2-mfcc-nn-streaming/components/kws/tf/dcnn.bin.musan.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "accelerator": "GPU",
    "colab": {
      "name": "dcnn13.bin.ipynb",
      "provenance": [],
      "collapsed_sections": [],
      "machine_shape": "hm"
    },
    "kernelspec": {
      "display_name": "Python 3",
      "name": "python3"
    }
  },
  "cells": [
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "5T3UOLCpGUmg",
        "outputId": "da36129a-070a-46f6-c282-192e1b15f174"
      },
      "source": [
        "import matplotlib.pyplot as plt\n",
        "import numpy as np\n",
        "from tensorflow import keras\n",
        "import tensorflow as tf\n",
        "print(tf.__version__)\n",
        "plt.rc('figure', figsize=(13, 4))"
      ],
      "execution_count": 1,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "2.6.0\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "l_6NYiKJ7qzN"
      },
      "source": [
        "# https://github.com/42io/dataset/tree/master/openslr_musan\n",
        "DATASET_FILE = '1zR4Llk9SyiEddFrXwDQuaKIr4kYerjOo'"
      ],
      "execution_count": 2,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "J4KjyC2X_ZyQ"
      },
      "source": [
        "!pip install -qq gdown\n",
        "!test -f \"$DATASET_FILE\" || gdown --id \"$DATASET_FILE\" -O \"$DATASET_FILE\""
      ],
      "execution_count": 3,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "kIdWKJAOu_p9"
      },
      "source": [
        "md5 = !md5sum \"$DATASET_FILE\"\n",
        "assert '3a65d6a64314e5c343ae52997d078ace' == md5[0].split()[0]"
      ],
      "execution_count": 4,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "tus9FNQ1pYnO"
      },
      "source": [
        "def create_ds(t):\n",
        "  x = 'x_' + t\n",
        "  y = 'y_' + t\n",
        "  d = tf.data.Dataset.from_tensor_slices((dset[x], dset[y]))\n",
        "  d = d.map(lambda x, y: [tf.cast(x, tf.float32), y])\n",
        "  d = d.map(lambda x, y: [x, tf.cast(y, tf.bool)])\n",
        "  d = d.map(lambda x, y: [tf.reshape(x, (49, 13))[1:-1], y])\n",
        "  d = d.cache()\n",
        "  return d"
      ],
      "execution_count": 5,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "zjcgSo1ahoz9",
        "outputId": "aa8bb039-2453-4066-db18-5912be0fff6b"
      },
      "source": [
        "dset = np.load(DATASET_FILE)\n",
        "print(dset['x_train'].shape)"
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "(1825800, 637)\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "oL2-YVKZYpK2",
        "outputId": "f637859c-86a2-40bf-fae4-0c16cf230a26"
      },
      "source": [
        "train_dataset = create_ds('train')\n",
        "test_dataset  = create_ds('test')\n",
        "valid_dataset = create_ds('valid')\n",
        "train_dataset.cardinality().numpy()"
      ],
      "execution_count": 7,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "1825800"
            ]
          },
          "metadata": {},
          "execution_count": 7
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "Xcj5MHTAKYwv"
      },
      "source": [
        "def spectrogram_masking(spectrogram, dim=1, masks_number=2, mask_max_size=5):\n",
        "  \"\"\"Spectrogram masking on frequency or time dimension.\n",
        "  Args:\n",
        "    spectrogram: Input spectrum [batch, time, frequency]\n",
        "    dim: dimension on which masking will be applied: 1 - time; 2 - frequency\n",
        "    masks_number: number of masks\n",
        "    mask_max_size: mask max size\n",
