
    "    plt.gray()\n",
    "    bare(ax)\n",
    "    ax = plt.subplot(2, n, i+1+n)\n",
    "    plt.imshow(decoded_imgs[i].reshape(28,28))\n",
    "    plt.hot()\n",
    "    bare(ax)\n",
    "plt.show()\n",
    "decoded_imgs = autoencoder.predict(X1)\n",
    "\n",
    "def bare(ax):\n",
    "    ax.get_xaxis().set_visible(False)\n",
    "    ax.get_yaxis().set_visible(False)\n",
    "    return ax\n",
    "\n",
    "n = 10\n",
    "plt.figure(figsize=(20,4))\n",
    "for i in range(n):\n",
    "    ax = plt.subplot(2, n, i+1)\n",
    "    plt.imshow(X1[i].reshape(28,28))\n",
    "    plt.gray()\n",
    "    bare(ax)\n",
    "    ax = plt.subplot(2, n, i+1+n)\n",
    "    plt.imshow(decoded_imgs[i].reshape(28,28))\n",
    "    plt.hot()\n",
    "    bare(ax)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "out1_t=[autoencoder.evaluate(X1_t[idx:idx+1,:],X1_t[idx:idx+1,:],verbose=0) for idx in range(int(X1_t.shape[0]))]\n",
    "out2_t=[autoencoder.evaluate(X2_t[idx:idx+1,:],X2_t[idx:idx+1,:],verbose=0) for idx in range(int(X2_t.shape[0]))]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([178., 339., 221., 126.,  61.,  30.,  22.,  34.,  16.,   1.]),\n",
       " array([0.18609485, 0.22982236, 0.27354987, 0.31727737, 0.36100488,\n",
       "        0.40473239, 0.4484599 , 0.49218741, 0.53591492, 0.57964242,\n",
       "        0.62336993]),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(out2_t)\n",
    "plt.hist(out1_t)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x288 with 20 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "decoded_imgs = autoencoder.predict(X1)\n",
    "\n",
    "def bare(ax):\n",
    "    ax.get_xaxis().set_visible(False)\n",
    "    ax.get_yaxis().set_visible(False)\n",
    "    return ax\n",
    "\n",
    "n = 10\n",
    "plt.figure(figsize=(20,4))\n",
    "for i in range(n):\n",
    "    ax = plt.subplot(2, n, i+1)\n",
    "    plt.imshow(X1[i].reshape(28,28))\n",
    "    plt.gray()\n",
    "    bare(ax)\n",
    "    ax = plt.subplot(2, n, i+1+n)\n",
    "    plt.imshow(decoded_imgs[i].reshape(28,28))\n",
    "    plt.hot()\n",
    "    bare(ax)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import keras\n",
    "import keras.backend as K\n",
    "from keras.optimizers import Adam\n",
    "from keras.callbacks import EarlyStopping, ModelCheckpoint, ReduceLROnPlateau\n",
    "from keras.layers import Input, Dense, Conv2D, MaxPooling2D, UpSampling2D\n",
    "from keras.models import Model\n",
    "early_stopping2 = EarlyStopping(patience=10, verbose=1)\n",
    "model_checkpoint2 = ModelCheckpoint(\"./keras.model_an_cnn\", save_best_only=True, verbose=1)\n",
    "reduce_lr2 = ReduceLROnPlateau(factor=0.1, patience=5, min_lr=0.00001, verbose=1)\n",
    "input_img = Input(shape=(28, 28, 1))  # adapt this if using `channels_first` image data format\n",
    "\n",
    "x = Conv2D(16, (3, 3), activation='relu', padding='same')(input_img)\n",
    "x = MaxPooling2D((2, 2), padding='same')(x)\n",
    "x = Conv2D(8, (4, 4), activation='relu', padding='same')(x)\n",
    "x = MaxPooling2D((2, 2), padding='same')(x)\n",
    "x = Conv2D(8, (5, 5), activation='relu', padding='same')(x)\n",
    "x = MaxPooling2D((2, 2), padding='same')(x)\n",
    "x = Conv2D(4, (5, 5), activation='relu', padding='same')(x)\n",
    "x = MaxPooling2D((2, 2), padding='same')(x)\n",
    "encoded = Conv2D(2, (5, 5), activation='relu', padding='same')(x)\n",
    "\n",
    "\n",
    "# at this point the representation is (4, 4, 8) i.e. 128-dimensional\n",
    "x = UpSampling2D((2, 2))(encoded)\n",
    "x = Conv2D(2, (5, 5), activation='relu',padding='same')(x)\n",
    "x = UpSampling2D((2, 2))(x)\n",
    "x = Conv2D(4, (5, 5), activation='relu',padding='same')(x)\n",
    "x = UpSampling2D((2, 2))(x)\n",
    "x = Conv2D(8, (4, 4), activation='relu',padding='same')(x)\n",
    "x = UpSampling2D((2, 2))(x)\n",
    "x = Conv2D(16, (3, 3), activation='relu')(x) #ojo con el padding! acá está la magia\n",
    "decoded = Conv2D(1, (3, 3), activation='sigmoid')(x)\n",
    "\n",
    "autoencoder2 = Model(input_img, decoded)\n",
    "autoencoder2.compile(optimizer='Adam', loss='binary_crossentropy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_2 (InputLayer)         (None, 28, 28, 1)         0         \n",
      "_________________________________________________________________\n",
      "conv2d_14 (Conv2D)           (None, 28, 28, 16)        160       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_3 (MaxPooling2 (None, 14, 14, 16)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_15 (Conv2D)           (None, 14, 14, 8)         2056      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_4 (MaxPooling2 (None, 7, 7, 8)           0         \n",
      "_________________________________________________________________\n",
      "conv2d_16 (Conv2D)           (None, 7, 7, 8)           1608      \n",
      "_________________________________________________________________\n",
      "max_pooling2d_5 (MaxPooling2 (None, 4, 4, 8)           0         \n",
      "_________________________________________________________________\n",
      "conv2d_17 (Conv2D)           (None, 4, 4, 4)           804       \n",
      "_________________________________________________________________\n",
      "max_pooling2d_6 (MaxPooling2 (None, 2, 2, 4)           0         \n",
      "_________________________________________________________________\n",
      "conv2d_18 (Conv2D)           (None, 2, 2, 2)           202       \n",
      "_________________________________________________________________\n",
      "up_sampling2d_3 (UpSampling2 (None, 4, 4, 2)           0         \n",
      "_________________________________________________________________\n",
      "conv2d_19 (Conv2D)           (None, 4, 4, 2)           102       \n",
      "_________________________________________________________________\n",
      "up_sampling2d_4 (UpSampling2 (None, 8, 8, 2)           0         \n",
      "_________________________________________________________________\n",
      "conv2d_20 (Conv2D)           (None, 8, 8, 4)           204       \n",
      "_________________________________________________________________\n",
      "up_sampling2d_5 (UpSampling2 (None, 16, 16, 4)         0         \n",
      "_________________________________________________________________\n",
      "conv2d_21 (Conv2D)           (None, 16, 16, 8)         520       \n",
      "_________________________________________________________________\n",
      "up_sampling2d_6 (UpSampling2 (None, 32, 32, 8)         0         \n",
      "_________________________________________________________________\n",
      "conv2d_22 (Conv2D)           (None, 30, 30, 16)        1168      \n",
      "_________________________________________________________________\n",
      "conv2d_23 (Conv2D)           (None, 28, 28, 1)         145       \n",
      "=================================================================\n",
      "Total params: 6,969\n",
      "Trainable params: 6,969\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "autoencoder2.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from keras.datasets import mnist\n",
    "(Xa, ya), (Xt, yt) = mnist.load_data()\n",
    "Xa = np.reshape(Xa, (len(Xa), 28, 28, 1))  # adapt this if using `channels_first` image data format\n",
    "Xt = np.reshape(Xt, (len(Xt), 28, 28, 1))  # adapt this if using `channels_first` image data format\n",
    "mask=(ya==8)\n",
    "y2=ya[mask]\n",
    "X2=Xa[mask,:,:,:]\n",
    "mask=(ya==7)\n",
    "y1=ya[mask]\n",
    "X1=Xa[mask,:,:,:]\n",
    "mask=(yt==8)\n",
    "y2_t=yt[mask]\n",
    "X2_t=Xt[mask,:,:,:]\n",
    "mask=(yt==7)\n",
    "y1_t=yt[mask]\n",
    "X1_t=Xt[mask,:,:,:]\n",
    "X1 = (X1.astype('float32')/255.0)\n",
    "X2 = (X2.astype('float32')/255.0)\n",
    "X1_t = (X1_t.astype('float32')/255.0)\n",
    "X2_t = (X2_t.astype('float32')/255.0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 5851 samples, validate on 974 samples\n",
      "Epoch 1/500\n",
      "5851/5851 [==============================] - 1s 160us/step - loss: 0.5681 - val_loss: 0.3774\n",
      "\n",
      "Epoch 00001: val_loss improved from inf to 0.37741, saving model to ./keras.model_an_cnn\n",
      "Epoch 2/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.2984 - val_loss: 0.2642\n",
      "\n",
      "Epoch 00002: val_loss improved from 0.37741 to 0.26422, saving model to ./keras.model_an_cnn\n",
      "Epoch 3/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.2563 - val_loss: 0.2486\n",
      "\n",
      "Epoch 00003: val_loss improved from 0.26422 to 0.24858, saving model to ./keras.model_an_cnn\n",
      "Epoch 4/500\n",
      "5851/5851 [==============================] - 0s 46us/step - loss: 0.2419 - val_loss: 0.2314\n",
      "\n",
      "Epoch 00004: val_loss improved from 0.24858 to 0.23144, saving model to ./keras.model_an_cnn\n",
      "Epoch 5/500\n",
      "5851/5851 [==============================] - 0s 39us/step - loss: 0.2228 - val_loss: 0.2185\n",
      "\n",
      "Epoch 00005: val_loss improved from 0.23144 to 0.21851, saving model to ./keras.model_an_cnn\n",
      "Epoch 6/500\n",
      "5851/5851 [==============================] - 0s 48us/step - loss: 0.2141 - val_loss: 0.2118\n",
      "\n",
      "Epoch 00006: val_loss improved from 0.21851 to 0.21177, saving model to ./keras.model_an_cnn\n",
      "Epoch 7/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.2080 - val_loss: 0.2064\n",
      "\n",
      "Epoch 00007: val_loss improved from 0.21177 to 0.20642, saving model to ./keras.model_an_cnn\n",
      "Epoch 8/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.2037 - val_loss: 0.2027\n",
      "\n",
      "Epoch 00008: val_loss improved from 0.20642 to 0.20273, saving model to ./keras.model_an_cnn\n",
      "Epoch 9/500\n",
      "5851/5851 [==============================] - 0s 40us/step - loss: 0.2006 - val_loss: 0.2007\n",
      "\n",
      "Epoch 00009: val_loss improved from 0.20273 to 0.20066, saving model to ./keras.model_an_cnn\n",
      "Epoch 10/500\n",
      "5851/5851 [==============================] - 0s 55us/step - loss: 0.1979 - val_loss: 0.1972\n",
      "\n",
      "Epoch 00010: val_loss improved from 0.20066 to 0.19717, saving model to ./keras.model_an_cnn\n",
      "Epoch 11/500\n",
      "5851/5851 [==============================] - 0s 47us/step - loss: 0.1956 - val_loss: 0.1949\n",
      "\n",
      "Epoch 00011: val_loss improved from 0.19717 to 0.19491, saving model to ./keras.model_an_cnn\n",
      "Epoch 12/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1935 - val_loss: 0.1941\n",
      "\n",
      "Epoch 00012: val_loss improved from 0.19491 to 0.19407, saving model to ./keras.model_an_cnn\n",
      "Epoch 13/500\n",
      "5851/5851 [==============================] - 0s 64us/step - loss: 0.1921 - val_loss: 0.1929\n",
      "\n",
      "Epoch 00013: val_loss improved from 0.19407 to 0.19293, saving model to ./keras.model_an_cnn\n",
      "Epoch 14/500\n",
      "5851/5851 [==============================] - 0s 40us/step - loss: 0.1903 - val_loss: 0.1907\n",
      "\n",
      "Epoch 00014: val_loss improved from 0.19293 to 0.19074, saving model to ./keras.model_an_cnn\n",
      "Epoch 15/500\n",
      "5851/5851 [==============================] - 0s 55us/step - loss: 0.1886 - val_loss: 0.1883\n",
      "\n",
      "Epoch 00015: val_loss improved from 0.19074 to 0.18830, saving model to ./keras.model_an_cnn\n",
      "Epoch 16/500\n",
      "5851/5851 [==============================] - 0s 42us/step - loss: 0.1873 - val_loss: 0.1881\n",
      "\n",
      "Epoch 00016: val_loss improved from 0.18830 to 0.18812, saving model to ./keras.model_an_cnn\n",
      "Epoch 17/500\n",
      "5851/5851 [==============================] - 0s 36us/step - loss: 0.1858 - val_loss: 0.1857\n",
      "\n",
      "Epoch 00017: val_loss improved from 0.18812 to 0.18572, saving model to ./keras.model_an_cnn\n",
      "Epoch 18/500\n",
      "5851/5851 [==============================] - 0s 39us/step - loss: 0.1850 - val_loss: 0.1851\n",
      "\n",
      "Epoch 00018: val_loss improved from 0.18572 to 0.18506, saving model to ./keras.model_an_cnn\n",
      "Epoch 19/500\n",
      "5851/5851 [==============================] - 0s 64us/step - loss: 0.1836 - val_loss: 0.1834\n",
      "\n",
      "Epoch 00019: val_loss improved from 0.18506 to 0.18344, saving model to ./keras.model_an_cnn\n",
      "Epoch 20/500\n",
      "5851/5851 [==============================] - 0s 71us/step - loss: 0.1824 - val_loss: 0.1828\n",
      "\n",
      "Epoch 00020: val_loss improved from 0.18344 to 0.18277, saving model to ./keras.model_an_cnn\n",
      "Epoch 21/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1816 - val_loss: 0.1829\n",
      "\n",
      "Epoch 00021: val_loss did not improve from 0.18277\n",
      "Epoch 22/500\n",
      "5851/5851 [==============================] - 0s 66us/step - loss: 0.1804 - val_loss: 0.1809\n",
      "\n",
      "Epoch 00022: val_loss improved from 0.18277 to 0.18090, saving model to ./keras.model_an_cnn\n",
      "Epoch 23/500\n",
      "5851/5851 [==============================] - 0s 64us/step - loss: 0.1796 - val_loss: 0.1802\n",
      "\n",
      "Epoch 00023: val_loss improved from 0.18090 to 0.18019, saving model to ./keras.model_an_cnn\n",
      "Epoch 24/500\n",
      "5851/5851 [==============================] - 0s 43us/step - loss: 0.1789 - val_loss: 0.1791\n",
      "\n",
      "Epoch 00024: val_loss improved from 0.18019 to 0.17913, saving model to ./keras.model_an_cnn\n",
      "Epoch 25/500\n",
      "5851/5851 [==============================] - 0s 36us/step - loss: 0.1782 - val_loss: 0.1787\n",
      "\n",
      "Epoch 00025: val_loss improved from 0.17913 to 0.17875, saving model to ./keras.model_an_cnn\n",
      "Epoch 26/500\n",
      "5851/5851 [==============================] - 0s 40us/step - loss: 0.1773 - val_loss: 0.1778\n",
      "\n",
      "Epoch 00026: val_loss improved from 0.17875 to 0.17777, saving model to ./keras.model_an_cnn\n",
      "Epoch 27/500\n",
      "5851/5851 [==============================] - 0s 58us/step - loss: 0.1766 - val_loss: 0.1775\n",
      "\n",
      "Epoch 00027: val_loss improved from 0.17777 to 0.17750, saving model to ./keras.model_an_cnn\n",
      "Epoch 28/500\n",
      "5851/5851 [==============================] - 0s 39us/step - loss: 0.1758 - val_loss: 0.1761\n",
      "\n",
      "Epoch 00028: val_loss improved from 0.17750 to 0.17609, saving model to ./keras.model_an_cnn\n",
      "Epoch 29/500\n",
      "5851/5851 [==============================] - 0s 56us/step - loss: 0.1755 - val_loss: 0.1759\n",
      "\n",
      "Epoch 00029: val_loss improved from 0.17609 to 0.17592, saving model to ./keras.model_an_cnn\n",
      "Epoch 30/500\n",
      "5851/5851 [==============================] - 0s 74us/step - loss: 0.1747 - val_loss: 0.1760\n",
      "\n",
      "Epoch 00030: val_loss did not improve from 0.17592\n",
      "Epoch 31/500\n",
      "5851/5851 [==============================] - 0s 71us/step - loss: 0.1740 - val_loss: 0.1749\n",
      "\n",
      "Epoch 00031: val_loss improved from 0.17592 to 0.17490, saving model to ./keras.model_an_cnn\n",
      "Epoch 32/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1729 - val_loss: 0.1736\n",
      "\n",
      "Epoch 00032: val_loss improved from 0.17490 to 0.17361, saving model to ./keras.model_an_cnn\n",
      "Epoch 33/500\n",
      "5851/5851 [==============================] - 0s 55us/step - loss: 0.1726 - val_loss: 0.1742\n",
      "\n",
      "Epoch 00033: val_loss did not improve from 0.17361\n",
      "Epoch 34/500\n",
      "5851/5851 [==============================] - 1s 117us/step - loss: 0.1724 - val_loss: 0.1736\n",
      "\n",
      "Epoch 00034: val_loss improved from 0.17361 to 0.17361, saving model to ./keras.model_an_cnn\n",
      "Epoch 35/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1714 - val_loss: 0.1723\n",
      "\n",
      "Epoch 00035: val_loss improved from 0.17361 to 0.17226, saving model to ./keras.model_an_cnn\n",
      "Epoch 36/500\n",
      "5851/5851 [==============================] - 1s 92us/step - loss: 0.1708 - val_loss: 0.1725\n",
      "\n",
      "Epoch 00036: val_loss did not improve from 0.17226\n",
      "Epoch 37/500\n",
      "5851/5851 [==============================] - 0s 62us/step - loss: 0.1705 - val_loss: 0.1710\n",
      "\n",
      "Epoch 00037: val_loss improved from 0.17226 to 0.17099, saving model to ./keras.model_an_cnn\n",
      "Epoch 38/500\n",
      "5851/5851 [==============================] - 0s 40us/step - loss: 0.1700 - val_loss: 0.1714\n",
      "\n",
      "Epoch 00038: val_loss did not improve from 0.17099\n",
      "Epoch 39/500\n",
      "5851/5851 [==============================] - 0s 50us/step - loss: 0.1694 - val_loss: 0.1703\n",
      "\n",
      "Epoch 00039: val_loss improved from 0.17099 to 0.17030, saving model to ./keras.model_an_cnn\n",
      "Epoch 40/500\n",
      "5851/5851 [==============================] - 0s 39us/step - loss: 0.1695 - val_loss: 0.1701\n",
      "\n",
      "Epoch 00040: val_loss improved from 0.17030 to 0.17006, saving model to ./keras.model_an_cnn\n",
      "Epoch 41/500\n",
      "5851/5851 [==============================] - 0s 62us/step - loss: 0.1688 - val_loss: 0.1695\n",
      "\n",
      "Epoch 00041: val_loss improved from 0.17006 to 0.16954, saving model to ./keras.model_an_cnn\n",
      "Epoch 42/500\n",
      "5851/5851 [==============================] - 0s 42us/step - loss: 0.1686 - val_loss: 0.1714\n",
      "\n",
      "Epoch 00042: val_loss did not improve from 0.16954\n",
      "Epoch 43/500\n",
      "5851/5851 [==============================] - 0s 56us/step - loss: 0.1683 - val_loss: 0.1686\n",
      "\n",
      "Epoch 00043: val_loss improved from 0.16954 to 0.16862, saving model to ./keras.model_an_cnn\n",
      "Epoch 44/500\n",
      "5851/5851 [==============================] - 0s 68us/step - loss: 0.1676 - val_loss: 0.1690\n",
      "\n",
      "Epoch 00044: val_loss did not improve from 0.16862\n",
      "Epoch 45/500\n",
      "5851/5851 [==============================] - 0s 82us/step - loss: 0.1674 - val_loss: 0.1686\n",
      "\n",
      "Epoch 00045: val_loss improved from 0.16862 to 0.16856, saving model to ./keras.model_an_cnn\n",
      "Epoch 46/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1672 - val_loss: 0.1681\n",
      "\n",
      "Epoch 00046: val_loss improved from 0.16856 to 0.16814, saving model to ./keras.model_an_cnn\n",
      "Epoch 47/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1672 - val_loss: 0.1683\n",
      "\n",
      "Epoch 00047: val_loss did not improve from 0.16814\n",
      "Epoch 48/500\n",
      "5851/5851 [==============================] - 0s 72us/step - loss: 0.1668 - val_loss: 0.1680\n",
      "\n",
      "Epoch 00048: val_loss improved from 0.16814 to 0.16798, saving model to ./keras.model_an_cnn\n",
      "Epoch 49/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1665 - val_loss: 0.1673\n",
      "\n",
      "Epoch 00049: val_loss improved from 0.16798 to 0.16735, saving model to ./keras.model_an_cnn\n",
      "Epoch 50/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1667 - val_loss: 0.1688\n",
      "\n",
      "Epoch 00050: val_loss did not improve from 0.16735\n",
      "Epoch 51/500\n",
      "5851/5851 [==============================] - 0s 77us/step - loss: 0.1663 - val_loss: 0.1685\n",
      "\n",
      "Epoch 00051: val_loss did not improve from 0.16735\n",
      "Epoch 52/500\n",
      "5851/5851 [==============================] - 0s 72us/step - loss: 0.1659 - val_loss: 0.1671\n",
      "\n",
      "Epoch 00052: val_loss improved from 0.16735 to 0.16706, saving model to ./keras.model_an_cnn\n",
      "Epoch 53/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1655 - val_loss: 0.1669\n",
      "\n",
      "Epoch 00053: val_loss improved from 0.16706 to 0.16690, saving model to ./keras.model_an_cnn\n",
