course_5.ipynb

{
  "cells": [
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "import pandas as pd\nimport numpy as np\nimport matplotlib.pyplot as plt\n%matplotlib inline\n\nimport utils\nutils.set_pretty_prints()",
      "execution_count": 2,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "import tensorflow as tf",
      "execution_count": 3,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "data = tf.keras.datasets.mnist.load_data()",
      "execution_count": 4,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "train, dev = data",
      "execution_count": 5,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "x_train, y_train = train\nx_dev, y_dev = dev",
      "execution_count": 7,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "x_train.shape",
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 8,
          "data": {
            "text/plain": "(60000, 28, 28)"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "y_train",
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 9,
          "data": {
            "text/plain": "array([5, 0, 4, ..., 5, 6, 8], dtype=uint8)"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "y_train.shape",
      "execution_count": 10,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 10,
          "data": {
            "text/plain": "(60000,)"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "(y_train == 8).sum()",
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 11,
          "data": {
            "text/plain": "5851"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "(y_train == 0).sum()",
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 12,
          "data": {
            "text/plain": "5923"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "y_dev",
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 13,
          "data": {
            "text/plain": "array([7, 2, 1, ..., 4, 5, 6], dtype=uint8)"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "y_dev.shape",
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 14,
          "data": {
            "text/plain": "(10000,)"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "(y_dev == 0).sum()",
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 15,
          "data": {
            "text/plain": "980"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "(y_dev == 8).sum()",
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 16,
          "data": {
            "text/plain": "974"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "np.unique(y_train)",
      "execution_count": 27,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 27,
          "data": {
            "text/plain": "array([0, 1, 2, 3, 4, 5, 6, 7, 8, 9], dtype=uint8)"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "np.set_printoptions(linewidth=500)",
      "execution_count": 21,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "x_train[0]",
      "execution_count": 22,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 22,
          "data": {
            "text/plain": "array([[  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   3,  18,  18,  18, 126, 136, 175,  26, 166, 255, 247, 127,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,  30,  36,  94, 154, 170, 253, 253, 253, 253, 253, 225, 172, 253, 242, 195,  64,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,  49, 238, 253, 253, 253, 253, 253, 253, 253, 253, 251,  93,  82,  82,  56,  39,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,  18, 219, 253, 253, 253, 253, 253, 198, 182, 247, 241,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,  80, 156, 107, 253, 253, 205,  11,   0,  43, 154,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,   0,  14,   1, 154, 253,  90,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 139, 253, 190,   2,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  11, 190, 253,  70,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  35, 241, 225, 160, 108,   1,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  81, 240, 253, 253, 119,  25,   0,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  45, 186, 253, 253, 150,  27,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  16,  93, 252, 253, 187,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0, 249, 253, 249,  64,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  46, 130, 183, 253, 253, 207,   2,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  39, 148, 229, 253, 253, 253, 250, 182,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,  24, 114, 221, 253, 253, 253, 253, 201,  78,   0,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,  23,  66, 213, 253, 253, 253, 253, 198,  81,   2,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,  18, 171, 219, 253, 253, 253, 253, 195,  80,   9,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,  55, 172, 226, 253, 253, 253, 253, 244, 133,  11,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0, 136, 253, 253, 253, 212, 135, 132,  16,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0],\n       [  0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0,   0]], dtype=uint8)"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "y_train[0]",
      "execution_count": 23,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 23,
          "data": {
            "text/plain": "5"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "plt.matshow(x_train[0])",
      "execution_count": 24,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 24,
          "data": {
            "text/plain": "<matplotlib.image.AxesImage at 0x7f0cf17a4e10>"
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": "<Figure size 288x288 with 1 Axes>"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "idxs = np.random.choice(6000, size=5)\nfor idx in idxs:\n    plt.figure()\n    plt.matshow(x_train[idx])\n    plt.title('LABEL = ' + str(y_train[idx]))\n    plt.show()",
      "execution_count": 26,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": "<Figure size 432x288 with 0 Axes>"
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": "<Figure size 288x288 with 1 Axes>"
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": "<Figure size 432x288 with 0 Axes>"
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": "<Figure size 288x288 with 1 Axes>"
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": "<Figure size 432x288 with 0 Axes>"
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": "<Figure size 288x288 with 1 Axes>"
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": "<Figure size 432x288 with 0 Axes>"