        "  Returns:\n",
        "    masked spectrogram\n",
        "  \"\"\"\n",
        "  if dim not in (1, 2):\n",
        "    raise ValueError('Wrong dim value: %d' % dim)\n",
        "  input_shape = spectrogram.shape\n",
        "  time_size, frequency_size = input_shape[1:3]\n",
        "  dim_size = input_shape[dim]  # size of dimension on which mask is applied\n",
        "  stripe_shape = [1, time_size, frequency_size]\n",
        "  for _ in range(masks_number):\n",
        "    mask_end = tf.random.uniform([], 0, mask_max_size, tf.int32)\n",
        "    mask_start = tf.random.uniform([], 0, dim_size - mask_end, tf.int32)\n",
        "\n",
        "    # initialize stripes with stripe_shape\n",
        "    stripe_ones_left = list(stripe_shape)\n",
        "    stripe_zeros_center = list(stripe_shape)\n",
        "    stripe_ones_right = list(stripe_shape)\n",
        "\n",
        "    # update stripes dim\n",
        "    stripe_ones_left[dim] = dim_size - mask_start - mask_end\n",
        "    stripe_zeros_center[dim] = mask_end\n",
        "    stripe_ones_right[dim] = mask_start\n",
        "\n",
        "    # generate mask\n",
        "    mask = tf.concat((\n",
        "        tf.ones(stripe_ones_left, spectrogram.dtype),\n",
        "        tf.zeros(stripe_zeros_center, spectrogram.dtype),\n",
        "        tf.ones(stripe_ones_right, spectrogram.dtype),\n",
        "    ), dim)\n",
        "    spectrogram = spectrogram * mask\n",
        "  return spectrogram"
      ],
      "execution_count": 8,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "x5m9wOJwCxxA"
      },
      "source": [
        "def train_model(model, train_dataset, valid_dataset):\n",
        "\n",
        "  early_stopping = keras.callbacks.EarlyStopping(\n",
        "        monitor='val_loss',\n",
        "        mode='min',\n",
        "        verbose=1,\n",
        "        patience=100,\n",
        "        restore_best_weights=True)\n",
        "\n",
        "  train_batch = train_dataset.shuffle(train_dataset.cardinality())\n",
        "  train_batch = train_batch.batch(128)\n",
        "  train_batch = train_batch.map(lambda x, y: (spectrogram_masking(x, 1, 3, 3), y))\n",
        "  train_batch = train_batch.map(lambda x, y: (spectrogram_masking(x, 2, 2, 2), y))\n",
        "  train_batch = train_batch.prefetch(tf.data.AUTOTUNE)\n",
        "\n",
        "  valid_batch = valid_dataset.batch(512)\n",
        "  valid_batch = valid_batch.prefetch(tf.data.AUTOTUNE)\n",
        "\n",
        "  # plt.imshow(list(train_batch.take(1))[0][0][0].numpy().T)\n",
        "  # plt.show()\n",
        "\n",
        "  history = model.fit(train_batch,\n",
        "                      validation_data=valid_batch,\n",
        "                      callbacks=[early_stopping],\n",
        "                      verbose=2,\n",
        "                      epochs=1000) # play with google colab time limit\n",
        "\n",
        "  model.set_weights(early_stopping.best_weights)\n",
        "  return history"
      ],
      "execution_count": 9,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "zbLWMOGgVqhv"
      },
      "source": [
        "# https://arxiv.org/pdf/2005.06720.pdf\n",
        "\n",
        "def streaming_input_output(streaming, t, inputs, otputs, x):\n",
        "  if streaming:\n",
        "    otputs.append(x)\n",
        "    x = keras.Input(shape=[t] + x.shape[2:])\n",
        "    inputs.append(x)\n",
        "  return x\n",