      "Epoch 54/500\n",
      "5851/5851 [==============================] - 0s 56us/step - loss: 0.1653 - val_loss: 0.1673\n",
      "\n",
      "Epoch 00054: val_loss did not improve from 0.16690\n",
      "Epoch 55/500\n",
      "5851/5851 [==============================] - 0s 42us/step - loss: 0.1653 - val_loss: 0.1667\n",
      "\n",
      "Epoch 00055: val_loss improved from 0.16690 to 0.16668, saving model to ./keras.model_an_cnn\n",
      "Epoch 56/500\n",
      "5851/5851 [==============================] - 0s 42us/step - loss: 0.1650 - val_loss: 0.1661\n",
      "\n",
      "Epoch 00056: val_loss improved from 0.16668 to 0.16614, saving model to ./keras.model_an_cnn\n",
      "Epoch 57/500\n",
      "5851/5851 [==============================] - 0s 36us/step - loss: 0.1647 - val_loss: 0.1654\n",
      "\n",
      "Epoch 00057: val_loss improved from 0.16614 to 0.16539, saving model to ./keras.model_an_cnn\n",
      "Epoch 58/500\n",
      "5851/5851 [==============================] - 0s 61us/step - loss: 0.1643 - val_loss: 0.1654\n",
      "\n",
      "Epoch 00058: val_loss did not improve from 0.16539\n",
      "Epoch 59/500\n",
      "5851/5851 [==============================] - 1s 129us/step - loss: 0.1645 - val_loss: 0.1651\n",
      "\n",
      "Epoch 00059: val_loss improved from 0.16539 to 0.16515, saving model to ./keras.model_an_cnn\n",
      "Epoch 60/500\n",
      "5851/5851 [==============================] - 0s 49us/step - loss: 0.1641 - val_loss: 0.1655\n",
      "\n",
      "Epoch 00060: val_loss did not improve from 0.16515\n",
      "Epoch 61/500\n",
      "5851/5851 [==============================] - 1s 96us/step - loss: 0.1641 - val_loss: 0.1651\n",
      "\n",
      "Epoch 00061: val_loss improved from 0.16515 to 0.16508, saving model to ./keras.model_an_cnn\n",
      "Epoch 62/500\n",
      "5851/5851 [==============================] - 0s 39us/step - loss: 0.1639 - val_loss: 0.1647\n",
      "\n",
      "Epoch 00062: val_loss improved from 0.16508 to 0.16469, saving model to ./keras.model_an_cnn\n",
      "Epoch 63/500\n",
      "5851/5851 [==============================] - 0s 51us/step - loss: 0.1635 - val_loss: 0.1647\n",
      "\n",
      "Epoch 00063: val_loss improved from 0.16469 to 0.16466, saving model to ./keras.model_an_cnn\n",
      "Epoch 64/500\n",
      "5851/5851 [==============================] - 0s 49us/step - loss: 0.1633 - val_loss: 0.1645\n",
      "\n",
      "Epoch 00064: val_loss improved from 0.16466 to 0.16450, saving model to ./keras.model_an_cnn\n",
      "Epoch 65/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1629 - val_loss: 0.1645\n",
      "\n",
      "Epoch 00065: val_loss improved from 0.16450 to 0.16446, saving model to ./keras.model_an_cnn\n",
      "Epoch 66/500\n",
      "5851/5851 [==============================] - 0s 64us/step - loss: 0.1631 - val_loss: 0.1643\n",
      "\n",
      "Epoch 00066: val_loss improved from 0.16446 to 0.16434, saving model to ./keras.model_an_cnn\n",
      "Epoch 67/500\n",
      "5851/5851 [==============================] - 0s 79us/step - loss: 0.1630 - val_loss: 0.1644\n",
      "\n",
      "Epoch 00067: val_loss did not improve from 0.16434\n",
      "Epoch 68/500\n",
      "5851/5851 [==============================] - 1s 100us/step - loss: 0.1629 - val_loss: 0.1637\n",
      "\n",
      "Epoch 00068: val_loss improved from 0.16434 to 0.16366, saving model to ./keras.model_an_cnn\n",
      "Epoch 69/500\n",
      "5851/5851 [==============================] - 0s 38us/step - loss: 0.1622 - val_loss: 0.1632\n",
      "\n",
      "Epoch 00069: val_loss improved from 0.16366 to 0.16323, saving model to ./keras.model_an_cnn\n",
      "Epoch 70/500\n",
      "5851/5851 [==============================] - 0s 43us/step - loss: 0.1622 - val_loss: 0.1641\n",
      "\n",
      "Epoch 00070: val_loss did not improve from 0.16323\n",
      "Epoch 71/500\n",
      "5851/5851 [==============================] - 0s 64us/step - loss: 0.1622 - val_loss: 0.1632\n",
      "\n",
      "Epoch 00071: val_loss improved from 0.16323 to 0.16316, saving model to ./keras.model_an_cnn\n",
      "Epoch 72/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1618 - val_loss: 0.1630\n",
      "\n",
      "Epoch 00072: val_loss improved from 0.16316 to 0.16301, saving model to ./keras.model_an_cnn\n",
      "Epoch 73/500\n",
      "5851/5851 [==============================] - 0s 42us/step - loss: 0.1617 - val_loss: 0.1628\n",
      "\n",
      "Epoch 00073: val_loss improved from 0.16301 to 0.16277, saving model to ./keras.model_an_cnn\n",
      "Epoch 74/500\n",
      "5851/5851 [==============================] - 0s 40us/step - loss: 0.1615 - val_loss: 0.1632\n",
      "\n",
      "Epoch 00074: val_loss did not improve from 0.16277\n",
      "Epoch 75/500\n",
      "5851/5851 [==============================] - 0s 51us/step - loss: 0.1616 - val_loss: 0.1637\n",
      "\n",
      "Epoch 00075: val_loss did not improve from 0.16277\n",
      "Epoch 76/500\n",
      "5851/5851 [==============================] - 0s 64us/step - loss: 0.1616 - val_loss: 0.1630\n",
      "\n",
      "Epoch 00076: val_loss did not improve from 0.16277\n",
      "Epoch 77/500\n",
      "5851/5851 [==============================] - 0s 81us/step - loss: 0.1613 - val_loss: 0.1633\n",
      "\n",
      "Epoch 00077: val_loss did not improve from 0.16277\n",
      "Epoch 78/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1605 - val_loss: 0.1618\n",
      "\n",
      "Epoch 00078: val_loss improved from 0.16277 to 0.16178, saving model to ./keras.model_an_cnn\n",
      "Epoch 79/500\n",
      "5851/5851 [==============================] - 0s 44us/step - loss: 0.1605 - val_loss: 0.1623\n",
      "\n",
      "Epoch 00079: val_loss did not improve from 0.16178\n",
      "Epoch 80/500\n",
      "5851/5851 [==============================] - 0s 66us/step - loss: 0.1605 - val_loss: 0.1637\n",
      "\n",
      "Epoch 00080: val_loss did not improve from 0.16178\n",
      "Epoch 81/500\n",
      "5851/5851 [==============================] - 0s 63us/step - loss: 0.1607 - val_loss: 0.1618\n",
      "\n",
      "Epoch 00081: val_loss did not improve from 0.16178\n",
      "Epoch 82/500\n",
      "5851/5851 [==============================] - 0s 71us/step - loss: 0.1602 - val_loss: 0.1617\n",
      "\n",
      "Epoch 00082: val_loss improved from 0.16178 to 0.16173, saving model to ./keras.model_an_cnn\n",
      "Epoch 83/500\n",
      "5851/5851 [==============================] - 0s 43us/step - loss: 0.1603 - val_loss: 0.1618\n",
      "\n",
      "Epoch 00083: val_loss did not improve from 0.16173\n",
      "\n",
      "Epoch 00083: ReduceLROnPlateau reducing learning rate to 0.00010000000474974513.\n",
      "Epoch 84/500\n",
      "5851/5851 [==============================] - 0s 39us/step - loss: 0.1592 - val_loss: 0.1607\n",
      "\n",
      "Epoch 00084: val_loss improved from 0.16173 to 0.16071, saving model to ./keras.model_an_cnn\n",
      "Epoch 85/500\n",
      "5851/5851 [==============================] - 0s 60us/step - loss: 0.1590 - val_loss: 0.1606\n",
      "\n",
      "Epoch 00085: val_loss improved from 0.16071 to 0.16061, saving model to ./keras.model_an_cnn\n",
      "Epoch 86/500\n",
      "5851/5851 [==============================] - 0s 54us/step - loss: 0.1590 - val_loss: 0.1606\n",
      "\n",
      "Epoch 00086: val_loss improved from 0.16061 to 0.16059, saving model to ./keras.model_an_cnn\n",
      "Epoch 87/500\n",
      "5851/5851 [==============================] - 0s 39us/step - loss: 0.1590 - val_loss: 0.1605\n",
      "\n",
      "Epoch 00087: val_loss improved from 0.16059 to 0.16052, saving model to ./keras.model_an_cnn\n",
      "Epoch 88/500\n",
      "5851/5851 [==============================] - 0s 72us/step - loss: 0.1589 - val_loss: 0.1606\n",
      "\n",
      "Epoch 00088: val_loss did not improve from 0.16052\n",
      "Epoch 89/500\n",
      "5851/5851 [==============================] - 0s 66us/step - loss: 0.1589 - val_loss: 0.1605\n",
      "\n",
      "Epoch 00089: val_loss improved from 0.16052 to 0.16049, saving model to ./keras.model_an_cnn\n",
      "Epoch 90/500\n",
      "5851/5851 [==============================] - 0s 53us/step - loss: 0.1589 - val_loss: 0.1605\n",
      "\n",
      "Epoch 00090: val_loss did not improve from 0.16049\n",
      "Epoch 91/500\n",
      "5851/5851 [==============================] - 1s 93us/step - loss: 0.1589 - val_loss: 0.1605\n",
      "\n",
      "Epoch 00091: val_loss improved from 0.16049 to 0.16046, saving model to ./keras.model_an_cnn\n",
      "Epoch 92/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1589 - val_loss: 0.1604\n",
      "\n",
      "Epoch 00092: val_loss improved from 0.16046 to 0.16041, saving model to ./keras.model_an_cnn\n",
      "Epoch 93/500\n",
      "5851/5851 [==============================] - 0s 38us/step - loss: 0.1589 - val_loss: 0.1604\n",
      "\n",
      "Epoch 00093: val_loss did not improve from 0.16041\n",
      "Epoch 94/500\n",
      "5851/5851 [==============================] - 0s 81us/step - loss: 0.1588 - val_loss: 0.1604\n",
      "\n",
      "Epoch 00094: val_loss did not improve from 0.16041\n",
      "\n",
      "Epoch 00094: ReduceLROnPlateau reducing learning rate to 1.0000000474974514e-05.\n",
      "Epoch 95/500\n",
      "5851/5851 [==============================] - 1s 97us/step - loss: 0.1588 - val_loss: 0.1604\n",
      "\n",
      "Epoch 00095: val_loss improved from 0.16041 to 0.16038, saving model to ./keras.model_an_cnn\n",
      "Epoch 96/500\n",
      "5851/5851 [==============================] - 0s 38us/step - loss: 0.1587 - val_loss: 0.1604\n",
      "\n",
      "Epoch 00096: val_loss improved from 0.16038 to 0.16037, saving model to ./keras.model_an_cnn\n",
      "Epoch 97/500\n",
      "5851/5851 [==============================] - 0s 55us/step - loss: 0.1587 - val_loss: 0.1604\n",
      "\n",
      "Epoch 00097: val_loss did not improve from 0.16037\n",
      "Epoch 98/500\n",
      "5851/5851 [==============================] - 1s 99us/step - loss: 0.1587 - val_loss: 0.1604\n",
      "\n",
      "Epoch 00098: val_loss improved from 0.16037 to 0.16037, saving model to ./keras.model_an_cnn\n",
      "Epoch 99/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1587 - val_loss: 0.1604\n",
      "\n",
      "Epoch 00099: val_loss did not improve from 0.16037\n",
      "Epoch 100/500\n",
      "5851/5851 [==============================] - 0s 77us/step - loss: 0.1587 - val_loss: 0.1604\n",
      "\n",
      "Epoch 00100: val_loss improved from 0.16037 to 0.16036, saving model to ./keras.model_an_cnn\n",
      "\n",
      "Epoch 00100: ReduceLROnPlateau reducing learning rate to 1e-05.\n",
      "Epoch 101/500\n",
      "5851/5851 [==============================] - 0s 45us/step - loss: 0.1587 - val_loss: 0.1604\n",
      "\n",
      "Epoch 00101: val_loss improved from 0.16036 to 0.16036, saving model to ./keras.model_an_cnn\n",
      "Epoch 102/500\n",
      "5851/5851 [==============================] - 0s 39us/step - loss: 0.1587 - val_loss: 0.1604\n",
      "\n",
      "Epoch 00102: val_loss improved from 0.16036 to 0.16036, saving model to ./keras.model_an_cnn\n",
      "Epoch 103/500\n",
      "5851/5851 [==============================] - 0s 36us/step - loss: 0.1587 - val_loss: 0.1604\n",
      "\n",
      "Epoch 00103: val_loss did not improve from 0.16036\n",
      "Epoch 104/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00104: val_loss improved from 0.16036 to 0.16035, saving model to ./keras.model_an_cnn\n",
      "Epoch 105/500\n",
      "5851/5851 [==============================] - 0s 48us/step - loss: 0.1587 - val_loss: 0.1604\n",
      "\n",
      "Epoch 00105: val_loss did not improve from 0.16035\n",
      "Epoch 106/500\n",
      "5851/5851 [==============================] - 1s 112us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00106: val_loss improved from 0.16035 to 0.16034, saving model to ./keras.model_an_cnn\n",
      "Epoch 107/500\n",
      "5851/5851 [==============================] - 0s 43us/step - loss: 0.1587 - val_loss: 0.1604\n",
      "\n",
      "Epoch 00107: val_loss did not improve from 0.16034\n",
      "Epoch 108/500\n",
      "5851/5851 [==============================] - 1s 105us/step - loss: 0.1587 - val_loss: 0.1604\n",
      "\n",
      "Epoch 00108: val_loss did not improve from 0.16034\n",
      "Epoch 109/500\n",
      "5851/5851 [==============================] - 0s 65us/step - loss: 0.1587 - val_loss: 0.1604\n",
      "\n",
      "Epoch 00109: val_loss did not improve from 0.16034\n",
      "Epoch 110/500\n",
      "5851/5851 [==============================] - 0s 73us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00110: val_loss improved from 0.16034 to 0.16034, saving model to ./keras.model_an_cnn\n",
      "Epoch 111/500\n",
      "5851/5851 [==============================] - 0s 38us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00111: val_loss did not improve from 0.16034\n",
      "Epoch 112/500\n",
      "5851/5851 [==============================] - 0s 52us/step - loss: 0.1587 - val_loss: 0.1604\n",
      "\n",
      "Epoch 00112: val_loss did not improve from 0.16034\n",
      "Epoch 113/500\n",
      "5851/5851 [==============================] - 0s 84us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00113: val_loss improved from 0.16034 to 0.16033, saving model to ./keras.model_an_cnn\n",
      "Epoch 114/500\n",
      "5851/5851 [==============================] - 0s 45us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00114: val_loss did not improve from 0.16033\n",
      "Epoch 115/500\n",
      "5851/5851 [==============================] - 1s 95us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00115: val_loss improved from 0.16033 to 0.16033, saving model to ./keras.model_an_cnn\n",
      "Epoch 116/500\n",
      "5851/5851 [==============================] - 0s 39us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00116: val_loss improved from 0.16033 to 0.16033, saving model to ./keras.model_an_cnn\n",
      "Epoch 117/500\n",
      "5851/5851 [==============================] - 0s 45us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00117: val_loss improved from 0.16033 to 0.16032, saving model to ./keras.model_an_cnn\n",
      "Epoch 118/500\n",
      "5851/5851 [==============================] - 0s 40us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00118: val_loss improved from 0.16032 to 0.16032, saving model to ./keras.model_an_cnn\n",
      "Epoch 119/500\n",
      "5851/5851 [==============================] - 0s 48us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00119: val_loss improved from 0.16032 to 0.16031, saving model to ./keras.model_an_cnn\n",
      "Epoch 120/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00120: val_loss did not improve from 0.16031\n",
      "Epoch 121/500\n",
      "5851/5851 [==============================] - 0s 49us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00121: val_loss improved from 0.16031 to 0.16031, saving model to ./keras.model_an_cnn\n",
      "Epoch 122/500\n",
      "5851/5851 [==============================] - 0s 71us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00122: val_loss improved from 0.16031 to 0.16031, saving model to ./keras.model_an_cnn\n",
      "Epoch 123/500\n",
      "5851/5851 [==============================] - 0s 49us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00123: val_loss did not improve from 0.16031\n",
      "Epoch 124/500\n",
      "5851/5851 [==============================] - 1s 114us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00124: val_loss did not improve from 0.16031\n",
      "Epoch 125/500\n",
      "5851/5851 [==============================] - 1s 100us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00125: val_loss did not improve from 0.16031\n",
      "Epoch 126/500\n",
      "5851/5851 [==============================] - 1s 134us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00126: val_loss improved from 0.16031 to 0.16029, saving model to ./keras.model_an_cnn\n",
      "Epoch 127/500\n",
      "5851/5851 [==============================] - 0s 44us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00127: val_loss did not improve from 0.16029\n",
      "Epoch 128/500\n",
      "5851/5851 [==============================] - 0s 71us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00128: val_loss did not improve from 0.16029\n",
      "Epoch 129/500\n",
      "5851/5851 [==============================] - 0s 71us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00129: val_loss did not improve from 0.16029\n",
      "Epoch 130/500\n",
      "5851/5851 [==============================] - 0s 66us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00130: val_loss improved from 0.16029 to 0.16028, saving model to ./keras.model_an_cnn\n",
      "Epoch 131/500\n",
      "5851/5851 [==============================] - 0s 52us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00131: val_loss did not improve from 0.16028\n",
      "Epoch 132/500\n",
      "5851/5851 [==============================] - 1s 98us/step - loss: 0.1586 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00132: val_loss did not improve from 0.16028\n",
      "Epoch 133/500\n",
      "5851/5851 [==============================] - 1s 97us/step - loss: 0.1587 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00133: val_loss improved from 0.16028 to 0.16027, saving model to ./keras.model_an_cnn\n",
      "Epoch 134/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1586 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00134: val_loss did not improve from 0.16027\n",
      "Epoch 135/500\n",
      "5851/5851 [==============================] - 0s 52us/step - loss: 0.1586 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00135: val_loss improved from 0.16027 to 0.16026, saving model to ./keras.model_an_cnn\n",
      "Epoch 136/500\n",
      "5851/5851 [==============================] - 0s 62us/step - loss: 0.1586 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00136: val_loss did not improve from 0.16026\n",
      "Epoch 137/500\n",
      "5851/5851 [==============================] - 1s 99us/step - loss: 0.1586 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00137: val_loss did not improve from 0.16026\n",
      "Epoch 138/500\n",
      "5851/5851 [==============================] - 0s 58us/step - loss: 0.1586 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00138: val_loss did not improve from 0.16026\n",
      "Epoch 139/500\n",
      "5851/5851 [==============================] - 0s 79us/step - loss: 0.1586 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00139: val_loss improved from 0.16026 to 0.16026, saving model to ./keras.model_an_cnn\n",
      "Epoch 140/500\n",
      "5851/5851 [==============================] - 0s 52us/step - loss: 0.1586 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00140: val_loss did not improve from 0.16026\n",
      "Epoch 141/500\n",
      "5851/5851 [==============================] - 0s 66us/step - loss: 0.1586 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00141: val_loss did not improve from 0.16026\n",
      "Epoch 142/500\n",
      "5851/5851 [==============================] - 1s 94us/step - loss: 0.1586 - val_loss: 0.1603\n",
      "\n",
      "Epoch 00142: val_loss did not improve from 0.16026\n",
      "Epoch 143/500\n",
      "5851/5851 [==============================] - 1s 90us/step - loss: 0.1586 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00143: val_loss improved from 0.16026 to 0.16025, saving model to ./keras.model_an_cnn\n",
      "Epoch 144/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1586 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00144: val_loss improved from 0.16025 to 0.16024, saving model to ./keras.model_an_cnn\n",
      "Epoch 145/500\n",