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": "<Figure size 288x288 with 1 Axes>"
          },
          "metadata": {
            "needs_background": "light"
          }
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": "<Figure size 432x288 with 0 Axes>"
          },
          "metadata": {}
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "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\n",
            "text/plain": "<Figure size 288x288 with 1 Axes>"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "v = [100, 100, 100, 100]\nsoftmax_v = [0.25, 0.25, 0.25, 0.25]",
      "execution_count": 29,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "def softmax(v):\n    return np.exp(v) / np.exp(v).sum()\n    \nsoftmax([100, 200, 300, 400, 500])",
      "execution_count": 37,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 37,
          "data": {
            "text/plain": "1.0"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "tf_input = tf.keras.layers.Input((28,28), name='intrare', dtype='float32')\ntf_x = tf.keras.layers.Flatten(name='mat_to_vect')(tf_input)\n\ntf_x = tf.keras.layers.Dense(512, name='linear1')(tf_x)\ntf_x = tf.keras.layers.Activation('relu', name='non_linearity1')(tf_x)\ntf_x = tf.keras.layers.Dropout(0.5, name='drop1')(tf_x)\n\ntf_x = tf.keras.layers.Dense(256, name='linear1')(tf_x)\ntf_x = tf.keras.layers.Activation('relu', name='non_linearity1')(tf_x)\ntf_x = tf.keras.layers.Dropout(0.5, name='drop1')(tf_x)\n\ntf_out = tf.keras.layers.Dense(10, activation='softmax', name='iesire')(tf_x)\n",
      "execution_count": 38,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "def get_model(input_shape=(28,28), layers=[512, 256], activation='relu', dropout=0.5):\n    tf_input = tf.keras.layers.Input(input_shape, name='intrare', dtype='float32')\n    tf_x = tf.keras.layers.Flatten(name='mat_to_vect')(tf_input)\n    \n    for i,layer in enumerate(layers):\n        tf_x = tf.keras.layers.Dense(layer, name='linear_'+str(i+1))(tf_x)\n        tf_x = tf.keras.layers.Activation(activation, name='non_linearitate_'+str(i+1))(tf_x)\n        tf_x = tf.keras.layers.Dropout(dropout, name='drop_'+str(i+1))(tf_x)\n\n    tf_out = tf.keras.layers.Dense(10, activation='softmax', name='iesire')(tf_x)\n    \n    model = tf.keras.models.Model(tf_input, tf_out, name='model_din_functie')\n    model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['acc'])\n    return model",
      "execution_count": 40,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "def train_model(model):\n    model.fit(x_train, y_train, validation_data=(x_dev, y_dev), epochs=10, batch_size=128)",
      "execution_count": 47,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "m1 = get_model(layers=[512, 256])\ntrain_model(m1)",
      "execution_count": 41,
      "outputs": [
        {
          "output_type": "stream",
          "text": "WARNING:tensorflow:From /home/nbuser/anaconda3_501/lib/python3.6/site-packages/tensorflow/python/ops/math_grad.py:1250: add_dispatch_support.<locals>.wrapper (from tensorflow.python.ops.array_ops) is deprecated and will be removed in a future version.\nInstructions for updating:\nUse tf.where in 2.0, which has the same broadcast rule as np.where\nTrain on 60000 samples, validate on 10000 samples\nEpoch 1/10\n60000/60000 [==============================] - 10s 163us/sample - loss: 10.7227 - acc: 0.3329 - val_loss: 10.0722 - val_acc: 0.3747\nEpoch 2/10\n60000/60000 [==============================] - 9s 156us/sample - loss: 9.9704 - acc: 0.3808 - val_loss: 8.7870 - val_acc: 0.4543\nEpoch 3/10\n60000/60000 [==============================] - 9s 149us/sample - loss: 9.2040 - acc: 0.4283 - val_loss: 8.9356 - val_acc: 0.4450\nEpoch 4/10\n60000/60000 [==============================] - 9s 145us/sample - loss: 7.6202 - acc: 0.5267 - val_loss: 6.2243 - val_acc: 0.6135\nEpoch 5/10\n60000/60000 [==============================] - 9s 152us/sample - loss: 7.1646 - acc: 0.5550 - val_loss: 6.7430 - val_acc: 0.5814\nEpoch 6/10\n60000/60000 [==============================] - 8s 135us/sample - loss: 6.8036 - acc: 0.5775 - val_loss: 6.5945 - val_acc: 0.5908\nEpoch 7/10\n60000/60000 [==============================] - 9s 152us/sample - loss: 6.8512 - acc: 0.5748 - val_loss: 5.9760 - val_acc: 0.6289\nEpoch 8/10\n60000/60000 [==============================] - 8s 139us/sample - loss: 7.1743 - acc: 0.5547 - val_loss: 5.8590 - val_acc: 0.6365\nEpoch 9/10\n60000/60000 [==============================] - 9s 154us/sample - loss: 6.6104 - acc: 0.5897 - val_loss: 6.0748 - val_acc: 0.6228\nEpoch 10/10\n60000/60000 [==============================] - 9s 149us/sample - loss: 6.7663 - acc: 0.5800 - val_loss: 6.2040 - val_acc: 0.6150\n",
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "execution_count": 41,
          "data": {
            "text/plain": "<tensorflow.python.keras.callbacks.History at 0x7f0d2065fbe0>"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "y_pred = m1.predict(x_dev)",
      "execution_count": 42,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "y_pred[:10]",
      "execution_count": 43,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 43,
          "data": {
            "text/plain": "array([[0., 0., 0., 0., 0., 0., 0., 1., 0., 0.],\n       [0., 0., 0., 0., 0., 0., 1., 0., 0., 0.],\n       [0., 1., 0., 0., 0., 0., 0., 0., 0., 0.],\n       [1., 0., 0., 0., 0., 0., 0., 0., 0., 0.],\n       [0., 0., 0., 0., 1., 0., 0., 0., 0., 0.],\n       [0., 1., 0., 0., 0., 0., 0., 0., 0., 0.],\n       [0., 0., 0., 0., 1., 0., 0., 0., 0., 0.],\n       [0., 0., 0., 0., 1., 0., 0., 0., 0., 0.],\n       [0., 0., 0., 0., 0., 0., 1., 0., 0., 0.],\n       [0., 0., 0., 0., 1., 0., 0., 0., 0., 0.]], dtype=float32)"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "y_pred_labels = y_pred.argmax(axis=1)",
      "execution_count": 44,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "y_pred_labels[:10]",
      "execution_count": 45,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 45,
          "data": {
            "text/plain": "array([7, 6, 1, 0, 4, 1, 4, 4, 6, 4])"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "(y_pred_labels == y_dev).sum() / y_dev.shape[0]",
      "execution_count": 46,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 46,
          "data": {
            "text/plain": "0.615"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "m2 = get_model(layers=[1024, 512])\ntrain_model(m2)",
      "execution_count": 48,
      "outputs": [
        {
          "output_type": "stream",
