train_colab_notebook.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "ISPR_project.ipynb",
      "version": "0.3.2",
      "provenance": [],
      "collapsed_sections": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/MarioBonse/Sketch-rnn/blob/master/train_colab_notebook.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7qNPZdlzbl1l",
        "colab_type": "code",
        "outputId": "11e2622c-fa28-46a9-9002-a3e2a9047910",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 70
        }
      },
      "source": [
        "!pip3 install --upgrade --quiet tensorflow-gpu\n",
        "!pip install --upgrade --quiet PyDrive\n",
        "!pip install -U tensorboardcolab\n",
        "!pip install svgwrite"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Requirement already up-to-date: tensorboardcolab in /usr/local/lib/python3.6/dist-packages (0.0.22)\n",
            "Requirement already satisfied: svgwrite in /usr/local/lib/python3.6/dist-packages (1.3.1)\n",
            "Requirement already satisfied: pyparsing>=2.0.1 in /usr/local/lib/python3.6/dist-packages (from svgwrite) (2.4.2)\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "jYt4wH2giQCr",
        "colab_type": "code",
        "outputId": "55e2a887-8079-4dd6-ee49-b49539f301b8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 134
        }
      },
      "source": [
        "!git clone https://github.com/MarioBonse/Sketch-rnn.git\n",
        "import os\n",
        "os.chdir(\"Sketch-rnn\")"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Cloning into 'Sketch-rnn'...\n",
            "remote: Enumerating objects: 10, done.\u001b[K\n",
            "remote: Counting objects:  10% (1/10)\u001b[K\rremote: Counting objects:  20% (2/10)\u001b[K\rremote: Counting objects:  30% (3/10)\u001b[K\rremote: Counting objects:  40% (4/10)\u001b[K\rremote: Counting objects:  50% (5/10)\u001b[K\rremote: Counting objects:  60% (6/10)\u001b[K\rremote: Counting objects:  70% (7/10)\u001b[K\rremote: Counting objects:  80% (8/10)\u001b[K\rremote: Counting objects:  90% (9/10)\u001b[K\rremote: Counting objects: 100% (10/10)\u001b[K\rremote: Counting objects: 100% (10/10), done.\u001b[K\n",
            "remote: Compressing objects: 100% (10/10), done.\u001b[K\n",
            "remote: Total 189 (delta 2), reused 6 (delta 0), pack-reused 179\n",
            "Receiving objects: 100% (189/189), 344.17 MiB | 32.69 MiB/s, done.\n",
            "Resolving deltas: 100% (74/74), done.\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "cfByb0ohor7g",
        "colab_type": "code",
        "outputId": "8eed2eb4-edf6-4976-d33c-007891e2eafd",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 269
        }
      },
      "source": [
        "import data_Manager \n",
        "from HyperParameters import HP\n",
        "import keras\n",
        "import matplotlib.pyplot as plt\n",
        "from train import *\n",
        "import tensorflow as tf\n",
        "from tensorboardcolab import *\n",
        "#import draw\n",
        "\n",
        "#tbc=TensorBoardColab()\n",
        "HP.data_folder = \"data/\"\n",
        "HP.data_name = \"carrot.npz\"\n",
        "HP.data_location = \"data/carrot.npz\"\n",
        "datas = data_Manager.Data()\n",
        "\n",
        "# create the callback for data augmentaion during training\n",
        "train_generator = data_Manager.DataGenerator(datas.train)\n",
        "validation_encoder = datas.valid\n",
        "validation_decoder = data_Manager.create_decoder_input(validation_encoder)\n",
        "vaidation = [validation_encoder, validation_decoder]\n",
        "HP.epochs = 20\n",
        "\n",
        "\n"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "remote: Enumerating objects: 6, done.\u001b[K\n",
            "remote: Counting objects:  16% (1/6)\u001b[K\rremote: Counting objects:  33% (2/6)\u001b[K\rremote: Counting objects:  50% (3/6)\u001b[K\rremote: Counting objects:  66% (4/6)\u001b[K\rremote: Counting objects:  83% (5/6)\u001b[K\rremote: Counting objects: 100% (6/6)\u001b[K\rremote: Counting objects: 100% (6/6), done.\u001b[K\n",
            "remote: Compressing objects: 100% (5/5), done.\u001b[K\n",
            "remote: Total 8 (delta 1), reused 5 (delta 1), pack-reused 2\u001b[K\n",
            "Unpacking objects: 100% (8/8), done.\n",
            "From https://github.com/MarioBonse/Sketch-rnn\n",
            "   edc7b57..15889ec  master     -> origin/master\n",