      "5851/5851 [==============================] - 0s 54us/step - loss: 0.1586 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00145: val_loss improved from 0.16024 to 0.16022, saving model to ./keras.model_an_cnn\n",
      "Epoch 146/500\n",
      "5851/5851 [==============================] - 0s 43us/step - loss: 0.1586 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00146: val_loss did not improve from 0.16022\n",
      "Epoch 147/500\n",
      "5851/5851 [==============================] - 0s 77us/step - loss: 0.1586 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00147: val_loss did not improve from 0.16022\n",
      "Epoch 148/500\n",
      "5851/5851 [==============================] - 1s 86us/step - loss: 0.1586 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00148: val_loss improved from 0.16022 to 0.16022, saving model to ./keras.model_an_cnn\n",
      "Epoch 149/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1586 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00149: val_loss improved from 0.16022 to 0.16022, saving model to ./keras.model_an_cnn\n",
      "Epoch 150/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1586 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00150: val_loss improved from 0.16022 to 0.16022, saving model to ./keras.model_an_cnn\n",
      "Epoch 151/500\n",
      "5851/5851 [==============================] - 0s 46us/step - loss: 0.1586 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00151: val_loss did not improve from 0.16022\n",
      "Epoch 152/500\n",
      "5851/5851 [==============================] - 1s 108us/step - loss: 0.1586 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00152: val_loss did not improve from 0.16022\n",
      "Epoch 153/500\n",
      "5851/5851 [==============================] - 0s 74us/step - loss: 0.1586 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00153: val_loss did not improve from 0.16022\n",
      "Epoch 154/500\n",
      "5851/5851 [==============================] - 0s 78us/step - loss: 0.1586 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00154: val_loss improved from 0.16022 to 0.16022, saving model to ./keras.model_an_cnn\n",
      "Epoch 155/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1586 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00155: val_loss improved from 0.16022 to 0.16020, saving model to ./keras.model_an_cnn\n",
      "Epoch 156/500\n",
      "5851/5851 [==============================] - 0s 38us/step - loss: 0.1586 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00156: val_loss improved from 0.16020 to 0.16019, saving model to ./keras.model_an_cnn\n",
      "Epoch 157/500\n",
      "5851/5851 [==============================] - 0s 42us/step - loss: 0.1586 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00157: val_loss did not improve from 0.16019\n",
      "Epoch 158/500\n",
      "5851/5851 [==============================] - 1s 85us/step - loss: 0.1586 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00158: val_loss did not improve from 0.16019\n",
      "Epoch 159/500\n",
      "5851/5851 [==============================] - 0s 71us/step - loss: 0.1586 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00159: val_loss did not improve from 0.16019\n",
      "Epoch 160/500\n",
      "5851/5851 [==============================] - 0s 77us/step - loss: 0.1585 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00160: val_loss improved from 0.16019 to 0.16018, saving model to ./keras.model_an_cnn\n",
      "Epoch 161/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1585 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00161: val_loss did not improve from 0.16018\n",
      "Epoch 162/500\n",
      "5851/5851 [==============================] - 0s 43us/step - loss: 0.1585 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00162: val_loss did not improve from 0.16018\n",
      "Epoch 163/500\n",
      "5851/5851 [==============================] - 0s 51us/step - loss: 0.1585 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00163: val_loss improved from 0.16018 to 0.16016, saving model to ./keras.model_an_cnn\n",
      "Epoch 164/500\n",
      "5851/5851 [==============================] - 0s 38us/step - loss: 0.1585 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00164: val_loss did not improve from 0.16016\n",
      "Epoch 165/500\n",
      "5851/5851 [==============================] - 0s 51us/step - loss: 0.1585 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00165: val_loss did not improve from 0.16016\n",
      "Epoch 166/500\n",
      "5851/5851 [==============================] - 1s 99us/step - loss: 0.1585 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00166: val_loss did not improve from 0.16016\n",
      "Epoch 167/500\n",
      "5851/5851 [==============================] - 0s 56us/step - loss: 0.1585 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00167: val_loss improved from 0.16016 to 0.16016, saving model to ./keras.model_an_cnn\n",
      "Epoch 168/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1585 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00168: val_loss did not improve from 0.16016\n",
      "Epoch 169/500\n",
      "5851/5851 [==============================] - 0s 63us/step - loss: 0.1585 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00169: val_loss improved from 0.16016 to 0.16015, saving model to ./keras.model_an_cnn\n",
      "Epoch 170/500\n",
      "5851/5851 [==============================] - 0s 47us/step - loss: 0.1585 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00170: val_loss did not improve from 0.16015\n",
      "Epoch 171/500\n",
      "5851/5851 [==============================] - 0s 83us/step - loss: 0.1585 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00171: val_loss improved from 0.16015 to 0.16012, saving model to ./keras.model_an_cnn\n",
      "Epoch 172/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1585 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00172: val_loss did not improve from 0.16012\n",
      "Epoch 173/500\n",
      "5851/5851 [==============================] - 0s 66us/step - loss: 0.1585 - val_loss: 0.1602\n",
      "\n",
      "Epoch 00173: val_loss did not improve from 0.16012\n",
      "Epoch 174/500\n",
      "5851/5851 [==============================] - 0s 69us/step - loss: 0.1585 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00174: val_loss did not improve from 0.16012\n",
      "Epoch 175/500\n",
      "5851/5851 [==============================] - 0s 83us/step - loss: 0.1585 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00175: val_loss improved from 0.16012 to 0.16011, saving model to ./keras.model_an_cnn\n",
      "Epoch 176/500\n",
      "5851/5851 [==============================] - 0s 60us/step - loss: 0.1585 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00176: val_loss did not improve from 0.16011\n",
      "Epoch 177/500\n",
      "5851/5851 [==============================] - 1s 133us/step - loss: 0.1585 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00177: val_loss did not improve from 0.16011\n",
      "Epoch 178/500\n",
      "5851/5851 [==============================] - 0s 76us/step - loss: 0.1585 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00178: val_loss did not improve from 0.16011\n",
      "Epoch 179/500\n",
      "5851/5851 [==============================] - 1s 90us/step - loss: 0.1585 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00179: val_loss did not improve from 0.16011\n",
      "Epoch 180/500\n",
      "5851/5851 [==============================] - 1s 125us/step - loss: 0.1585 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00180: val_loss did not improve from 0.16011\n",
      "Epoch 181/500\n",
      "5851/5851 [==============================] - 1s 88us/step - loss: 0.1585 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00181: val_loss did not improve from 0.16011\n",
      "Epoch 182/500\n",
      "5851/5851 [==============================] - 0s 59us/step - loss: 0.1585 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00182: val_loss improved from 0.16011 to 0.16010, saving model to ./keras.model_an_cnn\n",
      "Epoch 183/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1585 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00183: val_loss did not improve from 0.16010\n",
      "Epoch 184/500\n",
      "5851/5851 [==============================] - 0s 59us/step - loss: 0.1585 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00184: val_loss did not improve from 0.16010\n",
      "Epoch 185/500\n",
      "5851/5851 [==============================] - 1s 123us/step - loss: 0.1584 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00185: val_loss improved from 0.16010 to 0.16008, saving model to ./keras.model_an_cnn\n",
      "Epoch 186/500\n",
      "5851/5851 [==============================] - 0s 48us/step - loss: 0.1584 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00186: val_loss did not improve from 0.16008\n",
      "Epoch 187/500\n",
      "5851/5851 [==============================] - 0s 70us/step - loss: 0.1584 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00187: val_loss did not improve from 0.16008\n",
      "Epoch 188/500\n",
      "5851/5851 [==============================] - 1s 98us/step - loss: 0.1584 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00188: val_loss improved from 0.16008 to 0.16007, saving model to ./keras.model_an_cnn\n",
      "Epoch 189/500\n",
      "5851/5851 [==============================] - 0s 40us/step - loss: 0.1584 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00189: val_loss improved from 0.16007 to 0.16007, saving model to ./keras.model_an_cnn\n",
      "Epoch 190/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1584 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00190: val_loss improved from 0.16007 to 0.16007, saving model to ./keras.model_an_cnn\n",
      "Epoch 191/500\n",
      "5851/5851 [==============================] - 0s 40us/step - loss: 0.1584 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00191: val_loss improved from 0.16007 to 0.16005, saving model to ./keras.model_an_cnn\n",
      "Epoch 192/500\n",
      "5851/5851 [==============================] - 0s 50us/step - loss: 0.1584 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00192: val_loss did not improve from 0.16005\n",
      "Epoch 193/500\n",
      "5851/5851 [==============================] - 0s 65us/step - loss: 0.1584 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00193: val_loss improved from 0.16005 to 0.16005, saving model to ./keras.model_an_cnn\n",
      "Epoch 194/500\n",
      "5851/5851 [==============================] - 0s 45us/step - loss: 0.1584 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00194: val_loss did not improve from 0.16005\n",
      "Epoch 195/500\n",
      "5851/5851 [==============================] - 0s 62us/step - loss: 0.1584 - val_loss: 0.1601\n",
      "\n",
      "Epoch 00195: val_loss did not improve from 0.16005\n",
      "Epoch 196/500\n",
      "5851/5851 [==============================] - 1s 88us/step - loss: 0.1584 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00196: val_loss improved from 0.16005 to 0.16004, saving model to ./keras.model_an_cnn\n",
      "Epoch 197/500\n",
      "5851/5851 [==============================] - 0s 50us/step - loss: 0.1584 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00197: val_loss improved from 0.16004 to 0.16004, saving model to ./keras.model_an_cnn\n",
      "Epoch 198/500\n",
      "5851/5851 [==============================] - 0s 52us/step - loss: 0.1584 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00198: val_loss improved from 0.16004 to 0.16003, saving model to ./keras.model_an_cnn\n",
      "Epoch 199/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1584 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00199: val_loss did not improve from 0.16003\n",
      "Epoch 200/500\n",
      "5851/5851 [==============================] - 0s 75us/step - loss: 0.1584 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00200: val_loss did not improve from 0.16003\n",
      "Epoch 201/500\n",
      "5851/5851 [==============================] - 0s 69us/step - loss: 0.1584 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00201: val_loss improved from 0.16003 to 0.16001, saving model to ./keras.model_an_cnn\n",
      "Epoch 202/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1584 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00202: val_loss did not improve from 0.16001\n",
      "Epoch 203/500\n",
      "5851/5851 [==============================] - 0s 56us/step - loss: 0.1584 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00203: val_loss improved from 0.16001 to 0.16000, saving model to ./keras.model_an_cnn\n",
      "Epoch 204/500\n",
      "5851/5851 [==============================] - 0s 36us/step - loss: 0.1584 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00204: val_loss did not improve from 0.16000\n",
      "Epoch 205/500\n",
      "5851/5851 [==============================] - 0s 52us/step - loss: 0.1584 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00205: val_loss did not improve from 0.16000\n",
      "Epoch 206/500\n",
      "5851/5851 [==============================] - 0s 72us/step - loss: 0.1584 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00206: val_loss did not improve from 0.16000\n",
      "Epoch 207/500\n",
      "5851/5851 [==============================] - 0s 81us/step - loss: 0.1584 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00207: val_loss did not improve from 0.16000\n",
      "Epoch 208/500\n",
      "5851/5851 [==============================] - 1s 120us/step - loss: 0.1583 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00208: val_loss improved from 0.16000 to 0.15999, saving model to ./keras.model_an_cnn\n",
      "Epoch 209/500\n",
      "5851/5851 [==============================] - 0s 42us/step - loss: 0.1583 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00209: val_loss did not improve from 0.15999\n",
      "Epoch 210/500\n",
      "5851/5851 [==============================] - 0s 59us/step - loss: 0.1583 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00210: val_loss did not improve from 0.15999\n",
      "Epoch 211/500\n",
      "5851/5851 [==============================] - 0s 49us/step - loss: 0.1583 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00211: val_loss did not improve from 0.15999\n",
      "Epoch 212/500\n",
      "5851/5851 [==============================] - 0s 56us/step - loss: 0.1583 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00212: val_loss improved from 0.15999 to 0.15999, saving model to ./keras.model_an_cnn\n",
      "Epoch 213/500\n",
      "5851/5851 [==============================] - 0s 79us/step - loss: 0.1583 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00213: val_loss improved from 0.15999 to 0.15998, saving model to ./keras.model_an_cnn\n",
      "Epoch 214/500\n",
      "5851/5851 [==============================] - 0s 52us/step - loss: 0.1583 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00214: val_loss improved from 0.15998 to 0.15997, saving model to ./keras.model_an_cnn\n",
      "Epoch 215/500\n",
      "5851/5851 [==============================] - 0s 44us/step - loss: 0.1583 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00215: val_loss improved from 0.15997 to 0.15996, saving model to ./keras.model_an_cnn\n",
      "Epoch 216/500\n",
      "5851/5851 [==============================] - 0s 36us/step - loss: 0.1583 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00216: val_loss did not improve from 0.15996\n",
      "Epoch 217/500\n",
      "5851/5851 [==============================] - 0s 75us/step - loss: 0.1583 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00217: val_loss did not improve from 0.15996\n",
      "Epoch 218/500\n",
      "5851/5851 [==============================] - 1s 92us/step - loss: 0.1583 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00218: val_loss improved from 0.15996 to 0.15996, saving model to ./keras.model_an_cnn\n",
      "Epoch 219/500\n",
      "5851/5851 [==============================] - 0s 62us/step - loss: 0.1583 - val_loss: 0.1600\n",
      "\n",
      "Epoch 00219: val_loss improved from 0.15996 to 0.15995, saving model to ./keras.model_an_cnn\n",
      "Epoch 220/500\n",
      "5851/5851 [==============================] - 0s 40us/step - loss: 0.1583 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00220: val_loss improved from 0.15995 to 0.15995, saving model to ./keras.model_an_cnn\n",
      "Epoch 221/500\n",
      "5851/5851 [==============================] - 0s 51us/step - loss: 0.1583 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00221: val_loss improved from 0.15995 to 0.15994, saving model to ./keras.model_an_cnn\n",
      "Epoch 222/500\n",
      "5851/5851 [==============================] - 0s 51us/step - loss: 0.1583 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00222: val_loss improved from 0.15994 to 0.15994, saving model to ./keras.model_an_cnn\n",
      "Epoch 223/500\n",
      "5851/5851 [==============================] - 0s 46us/step - loss: 0.1583 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00223: val_loss improved from 0.15994 to 0.15993, saving model to ./keras.model_an_cnn\n",
      "Epoch 224/500\n",
      "5851/5851 [==============================] - 0s 58us/step - loss: 0.1583 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00224: val_loss improved from 0.15993 to 0.15993, saving model to ./keras.model_an_cnn\n",
      "Epoch 225/500\n",
      "5851/5851 [==============================] - 0s 50us/step - loss: 0.1583 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00225: val_loss did not improve from 0.15993\n",
      "Epoch 226/500\n",
      "5851/5851 [==============================] - 1s 111us/step - loss: 0.1583 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00226: val_loss did not improve from 0.15993\n",
      "Epoch 227/500\n",
      "5851/5851 [==============================] - 1s 145us/step - loss: 0.1583 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00227: val_loss improved from 0.15993 to 0.15992, saving model to ./keras.model_an_cnn\n",
      "Epoch 228/500\n",
      "5851/5851 [==============================] - 0s 56us/step - loss: 0.1583 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00228: val_loss improved from 0.15992 to 0.15992, saving model to ./keras.model_an_cnn\n",
      "Epoch 229/500\n",
      "5851/5851 [==============================] - 0s 38us/step - loss: 0.1583 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00229: val_loss did not improve from 0.15992\n",
      "Epoch 230/500\n",
      "5851/5851 [==============================] - 1s 91us/step - loss: 0.1583 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00230: val_loss improved from 0.15992 to 0.15992, saving model to ./keras.model_an_cnn\n",
      "Epoch 231/500\n",
      "5851/5851 [==============================] - 0s 42us/step - loss: 0.1583 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00231: val_loss did not improve from 0.15992\n",
      "Epoch 232/500\n",
      "5851/5851 [==============================] - 1s 142us/step - loss: 0.1582 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00232: val_loss improved from 0.15992 to 0.15991, saving model to ./keras.model_an_cnn\n",
      "Epoch 233/500\n",
      "5851/5851 [==============================] - 0s 67us/step - loss: 0.1582 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00233: val_loss improved from 0.15991 to 0.15991, saving model to ./keras.model_an_cnn\n",
      "Epoch 234/500\n",
      "5851/5851 [==============================] - 0s 57us/step - loss: 0.1582 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00234: val_loss improved from 0.15991 to 0.15989, saving model to ./keras.model_an_cnn\n",