          "text": "Train on 60000 samples, validate on 10000 samples\nEpoch 1/10\n60000/60000 [==============================] - 18s 297us/sample - loss: 12.5072 - acc: 0.2233 - val_loss: 11.9505 - val_acc: 0.2584\nEpoch 2/10\n60000/60000 [==============================] - 17s 290us/sample - loss: 12.2779 - acc: 0.2381 - val_loss: 12.1840 - val_acc: 0.2439\nEpoch 3/10\n44032/60000 [=====================>........] - ETA: 5s - loss: 12.0726 - acc: 0.2508",
          "name": "stdout"
        },
        {
          "output_type": "error",
          "ename": "KeyboardInterrupt",
          "evalue": "",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-48-7a37dfda33ec>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0mm2\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mget_model\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mlayers\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m1024\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m512\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 2\u001b[0;31m \u001b[0mtrain_model\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mm2\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
            "\u001b[0;32m<ipython-input-47-c1836a5236b2>\u001b[0m in \u001b[0;36mtrain_model\u001b[0;34m(model)\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0;32mdef\u001b[0m \u001b[0mtrain_model\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmodel\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 2\u001b[0;31m     \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfit\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx_train\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_train\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mvalidation_data\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx_dev\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_dev\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mepochs\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m10\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[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
            "\u001b[0;32m~/anaconda3_501/lib/python3.6/site-packages/tensorflow/python/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, validation_freq, max_queue_size, workers, use_multiprocessing, **kwargs)\u001b[0m\n\u001b[1;32m    641\u001b[0m         \u001b[0mmax_queue_size\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mmax_queue_size\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    642\u001b[0m         \u001b[0mworkers\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mworkers\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 643\u001b[0;31m         use_multiprocessing=use_multiprocessing)\n\u001b[0m\u001b[1;32m    644\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    645\u001b[0m   def evaluate(self,\n",
            "\u001b[0;32m~/anaconda3_501/lib/python3.6/site-packages/tensorflow/python/keras/engine/training_arrays.py\u001b[0m in \u001b[0;36mfit\u001b[0;34m(self, model, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_freq, **kwargs)\u001b[0m\n\u001b[1;32m    662\u001b[0m         \u001b[0mvalidation_steps\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mvalidation_steps\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    663\u001b[0m         \u001b[0mvalidation_freq\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mvalidation_freq\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 664\u001b[0;31m         steps_name='steps_per_epoch')\n\u001b[0m\u001b[1;32m    665\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    666\u001b[0m   def evaluate(self,\n",
            "\u001b[0;32m~/anaconda3_501/lib/python3.6/site-packages/tensorflow/python/keras/engine/training_arrays.py\u001b[0m in \u001b[0;36mmodel_iteration\u001b[0;34m(model, inputs, targets, sample_weights, batch_size, epochs, verbose, callbacks, val_inputs, val_targets, val_sample_weights, shuffle, initial_epoch, steps_per_epoch, validation_steps, validation_freq, mode, validation_in_fit, prepared_feed_values_from_dataset, steps_name, **kwargs)\u001b[0m\n\u001b[1;32m    381\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    382\u001b[0m         \u001b[0;31m# Get outputs.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 383\u001b[0;31m         \u001b[0mbatch_outs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mf\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mins_batch\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    384\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0misinstance\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbatch_outs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlist\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    385\u001b[0m           \u001b[0mbatch_outs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mbatch_outs\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m~/anaconda3_501/lib/python3.6/site-packages/tensorflow/python/keras/backend.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, inputs)\u001b[0m\n\u001b[1;32m   3508\u001b[0m         \u001b[0mvalue\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmath_ops\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcast\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mvalue\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtensor\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdtype\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3509\u001b[0m       \u001b[0mconverted_inputs\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mappend\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mvalue\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 3510\u001b[0;31m     \u001b[0moutputs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_graph_fn\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0mconverted_inputs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   3511\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3512\u001b[0m     \u001b[0;31m# EagerTensor.numpy() will often make a copy to ensure memory safety.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m~/anaconda3_501/lib/python3.6/site-packages/tensorflow/python/eager/function.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m    570\u001b[0m       raise TypeError(\"Keyword arguments {} unknown. Expected {}.