            "Updating edc7b57..15889ec\n",
            "Fast-forward\n",
            " .gitignore      |   7 \u001b[32m+\u001b[m\u001b[31m------\u001b[m\n",
            " data/carrot.npz | Bin \u001b[31m0\u001b[m -> \u001b[32m10052174\u001b[m bytes\n",
            " data/cat.npz    | Bin \u001b[31m0\u001b[m -> \u001b[32m14842947\u001b[m bytes\n",
            " 3 files changed, 1 insertion(+), 6 deletions(-)\n",
            " create mode 100644 data/carrot.npz\n",
            " create mode 100644 data/cat.npz\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "JMO1HvnekXNi",
        "colab_type": "code",
        "outputId": "fc645849-8ee9-43f5-bbd2-292e61434371",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 178
        }
      },
      "source": [
        "draw.draw_strokes(datas.train[0], svg_filename=\"carrot.svg\")"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<IPython.core.display.SVG object>"
            ],
            "image/svg+xml": "<svg baseProfile=\"full\" height=\"157.71101111776719\" version=\"1.1\" width=\"89.70702528493959\" xmlns=\"http://www.w3.org/2000/svg\" xmlns:ev=\"http://www.w3.org/2001/xml-events\" xmlns:xlink=\"http://www.w3.org/1999/xlink\"><defs/><rect fill=\"white\" height=\"157.71101111776719\" width=\"89.70702528493959\" x=\"0\" y=\"0\"/><path d=\"M37.32286991601573,127.69058263346449 m-1.1410064737051608,0.0 l-2.0538116526692893,-2.2820129474103217 -7.530642726454062,-14.148480273943994 l-1.597409063187225,-5.020428484302708 0.0,-31.71997996900347 l2.2820129474103217,-7.987045315936125 2.738415536892386,-2.966616831633418 l11.410064737051608,-8.215246610677157 5.020428484302708,-1.369207768446193 l9.812655673864382,0.22820129474103215 2.2820129474103217,2.2820129474103217 l1.1410064737051608,3.19481812637445 0.0,14.148480273943994 l-6.3896362527489,22.59192817936218 -1.369207768446193,2.510214242151354 l-0.9128051789641286,14.148480273943994 -0.9128051789641286,3.8794220105975468 l-6.1614349580078684,13.692077684461928 -2.2820129474103217,2.0538116526692893 l0.0,-2.738415536892386 m-15.061285452908123,-47.009466716652625 l3.6512207158565144,-1.597409063187225 10.269058263346446,-1.1410064737051608 l7.987045315936125,-2.2820129474103217 7.074240136971997,-0.6846038842230965 m-25.78674630573663,24.41753853729044 l1.8256103579282572,0.6846038842230965 9.356253084382319,0.0 l9.58445437912335,-2.738415536892386 m-16.20229192661328,-40.163427874421664 l-7.530642726454062,-5.705032368525804 -0.9128051789641286,-1.597409063187225 l-0.4564025894820643,-17.799700989800506 1.369207768446193,-0.6846038842230965 l1.597409063187225,0.9128051789641286 2.0538116526692893,2.0538116526692893 l2.738415536892386,1.369207768446193 2.966616831633418,3.8794220105975468 l1.369207768446193,0.4564025894820643 1.597409063187225,-3.6512207158565144 l2.2820129474103217,-13.692077684461928 1.369207768446193,0.0 l1.597409063187225,14.604882863426058 0.22820129474103215,1.1410064737051608 l1.1410064737051608,0.22820129474103215 4.107623305338579,-8.215246610677157 l2.738415536892386,-2.510214242151354 5.476831073784772,-2.0538116526692893 l6.617837547489933,-0.6846038842230965 -0.6846038842230965,4.564025894820643 l-2.966616831633418,6.617837547489933 -9.58445437912335,14.604882863426058 \" fill=\"none\" stroke=\"black\" stroke-width=\"1\"/></svg>"
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "lvfvYwFXdRTZ",
        "colab_type": "code",
        "outputId": "f5b133c6-80f3-45dd-ca0a-5097b74c6454",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "\n",
        "\"\"\"\n",
        "Create the model\n",
        "\n",
        "1. ENCODER\n",
        "\"\"\"\n",
        "encoder_input = tf.keras.layers.Input(batch_shape = (HP.batch_size, None, HP.input_dimention), name = \"encoder_input\" )\n",
        "\n",
        "encoderLSTM = tf.keras.layers.Bidirectional(tf.keras.layers.LSTM(HP.enc_hidden_size, return_sequences=False,\n",
        "    recurrent_dropout=HP.rec_dropout, name = \"LSTM_encoder\"), merge_mode='concat', name = \"BI_LSTM_encoder\")(encoder_input)\n",
        "\n",
        "hidden_state_mean = tf.keras.layers.Dense(HP.latent_dim, activation='linear', name = \"mean_MLP\")(encoderLSTM)\n",
        "\n",
        "variance_hat = tf.keras.layers.Dense(HP.latent_dim, activation='linear', name = \"variance_MLP\")(encoderLSTM)\n",
        "\n",
        "# from mean and variance to latent vairable z\n",