      "Epoch 235/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1582 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00235: val_loss did not improve from 0.15989\n",
      "Epoch 236/500\n",
      "5851/5851 [==============================] - 1s 90us/step - loss: 0.1582 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00236: val_loss did not improve from 0.15989\n",
      "Epoch 237/500\n",
      "5851/5851 [==============================] - 0s 74us/step - loss: 0.1582 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00237: val_loss improved from 0.15989 to 0.15988, saving model to ./keras.model_an_cnn\n",
      "Epoch 238/500\n",
      "5851/5851 [==============================] - 0s 40us/step - loss: 0.1582 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00238: val_loss did not improve from 0.15988\n",
      "Epoch 239/500\n",
      "5851/5851 [==============================] - 0s 69us/step - loss: 0.1582 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00239: val_loss did not improve from 0.15988\n",
      "Epoch 240/500\n",
      "5851/5851 [==============================] - 1s 108us/step - loss: 0.1582 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00240: val_loss did not improve from 0.15988\n",
      "Epoch 241/500\n",
      "5851/5851 [==============================] - 0s 58us/step - loss: 0.1582 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00241: val_loss improved from 0.15988 to 0.15987, saving model to ./keras.model_an_cnn\n",
      "Epoch 242/500\n",
      "5851/5851 [==============================] - 0s 46us/step - loss: 0.1582 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00242: val_loss did not improve from 0.15987\n",
      "Epoch 243/500\n",
      "5851/5851 [==============================] - 1s 88us/step - loss: 0.1582 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00243: val_loss did not improve from 0.15987\n",
      "Epoch 244/500\n",
      "5851/5851 [==============================] - 0s 79us/step - loss: 0.1582 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00244: val_loss improved from 0.15987 to 0.15986, saving model to ./keras.model_an_cnn\n",
      "Epoch 245/500\n",
      "5851/5851 [==============================] - 0s 47us/step - loss: 0.1582 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00245: val_loss improved from 0.15986 to 0.15985, saving model to ./keras.model_an_cnn\n",
      "Epoch 246/500\n",
      "5851/5851 [==============================] - 0s 49us/step - loss: 0.1582 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00246: val_loss improved from 0.15985 to 0.15985, saving model to ./keras.model_an_cnn\n",
      "Epoch 247/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1582 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00247: val_loss did not improve from 0.15985\n",
      "Epoch 248/500\n",
      "5851/5851 [==============================] - 0s 54us/step - loss: 0.1582 - val_loss: 0.1599\n",
      "\n",
      "Epoch 00248: val_loss did not improve from 0.15985\n",
      "Epoch 249/500\n",
      "5851/5851 [==============================] - 1s 111us/step - loss: 0.1582 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00249: val_loss improved from 0.15985 to 0.15984, saving model to ./keras.model_an_cnn\n",
      "Epoch 250/500\n",
      "5851/5851 [==============================] - 0s 39us/step - loss: 0.1582 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00250: val_loss improved from 0.15984 to 0.15983, saving model to ./keras.model_an_cnn\n",
      "Epoch 251/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1582 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00251: val_loss improved from 0.15983 to 0.15982, saving model to ./keras.model_an_cnn\n",
      "Epoch 252/500\n",
      "5851/5851 [==============================] - 0s 42us/step - loss: 0.1582 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00252: val_loss did not improve from 0.15982\n",
      "Epoch 253/500\n",
      "5851/5851 [==============================] - 0s 54us/step - loss: 0.1582 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00253: val_loss did not improve from 0.15982\n",
      "Epoch 254/500\n",
      "5851/5851 [==============================] - 1s 89us/step - loss: 0.1582 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00254: val_loss did not improve from 0.15982\n",
      "Epoch 255/500\n",
      "5851/5851 [==============================] - 0s 55us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00255: val_loss improved from 0.15982 to 0.15981, saving model to ./keras.model_an_cnn\n",
      "Epoch 256/500\n",
      "5851/5851 [==============================] - 0s 71us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00256: val_loss did not improve from 0.15981\n",
      "Epoch 257/500\n",
      "5851/5851 [==============================] - 1s 86us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00257: val_loss improved from 0.15981 to 0.15980, saving model to ./keras.model_an_cnn\n",
      "Epoch 258/500\n",
      "5851/5851 [==============================] - 0s 38us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00258: val_loss did not improve from 0.15980\n",
      "Epoch 259/500\n",
      "5851/5851 [==============================] - 0s 81us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00259: val_loss improved from 0.15980 to 0.15980, saving model to ./keras.model_an_cnn\n",
      "Epoch 260/500\n",
      "5851/5851 [==============================] - 0s 38us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00260: val_loss did not improve from 0.15980\n",
      "Epoch 261/500\n",
      "5851/5851 [==============================] - 0s 77us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00261: val_loss did not improve from 0.15980\n",
      "Epoch 262/500\n",
      "5851/5851 [==============================] - 1s 100us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00262: val_loss improved from 0.15980 to 0.15979, saving model to ./keras.model_an_cnn\n",
      "Epoch 263/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00263: val_loss did not improve from 0.15979\n",
      "Epoch 264/500\n",
      "5851/5851 [==============================] - 1s 109us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00264: val_loss improved from 0.15979 to 0.15978, saving model to ./keras.model_an_cnn\n",
      "Epoch 265/500\n",
      "5851/5851 [==============================] - 0s 58us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00265: val_loss did not improve from 0.15978\n",
      "Epoch 266/500\n",
      "5851/5851 [==============================] - 0s 83us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00266: val_loss did not improve from 0.15978\n",
      "Epoch 267/500\n",
      "5851/5851 [==============================] - 0s 85us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00267: val_loss improved from 0.15978 to 0.15977, saving model to ./keras.model_an_cnn\n",
      "Epoch 268/500\n",
      "5851/5851 [==============================] - 0s 46us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00268: val_loss did not improve from 0.15977\n",
      "Epoch 269/500\n",
      "5851/5851 [==============================] - 0s 64us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00269: val_loss improved from 0.15977 to 0.15977, saving model to ./keras.model_an_cnn\n",
      "Epoch 270/500\n",
      "5851/5851 [==============================] - 0s 38us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00270: val_loss did not improve from 0.15977\n",
      "Epoch 271/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00271: val_loss did not improve from 0.15977\n",
      "Epoch 272/500\n",
      "5851/5851 [==============================] - 0s 46us/step - loss: 0.1581 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00272: val_loss improved from 0.15977 to 0.15975, saving model to ./keras.model_an_cnn\n",
      "Epoch 273/500\n",
      "5851/5851 [==============================] - 0s 50us/step - loss: 0.1581 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00273: val_loss did not improve from 0.15975\n",
      "Epoch 274/500\n",
      "5851/5851 [==============================] - 1s 95us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00274: val_loss did not improve from 0.15975\n",
      "Epoch 275/500\n",
      "5851/5851 [==============================] - 1s 111us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00275: val_loss did not improve from 0.15975\n",
      "Epoch 276/500\n",
      "5851/5851 [==============================] - 1s 111us/step - loss: 0.1581 - val_loss: 0.1598\n",
      "\n",
      "Epoch 00276: val_loss did not improve from 0.15975\n",
      "Epoch 277/500\n",
      "5851/5851 [==============================] - 1s 101us/step - loss: 0.1581 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00277: val_loss did not improve from 0.15975\n",
      "Epoch 278/500\n",
      "5851/5851 [==============================] - 0s 85us/step - loss: 0.1581 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00278: val_loss improved from 0.15975 to 0.15972, saving model to ./keras.model_an_cnn\n",
      "Epoch 279/500\n",
      "5851/5851 [==============================] - 0s 78us/step - loss: 0.1581 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00279: val_loss did not improve from 0.15972\n",
      "Epoch 280/500\n",
      "5851/5851 [==============================] - 0s 76us/step - loss: 0.1581 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00280: val_loss improved from 0.15972 to 0.15972, saving model to ./keras.model_an_cnn\n",
      "Epoch 281/500\n",
      "5851/5851 [==============================] - 0s 56us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00281: val_loss did not improve from 0.15972\n",
      "Epoch 282/500\n",
      "5851/5851 [==============================] - 1s 90us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00282: val_loss did not improve from 0.15972\n",
      "Epoch 283/500\n",
      "5851/5851 [==============================] - 1s 121us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00283: val_loss did not improve from 0.15972\n",
      "Epoch 284/500\n",
      "5851/5851 [==============================] - 0s 69us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00284: val_loss improved from 0.15972 to 0.15970, saving model to ./keras.model_an_cnn\n",
      "Epoch 285/500\n",
      "5851/5851 [==============================] - 0s 38us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00285: val_loss did not improve from 0.15970\n",
      "Epoch 286/500\n",
      "5851/5851 [==============================] - 0s 51us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00286: val_loss improved from 0.15970 to 0.15969, saving model to ./keras.model_an_cnn\n",
      "Epoch 287/500\n",
      "5851/5851 [==============================] - 0s 74us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00287: val_loss did not improve from 0.15969\n",
      "Epoch 288/500\n",
      "5851/5851 [==============================] - 1s 95us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00288: val_loss improved from 0.15969 to 0.15969, saving model to ./keras.model_an_cnn\n",
      "Epoch 289/500\n",
      "5851/5851 [==============================] - 0s 48us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00289: val_loss improved from 0.15969 to 0.15968, saving model to ./keras.model_an_cnn\n",
      "Epoch 290/500\n",
      "5851/5851 [==============================] - 0s 46us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00290: val_loss did not improve from 0.15968\n",
      "Epoch 291/500\n",
      "5851/5851 [==============================] - 0s 75us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00291: val_loss improved from 0.15968 to 0.15968, saving model to ./keras.model_an_cnn\n",
      "Epoch 292/500\n",
      "5851/5851 [==============================] - 0s 36us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00292: val_loss did not improve from 0.15968\n",
      "Epoch 293/500\n",
      "5851/5851 [==============================] - 0s 64us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00293: val_loss improved from 0.15968 to 0.15968, saving model to ./keras.model_an_cnn\n",
      "Epoch 294/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00294: val_loss improved from 0.15968 to 0.15967, saving model to ./keras.model_an_cnn\n",
      "Epoch 295/500\n",
      "5851/5851 [==============================] - 0s 40us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00295: val_loss did not improve from 0.15967\n",
      "Epoch 296/500\n",
      "5851/5851 [==============================] - 0s 55us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00296: val_loss improved from 0.15967 to 0.15965, saving model to ./keras.model_an_cnn\n",
      "Epoch 297/500\n",
      "5851/5851 [==============================] - 0s 70us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00297: val_loss did not improve from 0.15965\n",
      "Epoch 298/500\n",
      "5851/5851 [==============================] - 0s 79us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00298: val_loss improved from 0.15965 to 0.15965, saving model to ./keras.model_an_cnn\n",
      "Epoch 299/500\n",
      "5851/5851 [==============================] - 0s 45us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00299: val_loss did not improve from 0.15965\n",
      "Epoch 300/500\n",
      "5851/5851 [==============================] - 0s 52us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00300: val_loss did not improve from 0.15965\n",
      "Epoch 301/500\n",
      "5851/5851 [==============================] - 0s 71us/step - loss: 0.1580 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00301: val_loss improved from 0.15965 to 0.15965, saving model to ./keras.model_an_cnn\n",
      "Epoch 302/500\n",
      "5851/5851 [==============================] - 0s 50us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00302: val_loss did not improve from 0.15965\n",
      "Epoch 303/500\n",
      "5851/5851 [==============================] - 0s 61us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00303: val_loss did not improve from 0.15965\n",
      "Epoch 304/500\n",
      "5851/5851 [==============================] - 0s 52us/step - loss: 0.1580 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00304: val_loss improved from 0.15965 to 0.15964, saving model to ./keras.model_an_cnn\n",
      "Epoch 305/500\n",
      "5851/5851 [==============================] - 0s 57us/step - loss: 0.1580 - val_loss: 0.1597\n",
      "\n",
      "Epoch 00305: val_loss did not improve from 0.15964\n",
      "Epoch 306/500\n",
      "5851/5851 [==============================] - 0s 77us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00306: val_loss improved from 0.15964 to 0.15963, saving model to ./keras.model_an_cnn\n",
      "Epoch 307/500\n",
      "5851/5851 [==============================] - 0s 40us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00307: val_loss did not improve from 0.15963\n",
      "Epoch 308/500\n",
      "5851/5851 [==============================] - 0s 46us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00308: val_loss did not improve from 0.15963\n",
      "Epoch 309/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00309: val_loss did not improve from 0.15963\n",
      "Epoch 310/500\n",
      "5851/5851 [==============================] - 0s 43us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00310: val_loss improved from 0.15963 to 0.15962, saving model to ./keras.model_an_cnn\n",
      "Epoch 311/500\n",
      "5851/5851 [==============================] - 0s 43us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00311: val_loss improved from 0.15962 to 0.15961, saving model to ./keras.model_an_cnn\n",
      "Epoch 312/500\n",
      "5851/5851 [==============================] - 0s 56us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00312: val_loss improved from 0.15961 to 0.15961, saving model to ./keras.model_an_cnn\n",
      "Epoch 313/500\n",
      "5851/5851 [==============================] - 0s 36us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00313: val_loss did not improve from 0.15961\n",
      "Epoch 314/500\n",
      "5851/5851 [==============================] - 0s 42us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00314: val_loss improved from 0.15961 to 0.15960, saving model to ./keras.model_an_cnn\n",
      "Epoch 315/500\n",
      "5851/5851 [==============================] - 0s 47us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00315: val_loss improved from 0.15960 to 0.15960, saving model to ./keras.model_an_cnn\n",
      "Epoch 316/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00316: val_loss did not improve from 0.15960\n",
      "Epoch 317/500\n",
      "5851/5851 [==============================] - 0s 49us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00317: val_loss improved from 0.15960 to 0.15958, saving model to ./keras.model_an_cnn\n",
      "Epoch 318/500\n",
      "5851/5851 [==============================] - 0s 46us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00318: val_loss did not improve from 0.15958\n",
      "Epoch 319/500\n",
      "5851/5851 [==============================] - 0s 69us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00319: val_loss did not improve from 0.15958\n",
      "Epoch 320/500\n",
      "5851/5851 [==============================] - 1s 109us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00320: val_loss did not improve from 0.15958\n",
      "Epoch 321/500\n",
      "5851/5851 [==============================] - 1s 96us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00321: val_loss did not improve from 0.15958\n",
      "Epoch 322/500\n",
      "5851/5851 [==============================] - 1s 90us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00322: val_loss improved from 0.15958 to 0.15958, saving model to ./keras.model_an_cnn\n",
      "Epoch 323/500\n",
      "5851/5851 [==============================] - 0s 76us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00323: val_loss improved from 0.15958 to 0.15958, saving model to ./keras.model_an_cnn\n",
      "Epoch 324/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00324: val_loss improved from 0.15958 to 0.15957, saving model to ./keras.model_an_cnn\n",
      "Epoch 325/500\n",
      "5851/5851 [==============================] - 0s 42us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00325: val_loss improved from 0.15957 to 0.15957, saving model to ./keras.model_an_cnn\n",
      "Epoch 326/500\n",
      "5851/5851 [==============================] - 0s 57us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00326: val_loss improved from 0.15957 to 0.15956, saving model to ./keras.model_an_cnn\n",
      "Epoch 327/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00327: val_loss improved from 0.15956 to 0.15956, saving model to ./keras.model_an_cnn\n",
      "Epoch 328/500\n",
      "5851/5851 [==============================] - 0s 47us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00328: val_loss did not improve from 0.15956\n",
      "Epoch 329/500\n",
      "5851/5851 [==============================] - 0s 61us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00329: val_loss improved from 0.15956 to 0.15955, saving model to ./keras.model_an_cnn\n",
      "Epoch 330/500\n",
      "5851/5851 [==============================] - 0s 42us/step - loss: 0.1579 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00330: val_loss did not improve from 0.15955\n",
      "Epoch 331/500\n",
      "5851/5851 [==============================] - 0s 52us/step - loss: 0.1578 - val_loss: 0.1596\n",
      "\n",
      "Epoch 00331: val_loss did not improve from 0.15955\n",
      "Epoch 332/500\n",
      "5851/5851 [==============================] - 0s 64us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00332: val_loss improved from 0.15955 to 0.15954, saving model to ./keras.model_an_cnn\n",