\".format(\n\u001b[1;32m    571\u001b[0m           list(kwargs.keys()), list(self._arg_keywords)))\n\u001b[0;32m--> 572\u001b[0;31m     \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_call_flat\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    573\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    574\u001b[0m   \u001b[0;32mdef\u001b[0m \u001b[0m_filtered_call\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0margs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m~/anaconda3_501/lib/python3.6/site-packages/tensorflow/python/eager/function.py\u001b[0m in \u001b[0;36m_call_flat\u001b[0;34m(self, args)\u001b[0m\n\u001b[1;32m    669\u001b[0m     \u001b[0;31m# Only need to override the gradient in graph mode and when we have outputs.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    670\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mcontext\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mexecuting_eagerly\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mor\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0moutputs\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 671\u001b[0;31m       \u001b[0moutputs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_inference_function\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcall\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mctx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0margs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    672\u001b[0m     \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    673\u001b[0m       \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_register_gradient\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m~/anaconda3_501/lib/python3.6/site-packages/tensorflow/python/eager/function.py\u001b[0m in \u001b[0;36mcall\u001b[0;34m(self, ctx, args)\u001b[0m\n\u001b[1;32m    443\u001b[0m             attrs=(\"executor_type\", executor_type,\n\u001b[1;32m    444\u001b[0m                    \"config_proto\", config),\n\u001b[0;32m--> 445\u001b[0;31m             ctx=ctx)\n\u001b[0m\u001b[1;32m    446\u001b[0m       \u001b[0;31m# Replace empty list with None\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    447\u001b[0m       \u001b[0moutputs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0moutputs\u001b[0m \u001b[0;32mor\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m~/anaconda3_501/lib/python3.6/site-packages/tensorflow/python/eager/execute.py\u001b[0m in \u001b[0;36mquick_execute\u001b[0;34m(op_name, num_outputs, inputs, attrs, ctx, name)\u001b[0m\n\u001b[1;32m     59\u001b[0m     tensors = pywrap_tensorflow.TFE_Py_Execute(ctx._handle, device_name,\n\u001b[1;32m     60\u001b[0m                                                \u001b[0mop_name\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0minputs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mattrs\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 61\u001b[0;31m                                                num_outputs)\n\u001b[0m\u001b[1;32m     62\u001b[0m   \u001b[0;32mexcept\u001b[0m \u001b[0mcore\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_NotOkStatusException\u001b[0m \u001b[0;32mas\u001b[0m \u001b[0me\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     63\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mname\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
          ]
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "m3 = get_model(layers=[256, 128])\ntrain_model(m3)",
      "execution_count": 49,
      "outputs": [
        {
          "output_type": "stream",
          "text": "Train on 60000 samples, validate on 10000 samples\nEpoch 1/10\n60000/60000 [==============================] - 8s 126us/sample - loss: 9.8987 - acc: 0.3814 - val_loss: 7.2917 - val_acc: 0.5465\nEpoch 2/10\n60000/60000 [==============================] - 6s 98us/sample - loss: 7.5787 - acc: 0.5275 - val_loss: 6.1426 - val_acc: 0.6180\nEpoch 3/10\n60000/60000 [==============================] - 5s 88us/sample - loss: 6.9846 - acc: 0.5649 - val_loss: 5.7926 - val_acc: 0.6394\nEpoch 4/10\n60000/60000 [==============================] - 6s 97us/sample - loss: 6.5268 - acc: 0.5934 - val_loss: 5.6466 - val_acc: 0.6484\nEpoch 5/10\n60000/60000 [==============================] - 6s 96us/sample - loss: 6.3217 - acc: 0.6062 - val_loss: 4.8699 - val_acc: 0.6971\nEpoch 6/10\n60000/60000 [==============================] - 5s 86us/sample - loss: 5.7892 - acc: 0.6395 - val_loss: 4.6564 - val_acc: 0.7103\nEpoch 7/10\n60000/60000 [==============================] - 6s 104us/sample - loss: 5.2092 - acc: 0.6758 - val_loss: 4.4846 - val_acc: 0.7213\nEpoch 8/10\n60000/60000 [==============================] - 5s 89us/sample - loss: 5.1251 - acc: 0.6810 - val_loss: 4.2476 - val_acc: 0.7360\nEpoch 9/10\n60000/60000 [==============================] - 5s 87us/sample - loss: 5.1692 - acc: 0.6783 - val_loss: 4.1431 - val_acc: 0.7424\nEpoch 10/10\n60000/60000 [==============================] - 5s 88us/sample - loss: 5.1411 - acc: 0.6800 - val_loss: 4.3754 - val_acc: 0.7282\n",
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "m4 = get_model(layers=[512], dropout=0.2)\ntrain_model(m4)",
      "execution_count": 50,
      "outputs": [
        {
          "output_type": "stream",
          "text": "Train on 60000 samples, validate on 10000 samples\nEpoch 1/10\n60000/60000 [==============================] - 7s 113us/sample - loss: 8.8851 - acc: 0.4469 - val_loss: 8.4469 - val_acc: 0.4753\nEpoch 2/10\n60000/60000 [==============================] - 7s 123us/sample - loss: 8.6735 - acc: 0.4612 - val_loss: 8.9310 - val_acc: 0.4454\nEpoch 3/10\n60000/60000 [==============================] - 7s 111us/sample - loss: 8.5100 - acc: 0.4717 - val_loss: 8.2358 - val_acc: 0.4888\nEpoch 4/10\n60000/60000 [==============================] - 7s 113us/sample - loss: 8.4751 - acc: 0.4739 - val_loss: 8.3517 - val_acc: 0.4815\nEpoch 5/10\n60000/60000 [==============================] - 9s 144us/sample - loss: 8.4624 - acc: 0.4747 - val_loss: 8.3539 - val_acc: 0.4815\nEpoch 6/10\n60000/60000 [==============================] - 7s 113us/sample - loss: 8.4605 - acc: 0.4748 - val_loss: 8.2184 - val_acc: 0.4897\nEpoch 7/10\n60000/60000 [==============================] - 7s 117us/sample - loss: 8.4557 - acc: 0.4752 - val_loss: 8.1807 - val_acc: 0.4924\nEpoch 8/10\n60000/60000 [==============================] - 8s 126us/sample - loss: 8.4220 - acc: 0.4773 - val_loss: 8.1268 - val_acc: 0.4957\nEpoch 9/10\n60000/60000 [==============================] - 7s 112us/sample - loss: 8.3072 - acc: 0.4845 - val_loss: 8.1434 - val_acc: 0.4947\nEpoch 10/10\n60000/60000 [==============================] - 7s 120us/sample - loss: 8.3818 - acc: 0.4798 - val_loss: 8.4334 - val_acc: 0.4765\n",