        "\n",
        "def build_model(streaming=False):\n",
        "\n",
        "  # resetting the layer name generation counter\n",
        "  keras.backend.clear_session()\n",
        "\n",
        "  inputs  = []\n",
        "  outputs = []\n",
        "\n",
        "  x_in = keras.Input(shape=[1, 1, 13] if streaming else [47, 13])\n",
        "  x = x_in if streaming else keras.backend.expand_dims(x_in, -2)\n",
        "\n",
        "  x = keras.layers.Conv2D(32, 1, use_bias=False)(x)\n",
        "  x = keras.layers.BatchNormalization()(x)\n",
        "  x = keras.layers.ReLU()(x)\n",
        "  x = keras.layers.SpatialDropout2D(x.shape[-1]/1280.0)(x)\n",
        "\n",
        "  for i in range(4):\n",
        "    x = streaming_input_output(streaming, 2, inputs, outputs, x)\n",
        "    x = keras.layers.SeparableConv2D(x.shape[-1], kernel_size=[2, 1], \n",
        "                                     dilation_rate=[1 if streaming else 2**i, 1], \n",
        "                                     use_bias=False)(x)\n",
        "    x = keras.layers.BatchNormalization()(x)\n",
        "    x = keras.layers.ReLU()(x)\n",
        "    x = keras.layers.SpatialDropout2D(x.shape[-1]/1280.0)(x)\n",
        "\n",
        "  x = streaming_input_output(streaming, 32, inputs, outputs, x)\n",
        "  x = keras.layers.AveragePooling2D([x.shape[1], 1])(x)\n",
        "  x = keras.layers.Flatten()(x)\n",
        "\n",
        "  x = keras.layers.Dense(x.shape[-1] // 2, use_bias=False)(x)\n",
        "  x = keras.layers.BatchNormalization()(x)\n",
        "  x = keras.layers.ReLU()(x)\n",
        "  x = keras.layers.Dropout(x.shape[-1]/1280.0)(x)\n",
        "\n",
        "  x = keras.layers.Dense(1)(x)\n",
        "\n",
        "  model = keras.Model(inputs=[x_in] + inputs, outputs=[x] + outputs)\n",
        "\n",
        "  return model"
      ],
      "execution_count": 10,
      "outputs": []
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "hP5H5d2z_SpG",
        "outputId": "164372f0-9bab-4f20-e9cd-3d9634041a69"
      },
      "source": [
        "model = build_model()\n",
        "model.summary()\n",
        "model.compile(\n",
        "  loss=keras.losses.BinaryCrossentropy(from_logits=True),\n",
        "  optimizer=keras.optimizers.Adam(),\n",
        "  metrics=['accuracy']\n",
        ")"
      ],
      "execution_count": 11,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Model: \"model\"\n",
            "_________________________________________________________________\n",
            "Layer (type)                 Output Shape              Param #   \n",
            "=================================================================\n",
            "input_1 (InputLayer)         [(None, 47, 13)]          0         \n",
            "_________________________________________________________________\n",
            "tf.expand_dims (TFOpLambda)  (None, 47, 1, 13)         0         \n",
            "_________________________________________________________________\n",
            "conv2d (Conv2D)              (None, 47, 1, 32)         416       \n",
            "_________________________________________________________________\n",
            "batch_normalization (BatchNo (None, 47, 1, 32)         128       \n",
            "_________________________________________________________________\n",
            "re_lu (ReLU)                 (None, 47, 1, 32)         0         \n",
            "_________________________________________________________________\n",