      "Epoch 333/500\n",
      "5851/5851 [==============================] - 0s 47us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00333: val_loss did not improve from 0.15954\n",
      "Epoch 334/500\n",
      "5851/5851 [==============================] - 0s 65us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00334: val_loss improved from 0.15954 to 0.15952, saving model to ./keras.model_an_cnn\n",
      "Epoch 335/500\n",
      "5851/5851 [==============================] - 0s 55us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00335: val_loss did not improve from 0.15952\n",
      "Epoch 336/500\n",
      "5851/5851 [==============================] - 0s 83us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00336: val_loss improved from 0.15952 to 0.15952, saving model to ./keras.model_an_cnn\n",
      "Epoch 337/500\n",
      "5851/5851 [==============================] - 0s 53us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00337: val_loss improved from 0.15952 to 0.15952, saving model to ./keras.model_an_cnn\n",
      "Epoch 338/500\n",
      "5851/5851 [==============================] - 0s 39us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00338: val_loss improved from 0.15952 to 0.15951, saving model to ./keras.model_an_cnn\n",
      "Epoch 339/500\n",
      "5851/5851 [==============================] - 0s 58us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00339: val_loss did not improve from 0.15951\n",
      "Epoch 340/500\n",
      "5851/5851 [==============================] - 1s 88us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00340: val_loss did not improve from 0.15951\n",
      "Epoch 341/500\n",
      "5851/5851 [==============================] - 1s 97us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00341: val_loss did not improve from 0.15951\n",
      "Epoch 342/500\n",
      "5851/5851 [==============================] - 1s 89us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00342: val_loss did not improve from 0.15951\n",
      "Epoch 343/500\n",
      "5851/5851 [==============================] - 1s 116us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00343: val_loss did not improve from 0.15951\n",
      "Epoch 344/500\n",
      "5851/5851 [==============================] - 0s 57us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00344: val_loss did not improve from 0.15951\n",
      "Epoch 345/500\n",
      "5851/5851 [==============================] - 0s 54us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00345: val_loss improved from 0.15951 to 0.15949, saving model to ./keras.model_an_cnn\n",
      "Epoch 346/500\n",
      "5851/5851 [==============================] - 0s 58us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00346: val_loss did not improve from 0.15949\n",
      "Epoch 347/500\n",
      "5851/5851 [==============================] - 1s 117us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00347: val_loss did not improve from 0.15949\n",
      "Epoch 348/500\n",
      "5851/5851 [==============================] - 1s 133us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00348: val_loss improved from 0.15949 to 0.15948, saving model to ./keras.model_an_cnn\n",
      "Epoch 349/500\n",
      "5851/5851 [==============================] - 0s 50us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00349: val_loss did not improve from 0.15948\n",
      "Epoch 350/500\n",
      "5851/5851 [==============================] - 0s 53us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00350: val_loss improved from 0.15948 to 0.15948, saving model to ./keras.model_an_cnn\n",
      "Epoch 351/500\n",
      "5851/5851 [==============================] - 0s 42us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00351: val_loss improved from 0.15948 to 0.15947, saving model to ./keras.model_an_cnn\n",
      "Epoch 352/500\n",
      "5851/5851 [==============================] - 0s 42us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00352: val_loss improved from 0.15947 to 0.15947, saving model to ./keras.model_an_cnn\n",
      "Epoch 353/500\n",
      "5851/5851 [==============================] - 0s 47us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00353: val_loss did not improve from 0.15947\n",
      "Epoch 354/500\n",
      "5851/5851 [==============================] - 0s 47us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00354: val_loss did not improve from 0.15947\n",
      "Epoch 355/500\n",
      "5851/5851 [==============================] - 0s 74us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00355: val_loss improved from 0.15947 to 0.15946, saving model to ./keras.model_an_cnn\n",
      "Epoch 356/500\n",
      "5851/5851 [==============================] - 0s 39us/step - loss: 0.1577 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00356: val_loss improved from 0.15946 to 0.15945, saving model to ./keras.model_an_cnn\n",
      "Epoch 357/500\n",
      "5851/5851 [==============================] - 0s 49us/step - loss: 0.1578 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00357: val_loss improved from 0.15945 to 0.15945, saving model to ./keras.model_an_cnn\n",
      "Epoch 358/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1577 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00358: val_loss did not improve from 0.15945\n",
      "Epoch 359/500\n",
      "5851/5851 [==============================] - 0s 42us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00359: val_loss improved from 0.15945 to 0.15944, saving model to ./keras.model_an_cnn\n",
      "Epoch 360/500\n",
      "5851/5851 [==============================] - 0s 44us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00360: val_loss did not improve from 0.15944\n",
      "Epoch 361/500\n",
      "5851/5851 [==============================] - 0s 58us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00361: val_loss did not improve from 0.15944\n",
      "Epoch 362/500\n",
      "5851/5851 [==============================] - 1s 124us/step - loss: 0.1577 - val_loss: 0.1595\n",
      "\n",
      "Epoch 00362: val_loss did not improve from 0.15944\n",
      "Epoch 363/500\n",
      "5851/5851 [==============================] - 0s 74us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00363: val_loss improved from 0.15944 to 0.15944, saving model to ./keras.model_an_cnn\n",
      "Epoch 364/500\n",
      "5851/5851 [==============================] - 0s 38us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00364: val_loss improved from 0.15944 to 0.15942, saving model to ./keras.model_an_cnn\n",
      "Epoch 365/500\n",
      "5851/5851 [==============================] - 0s 54us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00365: val_loss did not improve from 0.15942\n",
      "Epoch 366/500\n",
      "5851/5851 [==============================] - 0s 79us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00366: val_loss improved from 0.15942 to 0.15942, saving model to ./keras.model_an_cnn\n",
      "Epoch 367/500\n",
      "5851/5851 [==============================] - 0s 36us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00367: val_loss did not improve from 0.15942\n",
      "Epoch 368/500\n",
      "5851/5851 [==============================] - 0s 45us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00368: val_loss did not improve from 0.15942\n",
      "Epoch 369/500\n",
      "5851/5851 [==============================] - 0s 62us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00369: val_loss did not improve from 0.15942\n",
      "Epoch 370/500\n",
      "5851/5851 [==============================] - 0s 69us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00370: val_loss improved from 0.15942 to 0.15942, saving model to ./keras.model_an_cnn\n",
      "Epoch 371/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00371: val_loss improved from 0.15942 to 0.15941, saving model to ./keras.model_an_cnn\n",
      "Epoch 372/500\n",
      "5851/5851 [==============================] - 0s 47us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00372: val_loss did not improve from 0.15941\n",
      "Epoch 373/500\n",
      "5851/5851 [==============================] - 0s 79us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00373: val_loss improved from 0.15941 to 0.15941, saving model to ./keras.model_an_cnn\n",
      "Epoch 374/500\n",
      "5851/5851 [==============================] - 0s 62us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00374: val_loss improved from 0.15941 to 0.15940, saving model to ./keras.model_an_cnn\n",
      "Epoch 375/500\n",
      "5851/5851 [==============================] - 0s 42us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00375: val_loss improved from 0.15940 to 0.15938, saving model to ./keras.model_an_cnn\n",
      "Epoch 376/500\n",
      "5851/5851 [==============================] - 0s 46us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00376: val_loss improved from 0.15938 to 0.15937, saving model to ./keras.model_an_cnn\n",
      "Epoch 377/500\n",
      "5851/5851 [==============================] - 0s 43us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00377: val_loss did not improve from 0.15937\n",
      "Epoch 378/500\n",
      "5851/5851 [==============================] - 0s 64us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00378: val_loss did not improve from 0.15937\n",
      "Epoch 379/500\n",
      "5851/5851 [==============================] - 1s 111us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00379: val_loss improved from 0.15937 to 0.15937, saving model to ./keras.model_an_cnn\n",
      "Epoch 380/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00380: val_loss did not improve from 0.15937\n",
      "Epoch 381/500\n",
      "5851/5851 [==============================] - 0s 52us/step - loss: 0.1577 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00381: val_loss did not improve from 0.15937\n",
      "Epoch 382/500\n",
      "5851/5851 [==============================] - 0s 72us/step - loss: 0.1576 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00382: val_loss improved from 0.15937 to 0.15936, saving model to ./keras.model_an_cnn\n",
      "Epoch 383/500\n",
      "5851/5851 [==============================] - 0s 66us/step - loss: 0.1576 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00383: val_loss did not improve from 0.15936\n",
      "Epoch 384/500\n",
      "5851/5851 [==============================] - 1s 90us/step - loss: 0.1576 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00384: val_loss improved from 0.15936 to 0.15935, saving model to ./keras.model_an_cnn\n",
      "Epoch 385/500\n",
      "5851/5851 [==============================] - 0s 52us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00385: val_loss improved from 0.15935 to 0.15934, saving model to ./keras.model_an_cnn\n",
      "Epoch 386/500\n",
      "5851/5851 [==============================] - 0s 40us/step - loss: 0.1576 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00386: val_loss did not improve from 0.15934\n",
      "Epoch 387/500\n",
      "5851/5851 [==============================] - 0s 67us/step - loss: 0.1576 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00387: val_loss did not improve from 0.15934\n",
      "Epoch 388/500\n",
      "5851/5851 [==============================] - 0s 69us/step - loss: 0.1576 - val_loss: 0.1594\n",
      "\n",
      "Epoch 00388: val_loss did not improve from 0.15934\n",
      "Epoch 389/500\n",
      "5851/5851 [==============================] - 0s 56us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00389: val_loss did not improve from 0.15934\n",
      "Epoch 390/500\n",
      "5851/5851 [==============================] - 1s 94us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00390: val_loss did not improve from 0.15934\n",
      "Epoch 391/500\n",
      "5851/5851 [==============================] - 0s 76us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00391: val_loss improved from 0.15934 to 0.15932, saving model to ./keras.model_an_cnn\n",
      "Epoch 392/500\n",
      "5851/5851 [==============================] - 0s 40us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00392: val_loss did not improve from 0.15932\n",
      "Epoch 393/500\n",
      "5851/5851 [==============================] - 0s 76us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00393: val_loss improved from 0.15932 to 0.15931, saving model to ./keras.model_an_cnn\n",
      "Epoch 394/500\n",
      "5851/5851 [==============================] - 0s 47us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00394: val_loss did not improve from 0.15931\n",
      "Epoch 395/500\n",
      "5851/5851 [==============================] - 1s 104us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00395: val_loss did not improve from 0.15931\n",
      "Epoch 396/500\n",
      "5851/5851 [==============================] - 0s 81us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00396: val_loss did not improve from 0.15931\n",
      "Epoch 397/500\n",
      "5851/5851 [==============================] - 1s 95us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00397: val_loss did not improve from 0.15931\n",
      "Epoch 398/500\n",
      "5851/5851 [==============================] - 0s 71us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00398: val_loss did not improve from 0.15931\n",
      "Epoch 399/500\n",
      "5851/5851 [==============================] - 0s 81us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00399: val_loss did not improve from 0.15931\n",
      "Epoch 400/500\n",
      "5851/5851 [==============================] - 0s 70us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00400: val_loss improved from 0.15931 to 0.15930, saving model to ./keras.model_an_cnn\n",
      "Epoch 401/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00401: val_loss did not improve from 0.15930\n",
      "Epoch 402/500\n",
      "5851/5851 [==============================] - 0s 67us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00402: val_loss did not improve from 0.15930\n",
      "Epoch 403/500\n",
      "5851/5851 [==============================] - 0s 67us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00403: val_loss improved from 0.15930 to 0.15929, saving model to ./keras.model_an_cnn\n",
      "Epoch 404/500\n",
      "5851/5851 [==============================] - 0s 39us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00404: val_loss improved from 0.15929 to 0.15928, saving model to ./keras.model_an_cnn\n",
      "Epoch 405/500\n",
      "5851/5851 [==============================] - 0s 38us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00405: val_loss did not improve from 0.15928\n",
      "Epoch 406/500\n",
      "5851/5851 [==============================] - 0s 83us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00406: val_loss did not improve from 0.15928\n",
      "Epoch 407/500\n",
      "5851/5851 [==============================] - 0s 78us/step - loss: 0.1576 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00407: val_loss did not improve from 0.15928\n",
      "Epoch 408/500\n",
      "5851/5851 [==============================] - 0s 56us/step - loss: 0.1575 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00408: val_loss did not improve from 0.15928\n",
      "Epoch 409/500\n",
      "5851/5851 [==============================] - 0s 76us/step - loss: 0.1575 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00409: val_loss improved from 0.15928 to 0.15927, saving model to ./keras.model_an_cnn\n",
      "Epoch 410/500\n",
      "5851/5851 [==============================] - 0s 43us/step - loss: 0.1575 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00410: val_loss did not improve from 0.15927\n",
      "Epoch 411/500\n",
      "5851/5851 [==============================] - 1s 93us/step - loss: 0.1575 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00411: val_loss did not improve from 0.15927\n",
      "Epoch 412/500\n",
      "5851/5851 [==============================] - 1s 101us/step - loss: 0.1575 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00412: val_loss improved from 0.15927 to 0.15927, saving model to ./keras.model_an_cnn\n",
      "Epoch 413/500\n",
      "5851/5851 [==============================] - 0s 49us/step - loss: 0.1575 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00413: val_loss improved from 0.15927 to 0.15926, saving model to ./keras.model_an_cnn\n",
      "Epoch 414/500\n",
      "5851/5851 [==============================] - 0s 36us/step - loss: 0.1575 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00414: val_loss did not improve from 0.15926\n",
      "Epoch 415/500\n",
      "5851/5851 [==============================] - 0s 42us/step - loss: 0.1575 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00415: val_loss did not improve from 0.15926\n",
      "Epoch 416/500\n",
      "5851/5851 [==============================] - 0s 66us/step - loss: 0.1575 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00416: val_loss did not improve from 0.15926\n",
      "Epoch 417/500\n",
      "5851/5851 [==============================] - 0s 46us/step - loss: 0.1575 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00417: val_loss improved from 0.15926 to 0.15925, saving model to ./keras.model_an_cnn\n",
      "Epoch 418/500\n",
      "5851/5851 [==============================] - 0s 44us/step - loss: 0.1575 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00418: val_loss improved from 0.15925 to 0.15925, saving model to ./keras.model_an_cnn\n",
      "Epoch 419/500\n",
      "5851/5851 [==============================] - 0s 50us/step - loss: 0.1575 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00419: val_loss did not improve from 0.15925\n",
      "Epoch 420/500\n",
      "5851/5851 [==============================] - 0s 70us/step - loss: 0.1575 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00420: val_loss did not improve from 0.15925\n",
      "Epoch 421/500\n",
      "5851/5851 [==============================] - 1s 98us/step - loss: 0.1575 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00421: val_loss improved from 0.15925 to 0.15924, saving model to ./keras.model_an_cnn\n",
      "Epoch 422/500\n",
      "5851/5851 [==============================] - 0s 49us/step - loss: 0.1575 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00422: val_loss did not improve from 0.15924\n",
      "Epoch 423/500\n",
      "5851/5851 [==============================] - 1s 89us/step - loss: 0.1575 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00423: val_loss improved from 0.15924 to 0.15922, saving model to ./keras.model_an_cnn\n",
      "Epoch 424/500\n",
      "5851/5851 [==============================] - 0s 52us/step - loss: 0.1575 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00424: val_loss did not improve from 0.15922\n",
      "Epoch 425/500\n",
      "5851/5851 [==============================] - 0s 80us/step - loss: 0.1575 - val_loss: 0.1593\n",
      "\n",
      "Epoch 00425: val_loss did not improve from 0.15922\n",
      "Epoch 426/500\n",
      "5851/5851 [==============================] - 1s 109us/step - loss: 0.1575 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00426: val_loss improved from 0.15922 to 0.15921, saving model to ./keras.model_an_cnn\n",
      "Epoch 427/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1575 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00427: val_loss did not improve from 0.15921\n",
      "Epoch 428/500\n",
      "5851/5851 [==============================] - 0s 42us/step - loss: 0.1575 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00428: val_loss did not improve from 0.15921\n",
      "Epoch 429/500\n",