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "m5 = get_model(layers=[])\n",
      "execution_count": 51,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "m5.summary()",
      "execution_count": 52,
      "outputs": [
        {
          "output_type": "stream",
          "text": "Model: \"model_din_functie\"\n_________________________________________________________________\nLayer (type)                 Output Shape              Param #   \n=================================================================\nintrare (InputLayer)         [(None, 28, 28)]          0         \n_________________________________________________________________\nmat_to_vect (Flatten)        (None, 784)               0         \n_________________________________________________________________\niesire (Dense)               (None, 10)                7850      \n=================================================================\nTotal params: 7,850\nTrainable params: 7,850\nNon-trainable params: 0\n_________________________________________________________________\n",
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "train_model(m5)",
      "execution_count": 53,
      "outputs": [
        {
          "output_type": "stream",
          "text": "Train on 60000 samples, validate on 10000 samples\nEpoch 1/10\n60000/60000 [==============================] - 3s 48us/sample - loss: 8.7672 - acc: 0.4521 - val_loss: 6.9697 - val_acc: 0.5647\nEpoch 2/10\n60000/60000 [==============================] - 2s 38us/sample - loss: 6.1018 - acc: 0.6181 - val_loss: 4.6768 - val_acc: 0.7053\nEpoch 3/10\n60000/60000 [==============================] - 2s 42us/sample - loss: 4.2705 - acc: 0.7309 - val_loss: 3.9791 - val_acc: 0.7510\nEpoch 4/10\n60000/60000 [==============================] - 3s 44us/sample - loss: 4.0737 - acc: 0.7444 - val_loss: 4.0298 - val_acc: 0.7476\nEpoch 5/10\n60000/60000 [==============================] - 3s 44us/sample - loss: 3.9903 - acc: 0.7497 - val_loss: 3.8838 - val_acc: 0.7563\nEpoch 6/10\n60000/60000 [==============================] - 2s 41us/sample - loss: 3.9424 - acc: 0.7530 - val_loss: 4.0813 - val_acc: 0.7439\nEpoch 7/10\n60000/60000 [==============================] - 3s 45us/sample - loss: 3.9370 - acc: 0.7537 - val_loss: 4.1009 - val_acc: 0.7424\nEpoch 8/10\n60000/60000 [==============================] - 2s 40us/sample - loss: 3.9252 - acc: 0.7539 - val_loss: 3.8717 - val_acc: 0.7580\nEpoch 9/10\n60000/60000 [==============================] - 2s 41us/sample - loss: 3.9082 - acc: 0.7552 - val_loss: 3.9871 - val_acc: 0.7511\nEpoch 10/10\n60000/60000 [==============================] - 2s 39us/sample - loss: 3.8598 - acc: 0.7584 - val_loss: 3.8733 - val_acc: 0.7573\n",
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "x_train_norm = (x_train / 255).astype('float32')\nx_dev_norm = (x_dev / 255).astype('float32')\n\ndef train_model_norm(model):\n    model.fit(x_train_norm, y_train, validation_data=(x_dev_norm, y_dev), epochs=10, batch_size=128)",
      "execution_count": 54,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "m1_norm = get_model(layers=[512, 256])\ntrain_model_norm(m1_norm)",
      "execution_count": 55,
      "outputs": [
        {
          "output_type": "stream",
          "text": "Train on 60000 samples, validate on 10000 samples\nEpoch 1/10\n60000/60000 [==============================] - 12s 197us/sample - loss: 0.3865 - acc: 0.8832 - val_loss: 0.1298 - val_acc: 0.9609\nEpoch 2/10\n60000/60000 [==============================] - 11s 186us/sample - loss: 0.1746 - acc: 0.9482 - val_loss: 0.0986 - val_acc: 0.9695\nEpoch 3/10\n60000/60000 [==============================] - 10s 166us/sample - loss: 0.1345 - acc: 0.9595 - val_loss: 0.0899 - val_acc: 0.9726\nEpoch 4/10\n60000/60000 [==============================] - 11s 175us/sample - loss: 0.1142 - acc: 0.9648 - val_loss: 0.0749 - val_acc: 0.9763\nEpoch 5/10\n60000/60000 [==============================] - 10s 175us/sample - loss: 0.1040 - acc: 0.9687 - val_loss: 0.0703 - val_acc: 0.9798\nEpoch 6/10\n60000/60000 [==============================] - 10s 162us/sample - loss: 0.0916 - acc: 0.9719 - val_loss: 0.0698 - val_acc: 0.9791\nEpoch 7/10\n60000/60000 [==============================] - 9s 145us/sample - loss: 0.0822 - acc: 0.9747 - val_loss: 0.0657 - val_acc: 0.9809\nEpoch 8/10\n60000/60000 [==============================] - 8s 139us/sample - loss: 0.0772 - acc: 0.9765 - val_loss: 0.0661 - val_acc: 0.9813\nEpoch 9/10\n60000/60000 [==============================] - 8s 141us/sample - loss: 0.0719 - acc: 0.9775 - val_loss: 0.0612 - val_acc: 0.9818\nEpoch 10/10\n60000/60000 [==============================] - 10s 161us/sample - loss: 0.0670 - acc: 0.9792 - val_loss: 0.0663 - val_acc: 0.9805\n",
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "m1_norm.summary()",
      "execution_count": 57,
      "outputs": [
        {
          "output_type": "stream",
          "text": "Model: \"model_din_functie\"\n_________________________________________________________________\nLayer (type)                 Output Shape              Param #   \n=================================================================\nintrare (InputLayer)         [(None, 28, 28)]          0         \n_________________________________________________________________\nmat_to_vect (Flatten)        (None, 784)               0         \n_________________________________________________________________\nlinear_1 (Dense)             (None, 512)               401920    \n_________________________________________________________________\nnon_linearitate_1 (Activatio (None, 512)               0         \n_________________________________________________________________\ndrop_1 (Dropout)             (None, 512)               0         \n_________________________________________________________________\nlinear_2 (Dense)             (None, 256)               131328    \n_________________________________________________________________\nnon_linearitate_2 (Activatio (None, 256)               0         \n_________________________________________________________________\ndrop_2 (Dropout)             (None, 256)               0         \n_________________________________________________________________\niesire (Dense)               (None, 10)                2570      \n=================================================================\nTotal params: 535,818\nTrainable params: 535,818\nNon-trainable params: 0\n_________________________________________________________________\n",
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "m5_norm = get_model(layers=[])\nm5_norm.summary()",