            "spatial_dropout2d (SpatialDr (None, 47, 1, 32)         0         \n",
            "_________________________________________________________________\n",
            "separable_conv2d (SeparableC (None, 46, 1, 32)         1088      \n",
            "_________________________________________________________________\n",
            "batch_normalization_1 (Batch (None, 46, 1, 32)         128       \n",
            "_________________________________________________________________\n",
            "re_lu_1 (ReLU)               (None, 46, 1, 32)         0         \n",
            "_________________________________________________________________\n",
            "spatial_dropout2d_1 (Spatial (None, 46, 1, 32)         0         \n",
            "_________________________________________________________________\n",
            "separable_conv2d_1 (Separabl (None, 44, 1, 32)         1088      \n",
            "_________________________________________________________________\n",
            "batch_normalization_2 (Batch (None, 44, 1, 32)         128       \n",
            "_________________________________________________________________\n",
            "re_lu_2 (ReLU)               (None, 44, 1, 32)         0         \n",
            "_________________________________________________________________\n",
            "spatial_dropout2d_2 (Spatial (None, 44, 1, 32)         0         \n",
            "_________________________________________________________________\n",
            "separable_conv2d_2 (Separabl (None, 40, 1, 32)         1088      \n",
            "_________________________________________________________________\n",
            "batch_normalization_3 (Batch (None, 40, 1, 32)         128       \n",
            "_________________________________________________________________\n",
            "re_lu_3 (ReLU)               (None, 40, 1, 32)         0         \n",
            "_________________________________________________________________\n",
            "spatial_dropout2d_3 (Spatial (None, 40, 1, 32)         0         \n",
            "_________________________________________________________________\n",
            "separable_conv2d_3 (Separabl (None, 32, 1, 32)         1088      \n",
            "_________________________________________________________________\n",
            "batch_normalization_4 (Batch (None, 32, 1, 32)         128       \n",
            "_________________________________________________________________\n",
            "re_lu_4 (ReLU)               (None, 32, 1, 32)         0         \n",
            "_________________________________________________________________\n",
            "spatial_dropout2d_4 (Spatial (None, 32, 1, 32)         0         \n",
            "_________________________________________________________________\n",
            "average_pooling2d (AveragePo (None, 1, 1, 32)          0         \n",
            "_________________________________________________________________\n",
            "flatten (Flatten)            (None, 32)                0         \n",
            "_________________________________________________________________\n",
            "dense (Dense)                (None, 16)                512       \n",
            "_________________________________________________________________\n",
            "batch_normalization_5 (Batch (None, 16)                64        \n",
            "_________________________________________________________________\n",