        "z = tf.keras.layers.Lambda(sampling, output_shape=(HP.latent_dim,), name='z')([hidden_state_mean, variance_hat])\n",
        "\n",
        "# create the model with keras\n",
        "encoder = tf.keras.models.Model(encoder_input, [hidden_state_mean, variance_hat, z], name='encoder')\n",
        "encoder.summary()\n",
        "#tf.keras.utils.plot_model(encoder, to_file='vae_mlp_encoder.png', show_shapes=True)\n",
        "\n",
        "\"\"\" \n",
        "DECODER \n",
        "\"\"\"\n",
        "# the decoder LSTM has an input composed by the sequence and also the hidden state\n",
        "decoder_input_sequence = tf.keras.layers.Input(batch_shape = (HP.batch_size, HP.max_seq_length, HP.input_dimention), name = \"decoder_seq_input\" )\n",
        "# we have also the latent variable as input of the LSTM with the sequence\n",
        "inputLatentVariable = tf.keras.layers.RepeatVector(HP.max_seq_length)(z)\n",
        "# so we concatentate the two vector\n",
        "# as input we have the two vector above concatenated\n",
        "totalInput = tf.keras.layers.Concatenate()([decoder_input_sequence, inputLatentVariable])\n",
        "# Create LSTM for generation with input state = tanh(z)\n",
        "decoderLSTM = tf.keras.layers.LSTM(HP.dec_hidden_size, recurrent_dropout=HP.rec_dropout, \n",
        "                                    return_sequences=True, return_state=True, name = \"LSTM_decoder\")\n",
        "#\n",
        "init_state = tf.keras.layers.Dense(units=(2*decoderLSTM.units), activation='tanh', name = \"decoder_init_stat\")(z)\n",
        "h_0, c_0 = tf.split(init_state, num_or_size_splits=2, axis = 1)\n",
        "\n",
        "# creation of the LSTM\n",
        "decoder_output, _, _ = decoderLSTM(totalInput, initial_state = [h_0, c_0])\n",
        "\n",
        "# dense to output. THe dimention is, as explained in the paper equal to 3 + 6*M\n",
        "# 6 times M= number of mixture \n",
        "output_dimention = (3 + HP.M * 6)\n",
        "distribution_output = tf.keras.layers.Dense(output_dimention, name = \"output_layer\")(decoder_output)\n",
        "\n",
        "# Build Keras model\n",
        "seq_to_seq_VAE = tf.keras.models.Model([encoder_input, decoder_input_sequence], distribution_output)\n",
        "seq_to_seq_VAE.summary()\n",
        "\n"
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "WARNING: Logging before flag parsing goes to stderr.\n",
            "W0827 01:10:47.664777 140392123692928 deprecation.py:506] From /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/init_ops.py:1251: calling VarianceScaling.__init__ (from tensorflow.python.ops.init_ops) with dtype is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Call initializer instance with the dtype argument instead of passing it to the constructor\n",
            "W0827 01:10:47.673987 140392123692928 deprecation.py:506] From /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/init_ops.py:97: calling GlorotUniform.__init__ (from tensorflow.python.ops.init_ops) with dtype is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Call initializer instance with the dtype argument instead of passing it to the constructor\n",
            "W0827 01:10:47.675418 140392123692928 deprecation.py:506] From /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/init_ops.py:97: calling Orthogonal.__init__ (from tensorflow.python.ops.init_ops) with dtype is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Call initializer instance with the dtype argument instead of passing it to the constructor\n",
            "W0827 01:10:47.678017 140392123692928 deprecation.py:506] From /usr/local/lib/python3.6/dist-packages/tensorflow/python/ops/init_ops.py:97: calling Zeros.__init__ (from tensorflow.python.ops.init_ops) with dtype is deprecated and will be removed in a future version.\n",
            "Instructions for updating:\n",
            "Call initializer instance with the dtype argument instead of passing it to the constructor\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Model: \"encoder\"\n",
            "__________________________________________________________________________________________________\n",
            "Layer (type)                    Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            "encoder_input (InputLayer)      [(100, None, 5)]     0                                            \n",
            "__________________________________________________________________________________________________\n",