      "execution_count": 58,
      "outputs": [
        {
          "output_type": "stream",
          "text": "Model: \"model_din_functie\"\n_________________________________________________________________\nLayer (type)                 Output Shape              Param #   \n=================================================================\nintrare (InputLayer)         [(None, 28, 28)]          0         \n_________________________________________________________________\nmat_to_vect (Flatten)        (None, 784)               0         \n_________________________________________________________________\niesire (Dense)               (None, 10)                7850      \n=================================================================\nTotal params: 7,850\nTrainable params: 7,850\nNon-trainable params: 0\n_________________________________________________________________\n",
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "train_model_norm(m5_norm)",
      "execution_count": 59,
      "outputs": [
        {
          "output_type": "stream",
          "text": "Train on 60000 samples, validate on 10000 samples\nEpoch 1/10\n60000/60000 [==============================] - 3s 46us/sample - loss: 0.6739 - acc: 0.8323 - val_loss: 0.3844 - val_acc: 0.9006\nEpoch 2/10\n60000/60000 [==============================] - 2s 40us/sample - loss: 0.3626 - acc: 0.9020 - val_loss: 0.3208 - val_acc: 0.9123\nEpoch 3/10\n60000/60000 [==============================] - 3s 43us/sample - loss: 0.3203 - acc: 0.9116 - val_loss: 0.2967 - val_acc: 0.9165\nEpoch 4/10\n60000/60000 [==============================] - 2s 40us/sample - loss: 0.3000 - acc: 0.9161 - val_loss: 0.2847 - val_acc: 0.9196\nEpoch 5/10\n60000/60000 [==============================] - 2s 38us/sample - loss: 0.2882 - acc: 0.9195 - val_loss: 0.2794 - val_acc: 0.9222\nEpoch 6/10\n60000/60000 [==============================] - 2s 41us/sample - loss: 0.2799 - acc: 0.9215 - val_loss: 0.2734 - val_acc: 0.9242\nEpoch 7/10\n60000/60000 [==============================] - 2s 38us/sample - loss: 0.2742 - acc: 0.9231 - val_loss: 0.2701 - val_acc: 0.9255\nEpoch 8/10\n60000/60000 [==============================] - 3s 43us/sample - loss: 0.2696 - acc: 0.9248 - val_loss: 0.2690 - val_acc: 0.9244\nEpoch 9/10\n60000/60000 [==============================] - 3s 44us/sample - loss: 0.2655 - acc: 0.9257 - val_loss: 0.2679 - val_acc: 0.9248\nEpoch 10/10\n60000/60000 [==============================] - 2s 41us/sample - loss: 0.2625 - acc: 0.9268 - val_loss: 0.2654 - val_acc: 0.9250\n",
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "np_img = np.array([[0,0,0,1,1,1], [0,0,0,1,1,1], [0,0,0,1,1,1], [1,1,1,0,0,0], [1,1,1,0,0,0], [1,1,1,0,0,0]])\nnp_img",
      "execution_count": 60,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 60,
          "data": {
            "text/plain": "array([[0, 0, 0, 1, 1, 1],\n       [0, 0, 0, 1, 1, 1],\n       [0, 0, 0, 1, 1, 1],\n       [1, 1, 1, 0, 0, 0],\n       [1, 1, 1, 0, 0, 0],\n       [1, 1, 1, 0, 0, 0]])"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "np_conv_coefs = np.ones(shape=(3,3))\ntf_inp = tf.keras.layers.Input((6,6,1))\nlyr_conv = tf.keras.layers.Conv2D(filters=1, kernel_size=3, strides=1, use_bias=False)\ntf_x = lyr_conv(tf_inp)\nmodel = tf.keras.models.Model(tf_inp, tf_x)",
      "execution_count": 75,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "model.predict(np_img.reshape(1,6,6,1)).shape",
      "execution_count": 76,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 76,
          "data": {
            "text/plain": "(1, 4, 4, 1)"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "lyr_conv.get_weights()",
      "execution_count": 77,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 77,
          "data": {
            "text/plain": "[array([[[[ 0.57]],\n \n         [[-0.51]],\n \n         [[-0.27]]],\n \n \n        [[[ 0.08]],\n \n         [[-0.53]],\n \n         [[-0.1 ]]],\n \n \n        [[[-0.35]],\n \n         [[-0.58]],\n \n         [[-0.49]]]], dtype=float32)]"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "lyr_conv.get_weights()[0].shape",
      "execution_count": 78,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 78,
          "data": {
            "text/plain": "(3, 3, 1, 1)"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "lyr_conv.set_weights([np_conv_coefs.reshape(3,3,1,1)])",
      "execution_count": 80,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "lyr_conv.get_weights()[0]",
      "execution_count": 81,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 81,
          "data": {
            "text/plain": "array([[[[1.]],\n\n        [[1.]],\n\n        [[1.]]],\n\n\n       [[[1.]],\n\n        [[1.]],\n\n        [[1.]]],\n\n\n       [[[1.]],\n\n        [[1.]],\n\n        [[1.]]]], dtype=float32)"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "model.predict(np_img.reshape(1,6,6,1))",
      "execution_count": 82,
      "outputs": [
        {
          "output_type": "execute_result",
          "execution_count": 82,
          "data": {
            "text/plain": "array([[[[0.],\n         [3.],\n         [6.],\n         [9.]],\n\n        [[3.],\n         [4.],\n         [5.],\n         [6.]],\n\n        [[6.],\n         [5.],\n         [4.],\n         [3.]],\n\n        [[9.],\n         [6.],\n         [3.],\n         [0.]]]], dtype=float32)"
          },
          "metadata": {}
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "def get_model_conv(input_shape=(28,28), filters=[16, 32, 64], dropout=0.5, activation='relu', gmp=True):\n    tf_input = tf.keras.layers.Input(input_shape, name='intrare', dtype='float32')\n    tf_x = tf.keras.layers.Reshape(input_shape + (1,), name='prep_for_conv2d')(tf_input)\n    \n    for i,flt in enumerate(filters):\n        tf_x = tf.keras.layers.Conv2D(filters=flt, kernel_size=3, name='conv_'+str(i+1))(tf_x)\n        tf_x = tf.keras.layers.Activation(activation, name='non_linearitate_'+str(i+1))(tf_x)\n        tf_x = tf.keras.layers.Dropout(dropout, name='drop_'+str(i+1))(tf_x)\n    if gmp:\n        tf_x = tf.keras.layers.GlobalMaxPooling2D(name='volume_shrinker')(tf_x)\n    else:\n        tf_x = tf.keras.layers.Flatten(name='volum_to_vector')(tf_x)\n    tf_out = tf.keras.layers.Dense(10, activation='softmax', name='iesire')(tf_x)\n    \n    model = tf.keras.models.Model(tf_input, tf_out, name='model_din_functie')\n    model.compile(loss='sparse_categorical_crossentropy', optimizer='adam', metrics=['acc'])\n    return model    ",