            "re_lu_5 (ReLU)               (None, 16)                0         \n",
            "_________________________________________________________________\n",
            "dropout (Dropout)            (None, 16)                0         \n",
            "_________________________________________________________________\n",
            "dense_1 (Dense)              (None, 1)                 17        \n",
            "=================================================================\n",
            "Total params: 6,001\n",
            "Trainable params: 5,649\n",
            "Non-trainable params: 352\n",
            "_________________________________________________________________\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "MpFpu5bZeYj4",
        "outputId": "20807afc-440c-421c-d1e3-c63049b8f181"
      },
      "source": [
        "history = train_model(model, train_dataset, valid_dataset)"
      ],
      "execution_count": 12,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Epoch 1/1000\n",
            "14265/14265 - 384s - loss: 0.0880 - accuracy: 0.9657 - val_loss: 0.0442 - val_accuracy: 0.9860\n",
            "Epoch 2/1000\n",
            "14265/14265 - 327s - loss: 0.0541 - accuracy: 0.9802 - val_loss: 0.0374 - val_accuracy: 0.9875\n",
            "Epoch 3/1000\n",
            "14265/14265 - 327s - loss: 0.0487 - accuracy: 0.9823 - val_loss: 0.0325 - val_accuracy: 0.9898\n",
            "Epoch 4/1000\n",
            "14265/14265 - 327s - loss: 0.0452 - accuracy: 0.9836 - val_loss: 0.0338 - val_accuracy: 0.9882\n",
            "Epoch 5/1000\n",
            "14265/14265 - 327s - loss: 0.0437 - accuracy: 0.9842 - val_loss: 0.0306 - val_accuracy: 0.9911\n",
            "Epoch 6/1000\n",
            "14265/14265 - 326s - loss: 0.0421 - accuracy: 0.9849 - val_loss: 0.0267 - val_accuracy: 0.9918\n",
            "Epoch 7/1000\n",
            "14265/14265 - 326s - loss: 0.0410 - accuracy: 0.9852 - val_loss: 0.0292 - val_accuracy: 0.9911\n",
            "Epoch 8/1000\n",
            "14265/14265 - 327s - loss: 0.0400 - accuracy: 0.9855 - val_loss: 0.0285 - val_accuracy: 0.9905\n",
            "Epoch 9/1000\n",
            "14265/14265 - 326s - loss: 0.0395 - accuracy: 0.9859 - val_loss: 0.0316 - val_accuracy: 0.9905\n",
            "Epoch 10/1000\n",
            "14265/14265 - 326s - loss: 0.0387 - accuracy: 0.9861 - val_loss: 0.0269 - val_accuracy: 0.9918\n",
            "Epoch 11/1000\n",
            "14265/14265 - 325s - loss: 0.0381 - accuracy: 0.9863 - val_loss: 0.0266 - val_accuracy: 0.9907\n",
            "Epoch 12/1000\n",
            "14265/14265 - 326s - loss: 0.0377 - accuracy: 0.9865 - val_loss: 0.0306 - val_accuracy: 0.9896\n",
            "Epoch 13/1000\n",
            "14265/14265 - 326s - loss: 0.0373 - accuracy: 0.9866 - val_loss: 0.0283 - val_accuracy: 0.9905\n",
            "Epoch 14/1000\n",
            "14265/14265 - 326s - loss: 0.0368 - accuracy: 0.9868 - val_loss: 0.0257 - val_accuracy: 0.9922\n",
            "Epoch 15/1000\n",
            "14265/14265 - 325s - loss: 0.0365 - accuracy: 0.9869 - val_loss: 0.0254 - val_accuracy: 0.9920\n",
            "Epoch 16/1000\n",
            "14265/14265 - 325s - loss: 0.0365 - accuracy: 0.9869 - val_loss: 0.0257 - val_accuracy: 0.9918\n",
            "Epoch 17/1000\n",
            "14265/14265 - 325s - loss: 0.0359 - accuracy: 0.9871 - val_loss: 0.0241 - val_accuracy: 0.9926\n",
            "Epoch 18/1000\n",
            "14265/14265 - 325s - loss: 0.0358 - accuracy: 0.9872 - val_loss: 0.0241 - val_accuracy: 0.9918\n",