            "BI_LSTM_encoder (Bidirectional) (100, 512)           536576      encoder_input[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "mean_MLP (Dense)                (100, 256)           131328      BI_LSTM_encoder[0][0]            \n",
            "__________________________________________________________________________________________________\n",
            "variance_MLP (Dense)            (100, 256)           131328      BI_LSTM_encoder[0][0]            \n",
            "__________________________________________________________________________________________________\n",
            "z (Lambda)                      (100, 256)           0           mean_MLP[0][0]                   \n",
            "                                                                 variance_MLP[0][0]               \n",
            "==================================================================================================\n",
            "Total params: 799,232\n",
            "Trainable params: 799,232\n",
            "Non-trainable params: 0\n",
            "__________________________________________________________________________________________________\n",
            "Model: \"model\"\n",
            "__________________________________________________________________________________________________\n",
            "Layer (type)                    Output Shape         Param #     Connected to                     \n",
            "==================================================================================================\n",
            "encoder_input (InputLayer)      [(100, None, 5)]     0                                            \n",
            "__________________________________________________________________________________________________\n",
            "BI_LSTM_encoder (Bidirectional) (100, 512)           536576      encoder_input[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "mean_MLP (Dense)                (100, 256)           131328      BI_LSTM_encoder[0][0]            \n",
            "__________________________________________________________________________________________________\n",
            "variance_MLP (Dense)            (100, 256)           131328      BI_LSTM_encoder[0][0]            \n",
            "__________________________________________________________________________________________________\n",
            "z (Lambda)                      (100, 256)           0           mean_MLP[0][0]                   \n",
            "                                                                 variance_MLP[0][0]               \n",
            "__________________________________________________________________________________________________\n",
            "decoder_seq_input (InputLayer)  [(100, 200, 5)]      0                                            \n",
            "__________________________________________________________________________________________________\n",
            "repeat_vector (RepeatVector)    (100, 200, 256)      0           z[0][0]                          \n",
            "__________________________________________________________________________________________________\n",
            "concatenate (Concatenate)       (100, 200, 261)      0           decoder_seq_input[0][0]          \n",
            "                                                                 repeat_vector[0][0]              \n",
            "__________________________________________________________________________________________________\n",
            "LSTM_decoder (LSTM)             [(100, 200, 512), (1 1585152     concatenate[0][0]                \n",
            "__________________________________________________________________________________________________\n",
            "output_layer (Dense)            (100, 200, 123)      63099       LSTM_decoder[0][0]               \n",
            "==================================================================================================\n",
            "Total params: 2,447,483\n",
            "Trainable params: 2,447,483\n",
            "Non-trainable params: 0\n",
            "__________________________________________________________________________________________________\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "_6-ojE94Od9m",
        "colab_type": "text"
      },
      "source": [