      "execution_count": 86,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "mc1 = get_model_conv()\nmc1.summary()",
      "execution_count": 87,
      "outputs": [
        {
          "output_type": "stream",
          "text": "Model: \"model_din_functie\"\n_________________________________________________________________\nLayer (type)                 Output Shape              Param #   \n=================================================================\nintrare (InputLayer)         [(None, 28, 28)]          0         \n_________________________________________________________________\nprep_for_conv2d (Reshape)    (None, 28, 28, 1)         0         \n_________________________________________________________________\nconv_1 (Conv2D)              (None, 26, 26, 16)        160       \n_________________________________________________________________\nnon_linearitate_1 (Activatio (None, 26, 26, 16)        0         \n_________________________________________________________________\ndrop_1 (Dropout)             (None, 26, 26, 16)        0         \n_________________________________________________________________\nconv_2 (Conv2D)              (None, 24, 24, 32)        4640      \n_________________________________________________________________\nnon_linearitate_2 (Activatio (None, 24, 24, 32)        0         \n_________________________________________________________________\ndrop_2 (Dropout)             (None, 24, 24, 32)        0         \n_________________________________________________________________\nconv_3 (Conv2D)              (None, 22, 22, 64)        18496     \n_________________________________________________________________\nnon_linearitate_3 (Activatio (None, 22, 22, 64)        0         \n_________________________________________________________________\ndrop_3 (Dropout)             (None, 22, 22, 64)        0         \n_________________________________________________________________\nvolume_shrinker (GlobalMaxPo (None, 64)                0         \n_________________________________________________________________\niesire (Dense)               (None, 10)                650       \n=================================================================\nTotal params: 23,946\nTrainable params: 23,946\nNon-trainable params: 0\n_________________________________________________________________\n",
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "train_model_norm(mc1)",
      "execution_count": 88,
      "outputs": [
        {
          "output_type": "stream",
          "text": "Train on 60000 samples, validate on 10000 samples\nEpoch 1/10\n50176/60000 [========================>.....] - ETA: 37s - loss: 1.7651 - acc: 0.3634",
          "name": "stdout"
        },
        {
          "output_type": "error",
          "ename": "KeyboardInterrupt",
          "evalue": "",
          "traceback": [
            "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
            "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
            "\u001b[0;32m<ipython-input-88-7ffc48317c36>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mtrain_model_norm\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmc1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
            "\u001b[0;32m<ipython-input-54-1e29d21cbae9>\u001b[0m in \u001b[0;36mtrain_model_norm\u001b[0;34m(model)\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      4\u001b[0m \u001b[0;32mdef\u001b[0m \u001b[0mtrain_model_norm\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmodel\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 5\u001b[0;31m     \u001b[0mmodel\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mfit\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx_train_norm\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_train\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mvalidation_data\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mx_dev_norm\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0my_dev\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mepochs\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m10\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[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
            "\u001b[0;32m~/anaconda3_501/lib/python3.6/site-packages/tensorflow/python/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, validation_freq, max_queue_size, workers, use_multiprocessing, **kwargs)\u001b[0m\n\u001b[1;32m    641\u001b[0m         \u001b[0mmax_queue_size\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mmax_queue_size\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    642\u001b[0m         \u001b[0mworkers\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mworkers\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 643\u001b[0;31m         use_multiprocessing=use_multiprocessing)\n\u001b[0m\u001b[1;32m    644\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    645\u001b[0m   def evaluate(self,\n",
            "\u001b[0;32m~/anaconda3_501/lib/python3.6/site-packages/tensorflow/python/keras/engine/training_arrays.py\u001b[0m in \u001b[0;36mfit\u001b[0;34m(self, model, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_freq, **kwargs)\u001b[0m\n\u001b[1;32m    662\u001b[0m         \u001b[0mvalidation_steps\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mvalidation_steps\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    663\u001b[0m         \u001b[0mvalidation_freq\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mvalidation_freq\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 664\u001b[0;31m         steps_name='steps_per_epoch')\n\u001b[0m\u001b[1;32m    665\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    666\u001b[0m   def evaluate(self,\n",