            "Epoch 19/1000\n",
            "14265/14265 - 325s - loss: 0.0357 - accuracy: 0.9872 - val_loss: 0.0252 - val_accuracy: 0.9926\n",
            "Epoch 20/1000\n",
            "14265/14265 - 326s - loss: 0.0351 - accuracy: 0.9874 - val_loss: 0.0239 - val_accuracy: 0.9922\n",
            "Epoch 21/1000\n",
            "14265/14265 - 326s - loss: 0.0349 - accuracy: 0.9874 - val_loss: 0.0198 - val_accuracy: 0.9936\n",
            "Epoch 22/1000\n",
            "14265/14265 - 327s - loss: 0.0351 - accuracy: 0.9875 - val_loss: 0.0224 - val_accuracy: 0.9934\n",
            "Epoch 23/1000\n",
            "14265/14265 - 326s - loss: 0.0348 - accuracy: 0.9876 - val_loss: 0.0219 - val_accuracy: 0.9932\n",
            "Epoch 24/1000\n",
            "14265/14265 - 326s - loss: 0.0348 - accuracy: 0.9876 - val_loss: 0.0236 - val_accuracy: 0.9936\n",
            "Epoch 25/1000\n",
            "14265/14265 - 326s - loss: 0.0345 - accuracy: 0.9876 - val_loss: 0.0239 - val_accuracy: 0.9922\n",
            "Epoch 26/1000\n",
            "14265/14265 - 325s - loss: 0.0343 - accuracy: 0.9878 - val_loss: 0.0237 - val_accuracy: 0.9934\n",
            "Epoch 27/1000\n",
            "14265/14265 - 325s - loss: 0.0342 - accuracy: 0.9878 - val_loss: 0.0245 - val_accuracy: 0.9922\n",
            "Epoch 28/1000\n",
            "14265/14265 - 325s - loss: 0.0341 - accuracy: 0.9878 - val_loss: 0.0220 - val_accuracy: 0.9928\n",
            "Epoch 29/1000\n",
            "14265/14265 - 326s - loss: 0.0341 - accuracy: 0.9879 - val_loss: 0.0234 - val_accuracy: 0.9920\n",
            "Epoch 30/1000\n",
            "14265/14265 - 326s - loss: 0.0339 - accuracy: 0.9879 - val_loss: 0.0223 - val_accuracy: 0.9939\n",
            "Epoch 31/1000\n",
            "14265/14265 - 326s - loss: 0.0335 - accuracy: 0.9881 - val_loss: 0.0220 - val_accuracy: 0.9941\n",
            "Epoch 32/1000\n",
            "14265/14265 - 327s - loss: 0.0337 - accuracy: 0.9879 - val_loss: 0.0212 - val_accuracy: 0.9937\n",
            "Epoch 33/1000\n",
            "14265/14265 - 326s - loss: 0.0338 - accuracy: 0.9879 - val_loss: 0.0228 - val_accuracy: 0.9928\n",
            "Epoch 34/1000\n",
            "14265/14265 - 325s - loss: 0.0335 - accuracy: 0.9879 - val_loss: 0.0233 - val_accuracy: 0.9926\n",
            "Epoch 35/1000\n",
            "14265/14265 - 325s - loss: 0.0334 - accuracy: 0.9880 - val_loss: 0.0224 - val_accuracy: 0.9932\n",
            "Epoch 36/1000\n",
            "14265/14265 - 326s - loss: 0.0335 - accuracy: 0.9880 - val_loss: 0.0191 - val_accuracy: 0.9936\n",
            "Epoch 37/1000\n",
            "14265/14265 - 326s - loss: 0.0331 - accuracy: 0.9882 - val_loss: 0.0261 - val_accuracy: 0.9922\n",
            "Epoch 38/1000\n",
            "14265/14265 - 326s - loss: 0.0332 - accuracy: 0.9881 - val_loss: 0.0218 - val_accuracy: 0.9926\n",
            "Epoch 39/1000\n",
            "14265/14265 - 325s - loss: 0.0329 - accuracy: 0.9882 - val_loss: 0.0223 - val_accuracy: 0.9928\n",
            "Epoch 40/1000\n",
            "14265/14265 - 325s - loss: 0.0332 - accuracy: 0.9882 - val_loss: 0.0208 - val_accuracy: 0.9936\n",
            "Epoch 41/1000\n",
            "14265/14265 - 326s - loss: 0.0329 - accuracy: 0.9883 - val_loss: 0.0212 - val_accuracy: 0.9926\n",
            "Epoch 42/1000\n",
            "14265/14265 - 325s - loss: 0.0328 - accuracy: 0.9883 - val_loss: 0.0200 - val_accuracy: 0.9937\n",
            "Epoch 43/1000\n",