      "5851/5851 [==============================] - 1s 103us/step - loss: 0.1575 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00429: val_loss improved from 0.15921 to 0.15921, saving model to ./keras.model_an_cnn\n",
      "Epoch 430/500\n",
      "5851/5851 [==============================] - 0s 46us/step - loss: 0.1575 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00430: val_loss improved from 0.15921 to 0.15920, saving model to ./keras.model_an_cnn\n",
      "Epoch 431/500\n",
      "5851/5851 [==============================] - 0s 38us/step - loss: 0.1575 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00431: val_loss did not improve from 0.15920\n",
      "Epoch 432/500\n",
      "5851/5851 [==============================] - 1s 107us/step - loss: 0.1575 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00432: val_loss improved from 0.15920 to 0.15918, saving model to ./keras.model_an_cnn\n",
      "Epoch 433/500\n",
      "5851/5851 [==============================] - 0s 53us/step - loss: 0.1575 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00433: val_loss did not improve from 0.15918\n",
      "Epoch 434/500\n",
      "5851/5851 [==============================] - 1s 97us/step - loss: 0.1574 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00434: val_loss did not improve from 0.15918\n",
      "Epoch 435/500\n",
      "5851/5851 [==============================] - 0s 54us/step - loss: 0.1574 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00435: val_loss did not improve from 0.15918\n",
      "Epoch 436/500\n",
      "5851/5851 [==============================] - 0s 43us/step - loss: 0.1574 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00436: val_loss did not improve from 0.15918\n",
      "Epoch 437/500\n",
      "5851/5851 [==============================] - 1s 98us/step - loss: 0.1574 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00437: val_loss did not improve from 0.15918\n",
      "Epoch 438/500\n",
      "5851/5851 [==============================] - 1s 91us/step - loss: 0.1574 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00438: val_loss improved from 0.15918 to 0.15918, saving model to ./keras.model_an_cnn\n",
      "Epoch 439/500\n",
      "5851/5851 [==============================] - 0s 56us/step - loss: 0.1574 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00439: val_loss did not improve from 0.15918\n",
      "Epoch 440/500\n",
      "5851/5851 [==============================] - 0s 78us/step - loss: 0.1574 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00440: val_loss did not improve from 0.15918\n",
      "Epoch 441/500\n",
      "5851/5851 [==============================] - 0s 38us/step - loss: 0.1574 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00441: val_loss improved from 0.15918 to 0.15916, saving model to ./keras.model_an_cnn\n",
      "Epoch 442/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1574 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00442: val_loss improved from 0.15916 to 0.15916, saving model to ./keras.model_an_cnn\n",
      "Epoch 443/500\n",
      "5851/5851 [==============================] - 0s 40us/step - loss: 0.1574 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00443: val_loss improved from 0.15916 to 0.15915, saving model to ./keras.model_an_cnn\n",
      "Epoch 444/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1574 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00444: val_loss did not improve from 0.15915\n",
      "Epoch 445/500\n",
      "5851/5851 [==============================] - 0s 45us/step - loss: 0.1574 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00445: val_loss did not improve from 0.15915\n",
      "Epoch 446/500\n",
      "5851/5851 [==============================] - 0s 63us/step - loss: 0.1574 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00446: val_loss did not improve from 0.15915\n",
      "Epoch 447/500\n",
      "5851/5851 [==============================] - 0s 69us/step - loss: 0.1574 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00447: val_loss improved from 0.15915 to 0.15914, saving model to ./keras.model_an_cnn\n",
      "Epoch 448/500\n",
      "5851/5851 [==============================] - 0s 48us/step - loss: 0.1574 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00448: val_loss improved from 0.15914 to 0.15913, saving model to ./keras.model_an_cnn\n",
      "Epoch 449/500\n",
      "5851/5851 [==============================] - 0s 59us/step - loss: 0.1574 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00449: val_loss did not improve from 0.15913\n",
      "Epoch 450/500\n",
      "5851/5851 [==============================] - 1s 102us/step - loss: 0.1574 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00450: val_loss did not improve from 0.15913\n",
      "Epoch 451/500\n",
      "5851/5851 [==============================] - 1s 100us/step - loss: 0.1574 - val_loss: 0.1592\n",
      "\n",
      "Epoch 00451: val_loss did not improve from 0.15913\n",
      "Epoch 452/500\n",
      "5851/5851 [==============================] - 0s 56us/step - loss: 0.1574 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00452: val_loss did not improve from 0.15913\n",
      "Epoch 453/500\n",
      "5851/5851 [==============================] - 0s 63us/step - loss: 0.1574 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00453: val_loss improved from 0.15913 to 0.15913, saving model to ./keras.model_an_cnn\n",
      "Epoch 454/500\n",
      "5851/5851 [==============================] - 0s 66us/step - loss: 0.1574 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00454: val_loss improved from 0.15913 to 0.15911, saving model to ./keras.model_an_cnn\n",
      "Epoch 455/500\n",
      "5851/5851 [==============================] - 0s 53us/step - loss: 0.1574 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00455: val_loss did not improve from 0.15911\n",
      "Epoch 456/500\n",
      "5851/5851 [==============================] - 0s 79us/step - loss: 0.1574 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00456: val_loss improved from 0.15911 to 0.15910, saving model to ./keras.model_an_cnn\n",
      "Epoch 457/500\n",
      "5851/5851 [==============================] - 0s 66us/step - loss: 0.1574 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00457: val_loss did not improve from 0.15910\n",
      "Epoch 458/500\n",
      "5851/5851 [==============================] - 1s 100us/step - loss: 0.1574 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00458: val_loss did not improve from 0.15910\n",
      "Epoch 459/500\n",
      "5851/5851 [==============================] - 0s 66us/step - loss: 0.1574 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00459: val_loss did not improve from 0.15910\n",
      "Epoch 460/500\n",
      "5851/5851 [==============================] - 1s 101us/step - loss: 0.1574 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00460: val_loss did not improve from 0.15910\n",
      "Epoch 461/500\n",
      "5851/5851 [==============================] - 0s 84us/step - loss: 0.1574 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00461: val_loss improved from 0.15910 to 0.15909, saving model to ./keras.model_an_cnn\n",
      "Epoch 462/500\n",
      "5851/5851 [==============================] - 0s 37us/step - loss: 0.1573 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00462: val_loss did not improve from 0.15909\n",
      "Epoch 463/500\n",
      "5851/5851 [==============================] - 0s 57us/step - loss: 0.1573 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00463: val_loss improved from 0.15909 to 0.15908, saving model to ./keras.model_an_cnn\n",
      "Epoch 464/500\n",
      "5851/5851 [==============================] - 0s 50us/step - loss: 0.1573 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00464: val_loss did not improve from 0.15908\n",
      "Epoch 465/500\n",
      "5851/5851 [==============================] - 1s 112us/step - loss: 0.1573 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00465: val_loss did not improve from 0.15908\n",
      "Epoch 466/500\n",
      "5851/5851 [==============================] - 1s 91us/step - loss: 0.1573 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00466: val_loss did not improve from 0.15908\n",
      "Epoch 467/500\n",
      "5851/5851 [==============================] - 0s 82us/step - loss: 0.1573 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00467: val_loss improved from 0.15908 to 0.15908, saving model to ./keras.model_an_cnn\n",
      "Epoch 468/500\n",
      "5851/5851 [==============================] - 0s 39us/step - loss: 0.1573 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00468: val_loss did not improve from 0.15908\n",
      "Epoch 469/500\n",
      "5851/5851 [==============================] - 1s 103us/step - loss: 0.1573 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00469: val_loss did not improve from 0.15908\n",
      "Epoch 470/500\n",
      "5851/5851 [==============================] - 1s 101us/step - loss: 0.1573 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00470: val_loss improved from 0.15908 to 0.15905, saving model to ./keras.model_an_cnn\n",
      "Epoch 471/500\n",
      "5851/5851 [==============================] - 0s 44us/step - loss: 0.1573 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00471: val_loss did not improve from 0.15905\n",
      "Epoch 472/500\n",
      "5851/5851 [==============================] - 1s 98us/step - loss: 0.1573 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00472: val_loss did not improve from 0.15905\n",
      "Epoch 473/500\n",
      "5851/5851 [==============================] - 0s 83us/step - loss: 0.1573 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00473: val_loss did not improve from 0.15905\n",
      "Epoch 474/500\n",
      "5851/5851 [==============================] - 0s 84us/step - loss: 0.1573 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00474: val_loss did not improve from 0.15905\n",
      "Epoch 475/500\n",
      "5851/5851 [==============================] - 1s 98us/step - loss: 0.1573 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00475: val_loss did not improve from 0.15905\n",
      "Epoch 476/500\n",
      "5851/5851 [==============================] - 1s 102us/step - loss: 0.1573 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00476: val_loss improved from 0.15905 to 0.15904, saving model to ./keras.model_an_cnn\n",
      "Epoch 477/500\n",
      "5851/5851 [==============================] - 0s 73us/step - loss: 0.1573 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00477: val_loss did not improve from 0.15904\n",
      "Epoch 478/500\n",
      "5851/5851 [==============================] - 1s 89us/step - loss: 0.1573 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00478: val_loss did not improve from 0.15904\n",
      "Epoch 479/500\n",
      "5851/5851 [==============================] - 1s 123us/step - loss: 0.1573 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00479: val_loss improved from 0.15904 to 0.15904, saving model to ./keras.model_an_cnn\n",
      "Epoch 480/500\n",
      "5851/5851 [==============================] - 0s 52us/step - loss: 0.1573 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00480: val_loss did not improve from 0.15904\n",
      "Epoch 481/500\n",
      "5851/5851 [==============================] - 1s 87us/step - loss: 0.1573 - val_loss: 0.1591\n",
      "\n",
      "Epoch 00481: val_loss did not improve from 0.15904\n",
      "Epoch 482/500\n",
      "5851/5851 [==============================] - 1s 95us/step - loss: 0.1573 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00482: val_loss did not improve from 0.15904\n",
      "Epoch 483/500\n",
      "5851/5851 [==============================] - 1s 123us/step - loss: 0.1573 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00483: val_loss improved from 0.15904 to 0.15903, saving model to ./keras.model_an_cnn\n",
      "Epoch 484/500\n",
      "5851/5851 [==============================] - 0s 59us/step - loss: 0.1573 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00484: val_loss did not improve from 0.15903\n",
      "Epoch 485/500\n",
      "5851/5851 [==============================] - 0s 66us/step - loss: 0.1573 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00485: val_loss did not improve from 0.15903\n",
      "Epoch 486/500\n",
      "5851/5851 [==============================] - 0s 84us/step - loss: 0.1573 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00486: val_loss improved from 0.15903 to 0.15902, saving model to ./keras.model_an_cnn\n",
      "Epoch 487/500\n",
      "5851/5851 [==============================] - 0s 41us/step - loss: 0.1573 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00487: val_loss improved from 0.15902 to 0.15901, saving model to ./keras.model_an_cnn\n",
      "Epoch 488/500\n",
      "5851/5851 [==============================] - 0s 60us/step - loss: 0.1573 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00488: val_loss improved from 0.15901 to 0.15900, saving model to ./keras.model_an_cnn\n",
      "Epoch 489/500\n",
      "5851/5851 [==============================] - 0s 40us/step - loss: 0.1573 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00489: val_loss did not improve from 0.15900\n",
      "Epoch 490/500\n",
      "5851/5851 [==============================] - 0s 52us/step - loss: 0.1573 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00490: val_loss did not improve from 0.15900\n",
      "Epoch 491/500\n",
      "5851/5851 [==============================] - 1s 141us/step - loss: 0.1572 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00491: val_loss did not improve from 0.15900\n",
      "Epoch 492/500\n",
      "5851/5851 [==============================] - 1s 124us/step - loss: 0.1572 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00492: val_loss did not improve from 0.15900\n",
      "Epoch 493/500\n",
      "5851/5851 [==============================] - 1s 118us/step - loss: 0.1572 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00493: val_loss improved from 0.15900 to 0.15898, saving model to ./keras.model_an_cnn\n",
      "Epoch 494/500\n",
      "5851/5851 [==============================] - 0s 42us/step - loss: 0.1572 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00494: val_loss did not improve from 0.15898\n",
      "Epoch 495/500\n",
      "5851/5851 [==============================] - 0s 70us/step - loss: 0.1572 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00495: val_loss did not improve from 0.15898\n",
      "Epoch 496/500\n",
      "5851/5851 [==============================] - 1s 99us/step - loss: 0.1572 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00496: val_loss did not improve from 0.15898\n",
      "Epoch 497/500\n",
      "5851/5851 [==============================] - 1s 103us/step - loss: 0.1572 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00497: val_loss did not improve from 0.15898\n",
      "Epoch 498/500\n",
      "5851/5851 [==============================] - 0s 80us/step - loss: 0.1572 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00498: val_loss did not improve from 0.15898\n",
      "Epoch 499/500\n",
      "5851/5851 [==============================] - 1s 134us/step - loss: 0.1572 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00499: val_loss did not improve from 0.15898\n",
      "Epoch 500/500\n",
      "5851/5851 [==============================] - 1s 127us/step - loss: 0.1572 - val_loss: 0.1590\n",
      "\n",
      "Epoch 00500: val_loss did not improve from 0.15898\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7fbb080c2fd0>"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "autoencoder2.fit(X2, X2,\n",
    "                epochs=500,\n",
    "                batch_size=128,\n",
    "                shuffle=True,\n",
    "                callbacks=[early_stopping2, model_checkpoint2, reduce_lr2],\n",
    "                validation_data=(X2_t, X2_t))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras.models import load_model\n",
    "autoencoder2 = load_model(\"./keras.model_an_cnn\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "out1=[autoencoder2.evaluate(X1[idx:idx+1,:],X1[idx:idx+1,:],verbose=0) for idx in range(int(X1.shape[0]/10))]\n",
    "out2=[autoencoder2.evaluate(X2[idx:idx+1,:],X2[idx:idx+1,:],verbose=0) for idx in range(int(X2.shape[0]/10))]\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from matplotlib import pyplot as plt\n",
    "plt.hist(out1,alpha = 0.5)\n",
    "plt.hist(out2,alpha = 0.5)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([ 52., 244., 303., 191., 104.,  46.,  16.,  12.,   5.,   1.]),\n",
       " array([0.08686972, 0.11206098, 0.13725224, 0.1624435 , 0.18763477,\n",
       "        0.21282603, 0.23801729, 0.26320855, 0.28839982, 0.31359108,\n",
       "        0.33878234]),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "out1_t=[autoencoder2.evaluate(X1_t[idx:idx+1,:],X1_t[idx:idx+1,:],verbose=0) for idx in range(int(X1_t.shape[0]))]\n",
    "out2_t=[autoencoder2.evaluate(X2_t[idx:idx+1,:],X2_t[idx:idx+1,:],verbose=0) for idx in range(int(X2_t.shape[0]))]\n",
    "plt.hist(out1_t,alpha = 0.5)\n",
    "plt.hist(out2_t,alpha = 0.5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "idx_false=[idx for idx in range(int(X1.shape[0])) if autoencoder2.evaluate(X1[idx:idx+1,:],X1[idx:idx+1,:],verbose=0) <0.12]\n",
    "idx_true=[idx for idx in range(int(X2.shape[0])) if autoencoder2.evaluate(X2[idx:idx+1,:],X2[idx:idx+1,:],verbose=0) >0.26]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "13"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(idx_false)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "106"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "len(idx_true)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Números buenos tomados como anomalías\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x288 with 20 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Números malos tomados como normales\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x288 with 20 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "print(\"Números buenos tomados como anomalías\")\n",
    "X_false=X2[idx_true]\n",
    "decoded_imgs = autoencoder2.predict(X_false)\n",
    "def bare(ax):\n",
    "    ax.get_xaxis().set_visible(False)\n",
    "    ax.get_yaxis().set_visible(False)\n",
    "    return ax\n",
    "\n",
    "n = 10\n",
    "plt.figure(figsize=(20,4))\n",
    "for i in range(n):\n",
    "    ax = plt.subplot(2, n, i+1)\n",
    "    plt.imshow(X_false[i].reshape(28,28))\n",
    "    plt.gray()\n",
    "    bare(ax)\n",
    "    ax = plt.subplot(2, n, i+1+n)\n",
    "    plt.imshow(decoded_imgs[i].reshape(28,28))\n",
    "    plt.hot()\n",
    "    bare(ax)\n",
    "plt.show()\n",
    "print(\"Números malos tomados como normales\")\n",
    "X_false=X1[idx_false]\n",
    "decoded_imgs = autoencoder2.predict(X_false)\n",
    "def bare(ax):\n",
    "    ax.get_xaxis().set_visible(False)\n",
    "    ax.get_yaxis().set_visible(False)\n",
    "    return ax\n",
    "\n",
    "n = 10\n",
    "plt.figure(figsize=(20,4))\n",
    "for i in range(n):\n",
    "    ax = plt.subplot(2, n, i+1)\n",
    "    plt.imshow(X_false[i].reshape(28,28))\n",
    "    plt.gray()\n",
    "    bare(ax)\n",
    "    ax = plt.subplot(2, n, i+1+n)\n",
    "    plt.imshow(decoded_imgs[i].reshape(28,28))\n",
    "    plt.hot()\n",
    "    bare(ax)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Detección de anomalías usando U-NET como autoencoder"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "from random import randint\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "plt.style.use('seaborn-white')\n",
    "import seaborn as sns\n",
    "sns.set_style(\"white\")\n",
    "\n",
    "from sklearn.model_selection import train_test_split\n",
    "\n",
    "from skimage.transform import resize\n",