        "Define the loss"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "O8kUj1hlOb_C",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "    KL_weight = tf.keras.backend.variable(0.01, name = 'kl_weight')\n",
        "    KL_wheight_schedule = data_Manager.changing_KL_wheight(KL_weight)\n",
        "    def vae_loss(y_true, y_pred):\n",
        "        \"\"\" Calculate loss = reconstruction loss + KL loss for each data in minibatch \"\"\"\n",
        "        # E[log P(X|z)]\n",
        "        recon = reconstruction_loss(y_true, y_pred)\n",
        "        # D_KL(Q(z|X) || P(z|X)); calculate in closed form as both dist. are Gaussian\n",
        "        kl = kl_loss(hidden_state_mean, variance_hat)\n",
        "\n",
        "        return recon + kl*KL_weight"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "XtTfyMSsUOQ0",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "optimizer = tf.keras.optimizers.Adam(lr = HP.lr, clipvalue= HP.grad_clip)\n",
        "seq_to_seq_VAE.compile(optimizer=optimizer,  loss = vae_loss)\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "DacL6r3Dd4um",
        "colab_type": "code",
        "outputId": "9881efd1-7701-4f29-b84d-240f7450325f",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "source": [
        "device_name = tf.test.gpu_device_name()\n",
        "if device_name != '/device:GPU:0':\n",
        "  raise SystemError('GPU device not found')\n",
        "print('Found GPU at: {}'.format(device_name))"
      ],
      "execution_count": 9,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Found GPU at: /device:GPU:0\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "7axRtgETdjwG",
        "colab_type": "code",
        "outputId": "473f9ec0-4df4-49d6-e820-ca7009442570",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 794
        }
      },
      "source": [
        "with tf.device('/device:GPU:0'):\n",
        "  history = seq_to_seq_VAE.fit_generator(train_generator,\n",
        "                       validation_data= ([validation_encoder, validation_decoder],[validation_encoder]),\n",
        "                       steps_per_epoch=(datas.trainDimention)/HP.batch_size, \n",
        "                       epochs=HP.epochs, \n",
        "                       callbacks=[KL_wheight_schedule])\n",
        "# save the model\n",
        "seq_to_seq_VAE.save_weights(\"model_weight_carrot_20_epochs.h5\")"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "W0825 10:37:28.271658 140462281566080 deprecation.py:323] From /usr/local/lib/python3.6/dist-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.\n",
            "Instructions for updating:\n",
            "Use tf.where in 2.0, which has the same broadcast rule as np.where\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Epoch 1/20\n",
            "700/700 [==============================] - 901s 1s/step - loss: 0.8152 - val_loss: 0.6498\n",
            "Epoch 2/20\n",
            "700/700 [==============================] - 896s 1s/step - loss: 0.6140 - val_loss: 0.5962\n",
            "Epoch 3/20\n",
            "700/700 [==============================] - 896s 1s/step - loss: 0.5780 - val_loss: 0.5701\n",
            "Epoch 4/20\n",
            "700/700 [==============================] - 892s 1s/step - loss: 0.5584 - val_loss: 0.5552\n",
            "Epoch 5/20\n",
            "700/700 [==============================] - 894s 1s/step - loss: 0.5452 - val_loss: 0.5444\n",
            "Epoch 6/20\n",
            "700/700 [==============================] - 894s 1s/step - loss: 0.5355 - val_loss: 0.5357\n",
            "Epoch 7/20\n",
            "700/700 [==============================] - 895s 1s/step - loss: 0.5278 - val_loss: 0.5293\n",
            "Epoch 8/20\n",
            "700/700 [==============================] - 893s 1s/step - loss: 0.5214 - val_loss: 0.5236\n",
            "Epoch 9/20\n",
            "700/700 [==============================] - 894s 1s/step - loss: 0.5166 - val_loss: 0.5203\n",
            "Epoch 10/20\n",
            "700/700 [==============================] - 894s 1s/step - loss: 0.5121 - val_loss: 0.5154\n",
            "Epoch 11/20\n",
            "700/700 [==============================] - 894s 1s/step - loss: 0.5089 - val_loss: 0.5208\n",
            "Epoch 12/20\n",
            "700/700 [==============================] - 894s 1s/step - loss: 0.5068 - val_loss: 0.5101\n",
            "Epoch 13/20\n",