            "14265/14265 - 325s - loss: 0.0327 - accuracy: 0.9884 - val_loss: 0.0207 - val_accuracy: 0.9932\n",
            "Epoch 44/1000\n",
            "14265/14265 - 324s - loss: 0.0328 - accuracy: 0.9883 - val_loss: 0.0241 - val_accuracy: 0.9920\n",
            "Epoch 45/1000\n",
            "14265/14265 - 325s - loss: 0.0325 - accuracy: 0.9884 - val_loss: 0.0240 - val_accuracy: 0.9930\n",
            "Epoch 46/1000\n",
            "14265/14265 - 326s - loss: 0.0326 - accuracy: 0.9884 - val_loss: 0.0219 - val_accuracy: 0.9928\n",
            "Epoch 47/1000\n",
            "14265/14265 - 328s - loss: 0.0326 - accuracy: 0.9884 - val_loss: 0.0219 - val_accuracy: 0.9936\n",
            "Epoch 48/1000\n",
            "14265/14265 - 325s - loss: 0.0325 - accuracy: 0.9884 - val_loss: 0.0207 - val_accuracy: 0.9932\n",
            "Epoch 49/1000\n",
            "14265/14265 - 325s - loss: 0.0327 - accuracy: 0.9884 - val_loss: 0.0225 - val_accuracy: 0.9928\n",
            "Epoch 50/1000\n",
            "14265/14265 - 328s - loss: 0.0323 - accuracy: 0.9885 - val_loss: 0.0220 - val_accuracy: 0.9928\n",
            "Epoch 51/1000\n",
            "14265/14265 - 325s - loss: 0.0327 - accuracy: 0.9883 - val_loss: 0.0215 - val_accuracy: 0.9932\n",
            "Epoch 52/1000\n",
            "14265/14265 - 325s - loss: 0.0321 - accuracy: 0.9885 - val_loss: 0.0203 - val_accuracy: 0.9932\n",
            "Epoch 53/1000\n",
            "14265/14265 - 326s - loss: 0.0323 - accuracy: 0.9885 - val_loss: 0.0205 - val_accuracy: 0.9922\n",
            "Epoch 54/1000\n",
            "14265/14265 - 326s - loss: 0.0325 - accuracy: 0.9884 - val_loss: 0.0206 - val_accuracy: 0.9932\n",
            "Epoch 55/1000\n",
            "14265/14265 - 326s - loss: 0.0322 - accuracy: 0.9886 - val_loss: 0.0237 - val_accuracy: 0.9928\n",
            "Epoch 56/1000\n",
            "14265/14265 - 326s - loss: 0.0320 - accuracy: 0.9885 - val_loss: 0.0208 - val_accuracy: 0.9936\n",
            "Epoch 57/1000\n",
            "14265/14265 - 328s - loss: 0.0321 - accuracy: 0.9887 - val_loss: 0.0211 - val_accuracy: 0.9934\n",
            "Epoch 58/1000\n",
            "14265/14265 - 326s - loss: 0.0318 - accuracy: 0.9887 - val_loss: 0.0219 - val_accuracy: 0.9926\n",
            "Epoch 59/1000\n",
            "14265/14265 - 326s - loss: 0.0318 - accuracy: 0.9887 - val_loss: 0.0217 - val_accuracy: 0.9934\n",
            "Epoch 60/1000\n",
            "14265/14265 - 330s - loss: 0.0320 - accuracy: 0.9886 - val_loss: 0.0221 - val_accuracy: 0.9930\n",
            "Epoch 61/1000\n",
            "14265/14265 - 327s - loss: 0.0321 - accuracy: 0.9885 - val_loss: 0.0239 - val_accuracy: 0.9930\n",
            "Epoch 62/1000\n",
            "14265/14265 - 326s - loss: 0.0317 - accuracy: 0.9886 - val_loss: 0.0210 - val_accuracy: 0.9934\n",
            "Epoch 63/1000\n",
            "14265/14265 - 325s - loss: 0.0319 - accuracy: 0.9886 - val_loss: 0.0242 - val_accuracy: 0.9915\n",
            "Epoch 64/1000\n",
            "14265/14265 - 326s - loss: 0.0319 - accuracy: 0.9887 - val_loss: 0.0221 - val_accuracy: 0.9928\n",
            "Epoch 65/1000\n",
            "14265/14265 - 327s - loss: 0.0320 - accuracy: 0.9886 - val_loss: 0.0213 - val_accuracy: 0.9930\n",
            "Epoch 66/1000\n",
            "14265/14265 - 328s - loss: 0.0318 - accuracy: 0.9887 - val_loss: 0.0237 - val_accuracy: 0.9918\n",
            "Epoch 67/1000\n",
            "14265/14265 - 328s - loss: 0.0320 - accuracy: 0.9886 - val_loss: 0.0224 - val_accuracy: 0.9932\n",