            "\u001b[0;32m~/anaconda3_501/lib/python3.6/site-packages/tensorflow/python/keras/engine/training_arrays.py\u001b[0m in \u001b[0;36mmodel_iteration\u001b[0;34m(model, inputs, targets, sample_weights, batch_size, epochs, verbose, callbacks, val_inputs, val_targets, val_sample_weights, shuffle, initial_epoch, steps_per_epoch, validation_steps, validation_freq, mode, validation_in_fit, prepared_feed_values_from_dataset, steps_name, **kwargs)\u001b[0m\n\u001b[1;32m    381\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    382\u001b[0m         \u001b[0;31m# Get outputs.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 383\u001b[0;31m         \u001b[0mbatch_outs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mf\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mins_batch\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    384\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0misinstance\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbatch_outs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlist\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    385\u001b[0m           \u001b[0mbatch_outs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mbatch_outs\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m~/anaconda3_501/lib/python3.6/site-packages/tensorflow/python/keras/backend.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, inputs)\u001b[0m\n\u001b[1;32m   3508\u001b[0m         \u001b[0mvalue\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmath_ops\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcast\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mvalue\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mtensor\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdtype\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3509\u001b[0m       \u001b[0mconverted_inputs\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mappend\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mvalue\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 3510\u001b[0;31m     \u001b[0moutputs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_graph_fn\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0mconverted_inputs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   3511\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   3512\u001b[0m     \u001b[0;31m# EagerTensor.numpy() will often make a copy to ensure memory safety.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m~/anaconda3_501/lib/python3.6/site-packages/tensorflow/python/eager/function.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, *args, **kwargs)\u001b[0m\n\u001b[1;32m    570\u001b[0m       raise TypeError(\"Keyword arguments {} unknown. Expected {}.\".format(\n\u001b[1;32m    571\u001b[0m           list(kwargs.keys()), list(self._arg_keywords)))\n\u001b[0;32m--> 572\u001b[0;31m     \u001b[0;32mreturn\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_call_flat\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0margs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    573\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    574\u001b[0m   \u001b[0;32mdef\u001b[0m \u001b[0m_filtered_call\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0margs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mkwargs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m~/anaconda3_501/lib/python3.6/site-packages/tensorflow/python/eager/function.py\u001b[0m in \u001b[0;36m_call_flat\u001b[0;34m(self, args)\u001b[0m\n\u001b[1;32m    669\u001b[0m     \u001b[0;31m# Only need to override the gradient in graph mode and when we have outputs.\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    670\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mcontext\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mexecuting_eagerly\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mor\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0moutputs\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 671\u001b[0;31m       \u001b[0moutputs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_inference_function\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mcall\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mctx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0margs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    672\u001b[0m     \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    673\u001b[0m       \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_register_gradient\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m~/anaconda3_501/lib/python3.6/site-packages/tensorflow/python/eager/function.py\u001b[0m in \u001b[0;36mcall\u001b[0;34m(self, ctx, args)\u001b[0m\n\u001b[1;32m    443\u001b[0m             attrs=(\"executor_type\", executor_type,\n\u001b[1;32m    444\u001b[0m                    \"config_proto\", config),\n\u001b[0;32m--> 445\u001b[0;31m             ctx=ctx)\n\u001b[0m\u001b[1;32m    446\u001b[0m       \u001b[0;31m# Replace empty list with None\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    447\u001b[0m       \u001b[0moutputs\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0moutputs\u001b[0m \u001b[0;32mor\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;32m~/anaconda3_501/lib/python3.6/site-packages/tensorflow/python/eager/execute.py\u001b[0m in \u001b[0;36mquick_execute\u001b[0;34m(op_name, num_outputs, inputs, attrs, ctx, name)\u001b[0m\n\u001b[1;32m     59\u001b[0m     tensors = pywrap_tensorflow.TFE_Py_Execute(ctx._handle, device_name,\n\u001b[1;32m     60\u001b[0m                                                \u001b[0mop_name\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0minputs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mattrs\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 61\u001b[0;31m                                                num_outputs)\n\u001b[0m\u001b[1;32m     62\u001b[0m   \u001b[0;32mexcept\u001b[0m \u001b[0mcore\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_NotOkStatusException\u001b[0m \u001b[0;32mas\u001b[0m \u001b[0me\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     63\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mname\u001b[0m \u001b[0;32mis\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0;32mNone\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
            "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
          ]
        }
      ]
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "mc2 = get_model()",
      "execution_count": null,
      "outputs": []
    },
    {
      "metadata": {
        "trusted": true
      },
      "cell_type": "code",
      "source": "# TO DO:\n# 1. antrenare parametrii standard\n# 2. antrenare cu filtre marire (16, 64, 256)\n# 3. antrenare cu gmp=False\n# 4. antrenare cu gmp=False si filtre marite",
      "execution_count": null,
      "outputs": []
    }
  ],
  "metadata": {
    "kernelspec": {
      "name": "python36",
      "display_name": "Python 3.6",
      "language": "python"
    },
    "language_info": {
      "mimetype": "text/x-python",
      "nbconvert_exporter": "python",
      "name": "python",
      "pygments_lexer": "ipython3",
      "version": "3.6.6",
      "file_extension": ".py",
      "codemirror_mode": {
        "version": 3,
        "name": "ipython"
      }
    }
  },
  "nbformat": 4,
  "nbformat_minor": 4
}