    "\n",
    "from keras.preprocessing.image import load_img\n",
    "from keras import Model\n",
    "from keras.callbacks import EarlyStopping, ModelCheckpoint, ReduceLROnPlateau\n",
    "from keras.models import load_model\n",
    "from keras.optimizers import Adam\n",
    "from keras.utils.vis_utils import plot_model\n",
    "from keras.preprocessing.image import ImageDataGenerator\n",
    "from keras.layers import Input, Conv2D, Conv2DTranspose, MaxPooling2D, concatenate, Dropout,BatchNormalization\n",
    "from keras.layers import Conv2D, Concatenate, MaxPooling2D\n",
    "from keras.layers import UpSampling2D, Dropout, BatchNormalization\n",
    "from keras.initializers import VarianceScaling \n",
    "init=VarianceScaling(scale=2.0, mode='fan_in', distribution='normal', seed=None) #tomado del paper de unet\n",
    "early_stopping3 = EarlyStopping(patience=10, verbose=1)\n",
    "model_checkpoint3 = ModelCheckpoint(\"./keras.model_an_unet\", save_best_only=True, verbose=1)\n",
    "reduce_lr3 = ReduceLROnPlateau(factor=0.1, patience=5, min_lr=0.00001, verbose=1)\n",
    "def conv_block(m, dim, acti, bn, res, do=0):\n",
    "\tn = Conv2D(dim, 3, activation=acti, padding='same',kernel_initializer=init, bias_initializer='zeros')(m)\n",
    "\tn = BatchNormalization()(n) if bn else n\n",
    "\tn = Dropout(do)(n) if do else n\n",
    "\tn = Conv2D(dim, 3, activation=acti, padding='same',kernel_initializer=init, bias_initializer='zeros')(n)\n",
    "\tn = BatchNormalization()(n) if bn else n\n",
    "\treturn Concatenate()([m, n]) if res else n\n",
    "\n",
    "def level_block(m, dim, depth, inc, acti, do, bn, mp, up, res):\n",
    "\tif depth > 0:\n",
    "\t\tn = conv_block(m, dim, acti, bn, res)\n",
    "\t\tm = MaxPooling2D()(n) if mp else Conv2D(dim, 3, strides=2, padding='same')(n)\n",
    "\t\tm = level_block(m, int(inc*dim), depth-1, inc, acti, do, bn, mp, up, res)\n",
    "\t\tif up:\n",
    "\t\t\tm = UpSampling2D()(m)\n",
    "\t\t\tm = Conv2D(dim, 2, activation=acti, padding='same')(m)\n",
    "\t\telse:\n",
    "\t\t\tm = Conv2DTranspose(dim, 3, strides=2, activation=acti, padding='same')(m)\n",
    "\t\tn = Concatenate()([n, m])\n",
    "\t\tm = conv_block(n, dim, acti, bn, res)\n",
    "\telse:\n",
    "\t\tm = conv_block(m, dim, acti, bn, res, do)\n",
    "\treturn m\n",
    "\n",
    "def UNet(img_shape, out_ch=1, start_ch=64, depth=4, inc_rate=2., activation='relu', \n",
    "\t\t dropout=0.5, batchnorm=False, maxpool=True, upconv=True, residual=False):\n",
    "\ti = Input(shape=img_shape)\n",
    "\to = level_block(i, start_ch, depth, inc_rate, activation, dropout, batchnorm, maxpool, upconv, residual)\n",
    "\to = Conv2D(out_ch, 1, activation='sigmoid',kernel_initializer=init, bias_initializer='zeros')(o)\n",
    "\treturn Model(inputs=i, outputs=o)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "from keras.datasets import mnist\n",
    "(Xa, ya), (Xt, yt) = mnist.load_data()\n",
    "Xa = np.reshape(Xa, (len(Xa), 28, 28, 1))  # adapt this if using `channels_first` image data format\n",
    "Xt = np.reshape(Xt, (len(Xt), 28, 28, 1))  # adapt this if using `channels_first` image data format\n",
    "mask=(ya==8)\n",
    "y2=ya[mask]\n",
    "X2=Xa[mask,:,:,:]\n",
    "mask=(ya==7)\n",
    "y1=ya[mask]\n",
    "X1=Xa[mask,:,:,:]\n",
    "mask=(yt==8)\n",
    "y2_t=yt[mask]\n",
    "X2_t=Xt[mask,:,:,:]\n",
    "mask=(yt==7)\n",
    "y1_t=yt[mask]\n",
    "X1_t=Xt[mask,:,:,:]\n",
    "X1 = (X1.astype('float32')/255.0)\n",
    "X2 = (X2.astype('float32')/255.0)\n",
    "X1_t = (X1_t.astype('float32')/255.0)\n",
    "X2_t = (X2_t.astype('float32')/255.0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "autoencoder3 = UNet((28,28,1),start_ch=32,depth=2,batchnorm=True)\n",
    "autoencoder3.compile(loss=\"binary_crossentropy\", optimizer=\"adam\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 5851 samples, validate on 974 samples\n",
      "Epoch 1/500\n",
      "5851/5851 [==============================] - 4s 763us/step - loss: 0.1640 - val_loss: 0.0909\n",
      "\n",
      "Epoch 00001: val_loss improved from inf to 0.09095, saving model to ./keras.model_an_unet\n",
      "Epoch 2/500\n",
      "5851/5851 [==============================] - 2s 303us/step - loss: 0.0872 - val_loss: 0.0788\n",
      "\n",
      "Epoch 00002: val_loss improved from 0.09095 to 0.07880, saving model to ./keras.model_an_unet\n",
      "Epoch 3/500\n",
      "5851/5851 [==============================] - 2s 314us/step - loss: 0.0792 - val_loss: 0.0753\n",
      "\n",
      "Epoch 00003: val_loss improved from 0.07880 to 0.07526, saving model to ./keras.model_an_unet\n",
      "Epoch 4/500\n",
      "5851/5851 [==============================] - 2s 303us/step - loss: 0.0762 - val_loss: 0.0753\n",
      "\n",
      "Epoch 00004: val_loss did not improve from 0.07526\n",
      "Epoch 5/500\n",
      "5851/5851 [==============================] - 2s 330us/step - loss: 0.0738 - val_loss: 0.0725\n",
      "\n",
      "Epoch 00005: val_loss improved from 0.07526 to 0.07246, saving model to ./keras.model_an_unet\n",
      "Epoch 6/500\n",
      "5851/5851 [==============================] - 2s 328us/step - loss: 0.0720 - val_loss: 0.0822\n",
      "\n",
      "Epoch 00006: val_loss did not improve from 0.07246\n",
      "Epoch 7/500\n",
      "5851/5851 [==============================] - 2s 347us/step - loss: 0.0710 - val_loss: 0.0696\n",
      "\n",
      "Epoch 00007: val_loss improved from 0.07246 to 0.06959, saving model to ./keras.model_an_unet\n",
      "Epoch 8/500\n",
      "5851/5851 [==============================] - 2s 318us/step - loss: 0.0704 - val_loss: 0.0700\n",
      "\n",
      "Epoch 00008: val_loss did not improve from 0.06959\n",
      "Epoch 9/500\n",
      "5851/5851 [==============================] - 2s 350us/step - loss: 0.0699 - val_loss: 0.0698\n",
      "\n",
      "Epoch 00009: val_loss did not improve from 0.06959\n",
      "Epoch 10/500\n",
      "5851/5851 [==============================] - 2s 312us/step - loss: 0.0697 - val_loss: 0.0706\n",
      "\n",
      "Epoch 00010: val_loss did not improve from 0.06959\n",
      "Epoch 11/500\n",
      "5851/5851 [==============================] - 2s 328us/step - loss: 0.0695 - val_loss: 0.0693\n",
      "\n",
      "Epoch 00011: val_loss improved from 0.06959 to 0.06931, saving model to ./keras.model_an_unet\n",
      "Epoch 12/500\n",
      "5851/5851 [==============================] - 2s 309us/step - loss: 0.0693 - val_loss: 0.0691\n",
      "\n",
      "Epoch 00012: val_loss improved from 0.06931 to 0.06906, saving model to ./keras.model_an_unet\n",
      "Epoch 13/500\n",
      "5851/5851 [==============================] - 2s 313us/step - loss: 0.0693 - val_loss: 0.0689\n",
      "\n",
      "Epoch 00013: val_loss improved from 0.06906 to 0.06895, saving model to ./keras.model_an_unet\n",
      "Epoch 14/500\n",
      "5851/5851 [==============================] - 2s 319us/step - loss: 0.0695 - val_loss: 0.4139\n",
      "\n",
      "Epoch 00014: val_loss did not improve from 0.06895\n",
      "Epoch 15/500\n",
      "5851/5851 [==============================] - 2s 364us/step - loss: 0.0693 - val_loss: 0.0689\n",
      "\n",
      "Epoch 00015: val_loss improved from 0.06895 to 0.06892, saving model to ./keras.model_an_unet\n",
      "Epoch 16/500\n",
      "5851/5851 [==============================] - 2s 306us/step - loss: 0.0690 - val_loss: 0.0690\n",
      "\n",
      "Epoch 00016: val_loss did not improve from 0.06892\n",
      "Epoch 17/500\n",
      "5851/5851 [==============================] - 2s 326us/step - loss: 0.0690 - val_loss: 0.0688\n",
      "\n",
      "Epoch 00017: val_loss improved from 0.06892 to 0.06876, saving model to ./keras.model_an_unet\n",
      "Epoch 18/500\n",
      "5851/5851 [==============================] - 2s 331us/step - loss: 0.0690 - val_loss: 0.0688\n",
      "\n",
      "Epoch 00018: val_loss did not improve from 0.06876\n",
      "Epoch 19/500\n",
      "5851/5851 [==============================] - 2s 332us/step - loss: 0.0689 - val_loss: 0.0689\n",
      "\n",
      "Epoch 00019: val_loss did not improve from 0.06876\n",
      "Epoch 20/500\n",
      "5851/5851 [==============================] - 2s 354us/step - loss: 0.0688 - val_loss: 0.0687\n",
      "\n",
      "Epoch 00020: val_loss improved from 0.06876 to 0.06867, saving model to ./keras.model_an_unet\n",
      "Epoch 21/500\n",
      "5851/5851 [==============================] - 2s 321us/step - loss: 0.0688 - val_loss: 0.0688\n",
      "\n",
      "Epoch 00021: val_loss did not improve from 0.06867\n",
      "Epoch 22/500\n",
      "5851/5851 [==============================] - 2s 361us/step - loss: 0.0688 - val_loss: 0.0687\n",
      "\n",
      "Epoch 00022: val_loss improved from 0.06867 to 0.06865, saving model to ./keras.model_an_unet\n",
      "Epoch 23/500\n",
      "5851/5851 [==============================] - 2s 358us/step - loss: 0.0688 - val_loss: 0.0686\n",
      "\n",
      "Epoch 00023: val_loss improved from 0.06865 to 0.06863, saving model to ./keras.model_an_unet\n",
      "Epoch 24/500\n",
      "5851/5851 [==============================] - 2s 295us/step - loss: 0.0688 - val_loss: 0.0687\n",
      "\n",
      "Epoch 00024: val_loss did not improve from 0.06863\n",
      "Epoch 25/500\n",
      "5851/5851 [==============================] - 2s 327us/step - loss: 0.0687 - val_loss: 0.0686\n",
      "\n",
      "Epoch 00025: val_loss improved from 0.06863 to 0.06857, saving model to ./keras.model_an_unet\n",
      "Epoch 26/500\n",
      "5851/5851 [==============================] - 2s 338us/step - loss: 0.0687 - val_loss: 0.0686\n",
      "\n",
      "Epoch 00026: val_loss improved from 0.06857 to 0.06857, saving model to ./keras.model_an_unet\n",
      "Epoch 27/500\n",
      "5851/5851 [==============================] - 2s 296us/step - loss: 0.0686 - val_loss: 0.0686\n",
      "\n",
      "Epoch 00027: val_loss did not improve from 0.06857\n",
      "\n",
      "Epoch 00027: ReduceLROnPlateau reducing learning rate to 0.00010000000474974513.\n",
      "Epoch 28/500\n",
      "5851/5851 [==============================] - 2s 326us/step - loss: 0.0686 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00028: val_loss improved from 0.06857 to 0.06850, saving model to ./keras.model_an_unet\n",
      "Epoch 29/500\n",
      "5851/5851 [==============================] - 2s 299us/step - loss: 0.0686 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00029: val_loss did not improve from 0.06850\n",
      "Epoch 30/500\n",
      "5851/5851 [==============================] - 2s 320us/step - loss: 0.0686 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00030: val_loss did not improve from 0.06850\n",
      "Epoch 31/500\n",
      "5851/5851 [==============================] - 2s 355us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00031: val_loss did not improve from 0.06850\n",
      "Epoch 32/500\n",
      "5851/5851 [==============================] - 2s 292us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00032: val_loss did not improve from 0.06850\n",
      "Epoch 33/500\n",
      "5851/5851 [==============================] - 2s 326us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00033: val_loss did not improve from 0.06850\n",
      "\n",
      "Epoch 00033: ReduceLROnPlateau reducing learning rate to 1.0000000474974514e-05.\n",
      "Epoch 34/500\n",
      "5851/5851 [==============================] - 2s 391us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00034: val_loss improved from 0.06850 to 0.06849, saving model to ./keras.model_an_unet\n",
      "Epoch 35/500\n",
      "5851/5851 [==============================] - 2s 318us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00035: val_loss improved from 0.06849 to 0.06849, saving model to ./keras.model_an_unet\n",
      "Epoch 36/500\n",
      "5851/5851 [==============================] - 2s 373us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00036: val_loss did not improve from 0.06849\n",
      "Epoch 37/500\n",
      "5851/5851 [==============================] - 2s 343us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00037: val_loss did not improve from 0.06849\n",
      "Epoch 38/500\n",
      "5851/5851 [==============================] - 2s 323us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00038: val_loss did not improve from 0.06849\n",
      "\n",
      "Epoch 00038: ReduceLROnPlateau reducing learning rate to 1e-05.\n",
      "Epoch 39/500\n",
      "5851/5851 [==============================] - 2s 329us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00039: val_loss did not improve from 0.06849\n",
      "Epoch 40/500\n",
      "5851/5851 [==============================] - 2s 308us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00040: val_loss did not improve from 0.06849\n",
      "Epoch 41/500\n",
      "5851/5851 [==============================] - 2s 390us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00041: val_loss did not improve from 0.06849\n",
      "Epoch 42/500\n",
      "5851/5851 [==============================] - 2s 346us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00042: val_loss improved from 0.06849 to 0.06848, saving model to ./keras.model_an_unet\n",
      "Epoch 43/500\n",
      "5851/5851 [==============================] - 2s 319us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00043: val_loss improved from 0.06848 to 0.06848, saving model to ./keras.model_an_unet\n",
      "Epoch 44/500\n",
      "5851/5851 [==============================] - 2s 367us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00044: val_loss did not improve from 0.06848\n",
      "Epoch 45/500\n",
      "5851/5851 [==============================] - 2s 367us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00045: val_loss did not improve from 0.06848\n",
      "Epoch 46/500\n",
      "5851/5851 [==============================] - 2s 360us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00046: val_loss improved from 0.06848 to 0.06848, saving model to ./keras.model_an_unet\n",
      "Epoch 47/500\n",
      "5851/5851 [==============================] - 2s 313us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00047: val_loss did not improve from 0.06848\n",
      "Epoch 48/500\n",
      "5851/5851 [==============================] - 2s 370us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00048: val_loss did not improve from 0.06848\n",
      "Epoch 49/500\n",
      "5851/5851 [==============================] - 2s 373us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00049: val_loss did not improve from 0.06848\n",
      "Epoch 50/500\n",
      "5851/5851 [==============================] - 2s 335us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00050: val_loss improved from 0.06848 to 0.06848, saving model to ./keras.model_an_unet\n",
      "Epoch 51/500\n",
      "5851/5851 [==============================] - 2s 305us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00051: val_loss improved from 0.06848 to 0.06848, saving model to ./keras.model_an_unet\n",
      "Epoch 52/500\n",
      "5851/5851 [==============================] - 2s 331us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00052: val_loss did not improve from 0.06848\n",
      "Epoch 53/500\n",
      "5851/5851 [==============================] - 2s 333us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00053: val_loss did not improve from 0.06848\n",
      "Epoch 54/500\n",
      "5851/5851 [==============================] - 2s 306us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00054: val_loss improved from 0.06848 to 0.06848, saving model to ./keras.model_an_unet\n",
      "Epoch 55/500\n",
      "5851/5851 [==============================] - 2s 334us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00055: val_loss improved from 0.06848 to 0.06848, saving model to ./keras.model_an_unet\n",
      "Epoch 56/500\n",
      "5851/5851 [==============================] - 2s 326us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00056: val_loss did not improve from 0.06848\n",
      "Epoch 57/500\n",
      "5851/5851 [==============================] - 2s 325us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00057: val_loss did not improve from 0.06848\n",
      "Epoch 58/500\n",
      "5851/5851 [==============================] - 2s 359us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00058: val_loss improved from 0.06848 to 0.06848, saving model to ./keras.model_an_unet\n",
      "Epoch 59/500\n",
      "5851/5851 [==============================] - 2s 316us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00059: val_loss did not improve from 0.06848\n",
      "Epoch 60/500\n",
      "5851/5851 [==============================] - 2s 364us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00060: val_loss did not improve from 0.06848\n",
      "Epoch 61/500\n",
      "5851/5851 [==============================] - 2s 310us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00061: val_loss did not improve from 0.06848\n",
      "Epoch 62/500\n",
      "5851/5851 [==============================] - 2s 331us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00062: val_loss improved from 0.06848 to 0.06847, saving model to ./keras.model_an_unet\n",
      "Epoch 63/500\n",
      "5851/5851 [==============================] - 2s 374us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00063: val_loss did not improve from 0.06847\n",
      "Epoch 64/500\n",
      "5851/5851 [==============================] - 2s 305us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00064: val_loss improved from 0.06847 to 0.06847, saving model to ./keras.model_an_unet\n",
      "Epoch 65/500\n",
      "5851/5851 [==============================] - 2s 302us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00065: val_loss improved from 0.06847 to 0.06847, saving model to ./keras.model_an_unet\n",
      "Epoch 66/500\n",
      "5851/5851 [==============================] - 2s 359us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00066: val_loss did not improve from 0.06847\n",
      "Epoch 67/500\n",
      "5851/5851 [==============================] - 2s 314us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00067: val_loss improved from 0.06847 to 0.06847, saving model to ./keras.model_an_unet\n",
      "Epoch 68/500\n",
      "5851/5851 [==============================] - 2s 304us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00068: val_loss did not improve from 0.06847\n",
      "Epoch 69/500\n",
      "5851/5851 [==============================] - 2s 361us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00069: val_loss did not improve from 0.06847\n",
      "Epoch 70/500\n",
      "5851/5851 [==============================] - 2s 313us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00070: val_loss improved from 0.06847 to 0.06847, saving model to ./keras.model_an_unet\n",
      "Epoch 71/500\n",
      "5851/5851 [==============================] - 2s 341us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00071: val_loss did not improve from 0.06847\n",
      "Epoch 72/500\n",
      "5851/5851 [==============================] - 2s 323us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00072: val_loss improved from 0.06847 to 0.06847, saving model to ./keras.model_an_unet\n",
      "Epoch 73/500\n",
      "5851/5851 [==============================] - 2s 329us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00073: val_loss did not improve from 0.06847\n",
      "Epoch 74/500\n",
      "5851/5851 [==============================] - 2s 340us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00074: val_loss improved from 0.06847 to 0.06847, saving model to ./keras.model_an_unet\n",
      "Epoch 75/500\n",
      "5851/5851 [==============================] - 2s 317us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00075: val_loss improved from 0.06847 to 0.06847, saving model to ./keras.model_an_unet\n",