            "700/700 [==============================] - 893s 1s/step - loss: 0.5028 - val_loss: 0.5077\n",
            "Epoch 14/20\n",
            "700/700 [==============================] - 895s 1s/step - loss: 0.5002 - val_loss: 0.5052\n",
            "Epoch 15/20\n",
            "700/700 [==============================] - 894s 1s/step - loss: 0.4980 - val_loss: 0.5037\n",
            "Epoch 16/20\n",
            "700/700 [==============================] - 893s 1s/step - loss: 0.4957 - val_loss: 0.5016\n",
            "Epoch 17/20\n",
            "700/700 [==============================] - 887s 1s/step - loss: 0.4937 - val_loss: 0.5000\n",
            "Epoch 18/20\n",
            "700/700 [==============================] - 890s 1s/step - loss: 0.4919 - val_loss: 0.4987\n",
            "Epoch 19/20\n",
            "700/700 [==============================] - 898s 1s/step - loss: 0.4902 - val_loss: 0.4974\n",
            "Epoch 20/20\n",
            "700/700 [==============================] - 900s 1s/step - loss: 0.4887 - val_loss: 0.4966\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "q8auMLOlPF05",
        "colab_type": "text"
      },
      "source": [
        "Plot the history"
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "OMOfu0LYgLZt",
        "colab_type": "code",
        "outputId": "43af4fc3-9ff1-4c1f-a7a7-87c911983764",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 312
        }
      },
      "source": [
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "print(history.history.keys())\n",
        "# summarize history for loss\n",
        "plt.plot(history.history['loss'])\n",
        "plt.plot(history.history['val_loss'])\n",
        "plt.title('model loss')\n",
        "plt.ylabel('loss')\n",
        "plt.xlabel('epoch')\n",
        "plt.legend(['train', 'test'], loc='upper left')\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "dict_keys(['loss', 'val_loss'])\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYUAAAEWCAYAAACJ0YulAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nRK5CYII=\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kJFbuw9GAK2l",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        "seq_to_seq_VAE.load_weights(\"model_weight_carrot_20_epochs.h5\", by_name = True)\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Fy_NJOcr-SH3",
        "colab_type": "code",
        "outputId": "3c32ae0c-7560-4ac2-a94d-a5eeee192ea8",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 1000
        }
      },
      "source": [
        "HP.epocs = 50\n",
        "with tf.device('/device:GPU:0'):\n",
        "  history = seq_to_seq_VAE.fit_generator(train_generator,\n",
        "                       validation_data= ([validation_encoder, validation_decoder],[validation_encoder]),\n",
        "                       steps_per_epoch=(datas.trainDimention)/HP.batch_size, \n",
        "                       epochs=50, \n",
        "                       callbacks=[KL_wheight_schedule], initial_epoch = 20)\n",
        "# save the model\n",
        "seq_to_seq_VAE.save_weights(\"model_weight_carrot_50_epochs.h5\")\n"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "W0826 16:55:22.563134 139747641997184 deprecation.py:323] From /usr/local/lib/python3.6/dist-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.\n",
            "Instructions for updating:\n",
            "Use tf.where in 2.0, which has the same broadcast rule as np.where\n"
          ],
          "name": "stderr"
        },
        {
          "output_type": "stream",
          "text": [
            "Epoch 21/50\n",
            "700/700 [==============================] - 885s 1s/step - loss: 0.4881 - val_loss: 0.4952\n",
            "Epoch 22/50\n",
            "700/700 [==============================] - 883s 1s/step - loss: 0.4862 - val_loss: 0.4939\n",
            "Epoch 23/50\n",
            "700/700 [==============================] - 882s 1s/step - loss: 0.4849 - val_loss: 0.4934\n",
            "Epoch 24/50\n",
            "700/700 [==============================] - 883s 1s/step - loss: 0.4836 - val_loss: 0.4920\n",
            "Epoch 25/50\n",
            "700/700 [==============================] - 885s 1s/step - loss: 0.4824 - val_loss: 0.4916\n",
            "Epoch 26/50\n",
            "700/700 [==============================] - 882s 1s/step - loss: 0.4815 - val_loss: 0.4905\n",
            "Epoch 27/50\n",