      "Epoch 76/500\n",
      "5851/5851 [==============================] - 2s 309us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00076: val_loss did not improve from 0.06847\n",
      "Epoch 77/500\n",
      "5851/5851 [==============================] - 2s 339us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00077: val_loss improved from 0.06847 to 0.06847, saving model to ./keras.model_an_unet\n",
      "Epoch 78/500\n",
      "5851/5851 [==============================] - 2s 326us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00078: val_loss did not improve from 0.06847\n",
      "Epoch 79/500\n",
      "5851/5851 [==============================] - 2s 376us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00079: val_loss did not improve from 0.06847\n",
      "Epoch 80/500\n",
      "5851/5851 [==============================] - 2s 337us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00080: val_loss improved from 0.06847 to 0.06847, saving model to ./keras.model_an_unet\n",
      "Epoch 81/500\n",
      "5851/5851 [==============================] - 2s 297us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00081: val_loss improved from 0.06847 to 0.06846, saving model to ./keras.model_an_unet\n",
      "Epoch 82/500\n",
      "5851/5851 [==============================] - 2s 337us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00082: val_loss did not improve from 0.06846\n",
      "Epoch 83/500\n",
      "5851/5851 [==============================] - 2s 356us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00083: val_loss did not improve from 0.06846\n",
      "Epoch 84/500\n",
      "5851/5851 [==============================] - 2s 308us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00084: val_loss improved from 0.06846 to 0.06846, saving model to ./keras.model_an_unet\n",
      "Epoch 85/500\n",
      "5851/5851 [==============================] - 2s 304us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00085: val_loss did not improve from 0.06846\n",
      "Epoch 86/500\n",
      "5851/5851 [==============================] - 2s 326us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00086: val_loss did not improve from 0.06846\n",
      "Epoch 87/500\n",
      "5851/5851 [==============================] - 2s 324us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00087: val_loss did not improve from 0.06846\n",
      "Epoch 88/500\n",
      "5851/5851 [==============================] - 2s 314us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00088: val_loss improved from 0.06846 to 0.06846, saving model to ./keras.model_an_unet\n",
      "Epoch 89/500\n",
      "5851/5851 [==============================] - 2s 330us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00089: val_loss did not improve from 0.06846\n",
      "Epoch 90/500\n",
      "5851/5851 [==============================] - 2s 360us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00090: val_loss did not improve from 0.06846\n",
      "Epoch 91/500\n",
      "5851/5851 [==============================] - 2s 328us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00091: val_loss improved from 0.06846 to 0.06846, saving model to ./keras.model_an_unet\n",
      "Epoch 92/500\n",
      "5851/5851 [==============================] - 2s 357us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00092: val_loss improved from 0.06846 to 0.06846, saving model to ./keras.model_an_unet\n",
      "Epoch 93/500\n",
      "5851/5851 [==============================] - 2s 333us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00093: val_loss improved from 0.06846 to 0.06846, saving model to ./keras.model_an_unet\n",
      "Epoch 94/500\n",
      "5851/5851 [==============================] - 2s 303us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00094: val_loss did not improve from 0.06846\n",
      "Epoch 95/500\n",
      "5851/5851 [==============================] - 2s 321us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00095: val_loss did not improve from 0.06846\n",
      "Epoch 96/500\n",
      "5851/5851 [==============================] - 2s 376us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00096: val_loss improved from 0.06846 to 0.06846, saving model to ./keras.model_an_unet\n",
      "Epoch 97/500\n",
      "5851/5851 [==============================] - 2s 301us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00097: val_loss improved from 0.06846 to 0.06846, saving model to ./keras.model_an_unet\n",
      "Epoch 98/500\n",
      "5851/5851 [==============================] - 2s 325us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00098: val_loss did not improve from 0.06846\n",
      "Epoch 99/500\n",
      "5851/5851 [==============================] - 2s 328us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00099: val_loss did not improve from 0.06846\n",
      "Epoch 100/500\n",
      "5851/5851 [==============================] - 2s 336us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00100: val_loss improved from 0.06846 to 0.06846, saving model to ./keras.model_an_unet\n",
      "Epoch 101/500\n",
      "5851/5851 [==============================] - 2s 334us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00101: val_loss did not improve from 0.06846\n",
      "Epoch 102/500\n",
      "5851/5851 [==============================] - 2s 319us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00102: val_loss improved from 0.06846 to 0.06846, saving model to ./keras.model_an_unet\n",
      "Epoch 103/500\n",
      "5851/5851 [==============================] - 2s 346us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00103: val_loss did not improve from 0.06846\n",
      "Epoch 104/500\n",
      "5851/5851 [==============================] - 2s 344us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00104: val_loss did not improve from 0.06846\n",
      "Epoch 105/500\n",
      "5851/5851 [==============================] - 2s 323us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00105: val_loss improved from 0.06846 to 0.06846, saving model to ./keras.model_an_unet\n",
      "Epoch 106/500\n",
      "5851/5851 [==============================] - 2s 299us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00106: val_loss did not improve from 0.06846\n",
      "Epoch 107/500\n",
      "5851/5851 [==============================] - 2s 424us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00107: val_loss did not improve from 0.06846\n",
      "Epoch 108/500\n",
      "5851/5851 [==============================] - 2s 370us/step - loss: 0.0685 - val_loss: 0.0685\n",
      "\n",
      "Epoch 00108: val_loss improved from 0.06846 to 0.06845, saving model to ./keras.model_an_unet\n",
      "Epoch 109/500\n",
      "5664/5851 [============================>.] - ETA: 0s - loss: 0.0685"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-4-082810fd24fa>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      4\u001b[0m                 \u001b[0mshuffle\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;32mTrue\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      5\u001b[0m                 \u001b[0mcallbacks\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mearly_stopping3\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mmodel_checkpoint3\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mreduce_lr3\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 6\u001b[0;31m                 validation_data=(X2_t, X2_t))\n\u001b[0m",
      "\u001b[0;32m~/.conda/envs/doctorado-gpu/lib/python3.6/site-packages/keras/engine/training.py\u001b[0m in \u001b[0;36mfit\u001b[0;34m(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, **kwargs)\u001b[0m\n\u001b[1;32m   1037\u001b[0m                                         \u001b[0minitial_epoch\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0minitial_epoch\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1038\u001b[0m                                         \u001b[0msteps_per_epoch\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0msteps_per_epoch\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1039\u001b[0;31m                                         validation_steps=validation_steps)\n\u001b[0m\u001b[1;32m   1040\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1041\u001b[0m     def evaluate(self, x=None, y=None,\n",
      "\u001b[0;32m~/.conda/envs/doctorado-gpu/lib/python3.6/site-packages/keras/engine/training_arrays.py\u001b[0m in \u001b[0;36mfit_loop\u001b[0;34m(model, f, ins, out_labels, batch_size, epochs, verbose, callbacks, val_f, val_ins, shuffle, callback_metrics, initial_epoch, steps_per_epoch, validation_steps)\u001b[0m\n\u001b[1;32m    193\u001b[0m                 \u001b[0mbatch_logs\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'batch'\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mbatch_index\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    194\u001b[0m                 \u001b[0mbatch_logs\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'size'\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbatch_ids\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 195\u001b[0;31m                 \u001b[0mcallbacks\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mon_batch_begin\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbatch_index\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mbatch_logs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    196\u001b[0m                 \u001b[0;32mfor\u001b[0m \u001b[0mi\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mindices_for_conversion_to_dense\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    197\u001b[0m                     \u001b[0mins_batch\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mi\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mins_batch\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mi\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtoarray\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.conda/envs/doctorado-gpu/lib/python3.6/site-packages/keras/callbacks.py\u001b[0m in \u001b[0;36mon_batch_begin\u001b[0;34m(self, batch, logs)\u001b[0m\n\u001b[1;32m     91\u001b[0m             \u001b[0mcallback\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mon_batch_begin\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbatch\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlogs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     92\u001b[0m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_delta_ts_batch_begin\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mappend\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtime\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtime\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m-\u001b[0m \u001b[0mt_before_callbacks\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 93\u001b[0;31m         \u001b[0mdelta_t_median\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmedian\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_delta_ts_batch_begin\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     94\u001b[0m         if (self._delta_t_batch > 0. and\n\u001b[1;32m     95\u001b[0m            \u001b[0mdelta_t_median\u001b[0m \u001b[0;34m>\u001b[0m \u001b[0;36m0.95\u001b[0m \u001b[0;34m*\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_delta_t_batch\u001b[0m \u001b[0;32mand\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.conda/envs/doctorado-gpu/lib/python3.6/site-packages/numpy/lib/function_base.py\u001b[0m in \u001b[0;36mmedian\u001b[0;34m(a, axis, out, overwrite_input, keepdims)\u001b[0m\n\u001b[1;32m   3334\u001b[0m     \"\"\"\n\u001b[1;32m   3335\u001b[0m     r, k = _ureduce(a, func=_median, axis=axis, out=out,\n\u001b[0;32m-> 3336\u001b[0;31m                     overwrite_input=overwrite_input)\n\u001b[0m\u001b[1;32m   3337\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mkeepdims\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3338\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0mr\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mreshape\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mk\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.conda/envs/doctorado-gpu/lib/python3.6/site-packages/numpy/lib/function_base.py\u001b[0m in \u001b[0;36m_ureduce\u001b[0;34m(a, func, **kwargs)\u001b[0m\n\u001b[1;32m   3248\u001b[0m         \u001b[0mkeepdim\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m*\u001b[0m \u001b[0ma\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mndim\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3249\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 3250\u001b[0;31m     \u001b[0mr\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mfunc\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0ma\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   3251\u001b[0m     \u001b[0;32mreturn\u001b[0m \u001b[0mr\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mkeepdim\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3252\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.conda/envs/doctorado-gpu/lib/python3.6/site-packages/numpy/lib/function_base.py\u001b[0m in \u001b[0;36m_median\u001b[0;34m(a, axis, out, overwrite_input)\u001b[0m\n\u001b[1;32m   3388\u001b[0m         \u001b[0;31m# warn and return nans like mean would\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3389\u001b[0m         \u001b[0mrout\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmean\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mpart\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mindexer\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maxis\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0maxis\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mout\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mout\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 3390\u001b[0;31m         \u001b[0;32mreturn\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mlib\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mutils\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_median_nancheck\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mpart\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mrout\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maxis\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mout\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   3391\u001b[0m     \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3392\u001b[0m         \u001b[0;31m# if there are no nans\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.conda/envs/doctorado-gpu/lib/python3.6/site-packages/numpy/lib/utils.py\u001b[0m in \u001b[0;36m_median_nancheck\u001b[0;34m(data, result, axis, out)\u001b[0m\n\u001b[1;32m   1139\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mdata\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msize\u001b[0m \u001b[0;34m==\u001b[0m \u001b[0;36m0\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1140\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0mresult\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1141\u001b[0;31m     \u001b[0mdata\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmoveaxis\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdata\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0maxis\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m-\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   1142\u001b[0m     \u001b[0mn\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0misnan\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdata\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m...\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m-\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1143\u001b[0m     \u001b[0;31m# masked NaN values are ok\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m~/.conda/envs/doctorado-gpu/lib/python3.6/site-packages/numpy/core/numeric.py\u001b[0m in \u001b[0;36mmoveaxis\u001b[0;34m(a, source, destination)\u001b[0m\n\u001b[1;32m   1578\u001b[0m         \u001b[0mtranspose\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0ma\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtranspose\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1579\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1580\u001b[0;31m     \u001b[0msource\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnormalize_axis_tuple\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0msource\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0ma\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mndim\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m'source'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   1581\u001b[0m     \u001b[0mdestination\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnormalize_axis_tuple\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdestination\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0ma\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mndim\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m'destination'\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1582\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0msource\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m!=\u001b[0m \u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdestination\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "autoencoder3.fit(X2, X2,\n",
    "                epochs=500,\n",
    "                batch_size=32,\n",
    "                shuffle=True,\n",
    "                callbacks=[early_stopping3, model_checkpoint3, reduce_lr3],\n",
    "                validation_data=(X2_t, X2_t))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([  1.,  11.,  24.,  91., 271., 317., 177.,  63.,  15.,   4.]),\n",
       " array([0.02802917, 0.03566595, 0.04330274, 0.05093952, 0.0585763 ,\n",
       "        0.06621309, 0.07384987, 0.08148665, 0.08912343, 0.09676022,\n",
       "        0.104397  ]),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "out1_t=[autoencoder3.evaluate(X1_t[idx:idx+1,:],X1_t[idx:idx+1,:],verbose=0) for idx in range(int(X1_t.shape[0]))]\n",
    "out2_t=[autoencoder3.evaluate(X2_t[idx:idx+1,:],X2_t[idx:idx+1,:],verbose=0) for idx in range(int(X2_t.shape[0]))]\n",
    "plt.hist(out1_t)\n",
    "plt.hist(out2_t)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[name: \"/device:CPU:0\"\n",
      "device_type: \"CPU\"\n",
      "memory_limit: 268435456\n",
      "locality {\n",
      "}\n",
      "incarnation: 11239420057561585026\n",
      ", name: \"/device:GPU:0\"\n",
      "device_type: \"GPU\"\n",
      "memory_limit: 11910869812\n",
      "locality {\n",
      "  bus_id: 1\n",
      "  links {\n",
      "  }\n",
      "}\n",
      "incarnation: 10712293712288644875\n",
      "physical_device_desc: \"device: 0, name: TITAN Xp, pci bus id: 0000:02:00.0, compute capability: 6.1\"\n",
      "]\n"
     ]
    }
   ],
   "source": [
    "from tensorflow.python.client import device_lib\n",
    "print(device_lib.list_local_devices())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1440x288 with 20 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "decoded_imgs = autoencoder3.predict(X1)\n",
    "def bare(ax):\n",
    "    ax.get_xaxis().set_visible(False)\n",
    "    ax.get_yaxis().set_visible(False)\n",
    "    return ax\n",
    "\n",
    "n = 10\n",
    "plt.figure(figsize=(20,4))\n",
    "for i in range(n):\n",
    "    ax = plt.subplot(2, n, i+1)\n",
    "    plt.imshow(X1[i].reshape(28,28))\n",
    "    plt.gray()\n",
    "    bare(ax)\n",
    "    ax = plt.subplot(2, n, i+1+n)\n",
    "    plt.imshow(decoded_imgs[i].reshape(28,28))\n",
    "    plt.hot()\n",
    "    bare(ax)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python (tensorflow-gpu)",
   "language": "python",
   "name": "tensorflow-gpu"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.5"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}