            "700/700 [==============================] - 881s 1s/step - loss: 0.4804 - val_loss: 0.4895\n",
            "Epoch 28/50\n",
            "700/700 [==============================] - 883s 1s/step - loss: 0.4794 - val_loss: 0.4890\n",
            "Epoch 29/50\n",
            "700/700 [==============================] - 880s 1s/step - loss: 0.4787 - val_loss: 0.4888\n",
            "Epoch 30/50\n",
            "700/700 [==============================] - 882s 1s/step - loss: 0.4779 - val_loss: 0.4877\n",
            "Epoch 31/50\n",
            "700/700 [==============================] - 882s 1s/step - loss: 0.4770 - val_loss: 0.4881\n",
            "Epoch 32/50\n",
            "700/700 [==============================] - 882s 1s/step - loss: 0.4765 - val_loss: 0.4868\n",
            "Epoch 33/50\n",
            "700/700 [==============================] - 880s 1s/step - loss: 0.4762 - val_loss: 0.4865\n",
            "Epoch 34/50\n",
            "700/700 [==============================] - 878s 1s/step - loss: 0.4754 - val_loss: 0.4866\n",
            "Epoch 35/50\n",
            "700/700 [==============================] - 876s 1s/step - loss: 0.4746 - val_loss: 0.4855\n",
            "Epoch 36/50\n",
            "700/700 [==============================] - 880s 1s/step - loss: 0.4740 - val_loss: 0.4857\n",
            "Epoch 37/50\n",
            "700/700 [==============================] - 880s 1s/step - loss: 0.4733 - val_loss: 0.4849\n",
            "Epoch 38/50\n",
            "700/700 [==============================] - 879s 1s/step - loss: 0.4731 - val_loss: 0.4847\n",
            "Epoch 39/50\n",
            "700/700 [==============================] - 880s 1s/step - loss: 0.4723 - val_loss: 0.4842\n",
            "Epoch 40/50\n",
            "700/700 [==============================] - 880s 1s/step - loss: 0.4718 - val_loss: 0.4842\n",
            "Epoch 41/50\n",
            "700/700 [==============================] - 880s 1s/step - loss: 0.4714 - val_loss: 0.4833\n",
            "Epoch 42/50\n",
            "700/700 [==============================] - 877s 1s/step - loss: 0.4708 - val_loss: 0.4836\n",
            "Epoch 43/50\n",
            "700/700 [==============================] - 880s 1s/step - loss: 0.4704 - val_loss: 0.4831\n",
            "Epoch 44/50\n",
            "700/700 [==============================] - 880s 1s/step - loss: 0.4700 - val_loss: 0.4823\n",
            "Epoch 45/50\n",
            "700/700 [==============================] - 883s 1s/step - loss: 0.4734 - val_loss: 0.4862\n",
            "Epoch 46/50\n",
            "700/700 [==============================] - 879s 1s/step - loss: 0.4720 - val_loss: 0.4828\n",
            "Epoch 47/50\n",
            "700/700 [==============================] - 880s 1s/step - loss: 0.4698 - val_loss: 0.4822\n",
            "Epoch 48/50\n",
            "700/700 [==============================] - 879s 1s/step - loss: 0.4687 - val_loss: 0.4816\n",
            "Epoch 49/50\n",
            "700/700 [==============================] - 880s 1s/step - loss: 0.4682 - val_loss: 0.4816\n",
            "Epoch 50/50\n",
            "700/700 [==============================] - 879s 1s/step - loss: 0.4678 - val_loss: 0.4813\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "N2kRnCnZ_a7t",
        "colab_type": "code",
        "outputId": "8cca2e3b-611d-472f-fd40-4261472b64c3",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 313
        }
      },
      "source": [
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "print(history.history.keys())\n",
        "# summarize history for loss\n",
        "plt.plot(history.history['loss'])\n",
        "plt.plot(history.history['val_loss'])\n",
        "plt.title('model loss')\n",
        "plt.ylabel('loss')\n",
        "plt.xlabel('epoch')\n",
        "plt.legend(['train', 'test'], loc='upper left')\n",
        "plt.show()"
      ],
      "execution_count": 0,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "dict_keys(['loss', 'val_loss'])\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAZIAAAEWCAYAAABMoxE0AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nFUskxhhjvGKJxBhjjFcskRhjjPHK/wOhisRLt/MmnAAAAABJRU5ErkJggg==\n",
            "text/plain": [
              "<Figure size 432x288 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "e6Ix-Ij6-SQO",
        "colab_type": "code",
        "colab": {}
      },
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}