            "Epoch 68/1000\n",
            "14265/14265 - 326s - loss: 0.0318 - accuracy: 0.9887 - val_loss: 0.0213 - val_accuracy: 0.9932\n",
            "Epoch 69/1000\n",
            "14265/14265 - 326s - loss: 0.0317 - accuracy: 0.9888 - val_loss: 0.0214 - val_accuracy: 0.9934\n",
            "Epoch 70/1000\n",
            "14265/14265 - 327s - loss: 0.0320 - accuracy: 0.9886 - val_loss: 0.0228 - val_accuracy: 0.9928\n",
            "Epoch 71/1000\n",
            "14265/14265 - 326s - loss: 0.0314 - accuracy: 0.9887 - val_loss: 0.0230 - val_accuracy: 0.9930\n",
            "Epoch 72/1000\n",
            "14265/14265 - 326s - loss: 0.0315 - accuracy: 0.9888 - val_loss: 0.0219 - val_accuracy: 0.9917\n",
            "Epoch 73/1000\n",
            "14265/14265 - 327s - loss: 0.0314 - accuracy: 0.9888 - val_loss: 0.0249 - val_accuracy: 0.9917\n",
            "Epoch 74/1000\n",
            "14265/14265 - 327s - loss: 0.0314 - accuracy: 0.9888 - val_loss: 0.0232 - val_accuracy: 0.9930\n",
            "Epoch 75/1000\n",
            "14265/14265 - 327s - loss: 0.0315 - accuracy: 0.9887 - val_loss: 0.0237 - val_accuracy: 0.9928\n",
            "Epoch 76/1000\n",
            "14265/14265 - 326s - loss: 0.0319 - accuracy: 0.9887 - val_loss: 0.0220 - val_accuracy: 0.9934\n",
            "Epoch 77/1000\n",
            "14265/14265 - 326s - loss: 0.0315 - accuracy: 0.9888 - val_loss: 0.0211 - val_accuracy: 0.9928\n",
            "Epoch 78/1000\n",
            "14265/14265 - 326s - loss: 0.0312 - accuracy: 0.9889 - val_loss: 0.0215 - val_accuracy: 0.9930\n",
            "Epoch 79/1000\n",
            "14265/14265 - 326s - loss: 0.0313 - accuracy: 0.9887 - val_loss: 0.0220 - val_accuracy: 0.9936\n",
            "Epoch 80/1000\n",
            "14265/14265 - 326s - loss: 0.0310 - accuracy: 0.9889 - val_loss: 0.0236 - val_accuracy: 0.9928\n",
            "Epoch 81/1000\n",
            "14265/14265 - 326s - loss: 0.0315 - accuracy: 0.9887 - val_loss: 0.0256 - val_accuracy: 0.9926\n",
            "Epoch 82/1000\n",
            "14265/14265 - 325s - loss: 0.0312 - accuracy: 0.9889 - val_loss: 0.0220 - val_accuracy: 0.9930\n",
            "Epoch 83/1000\n",
            "14265/14265 - 325s - loss: 0.0313 - accuracy: 0.9888 - val_loss: 0.0216 - val_accuracy: 0.9936\n",
            "Epoch 84/1000\n",
            "14265/14265 - 325s - loss: 0.0312 - accuracy: 0.9889 - val_loss: 0.0222 - val_accuracy: 0.9930\n",
            "Epoch 85/1000\n",
            "14265/14265 - 326s - loss: 0.0310 - accuracy: 0.9890 - val_loss: 0.0206 - val_accuracy: 0.9941\n",
            "Epoch 86/1000\n",
            "14265/14265 - 326s - loss: 0.0313 - accuracy: 0.9889 - val_loss: 0.0219 - val_accuracy: 0.9941\n",
            "Epoch 87/1000\n",
            "14265/14265 - 326s - loss: 0.0310 - accuracy: 0.9889 - val_loss: 0.0232 - val_accuracy: 0.9937\n",
            "Epoch 88/1000\n",
            "14265/14265 - 326s - loss: 0.0308 - accuracy: 0.9890 - val_loss: 0.0207 - val_accuracy: 0.9932\n",
            "Epoch 89/1000\n",
            "14265/14265 - 326s - loss: 0.0308 - accuracy: 0.9890 - val_loss: 0.0238 - val_accuracy: 0.9918\n",
            "Epoch 90/1000\n",
            "14265/14265 - 325s - loss: 0.0309 - accuracy: 0.9889 - val_loss: 0.0219 - val_accuracy: 0.9932\n",
            "Epoch 91/1000\n",
            "14265/14265 - 326s - loss: 0.0306 - accuracy: 0.9890 - val_loss: 0.0234 - val_accuracy: 0.9936\n",
            "Epoch 92/1000\n",
            "14265/14265 - 326s - loss: 0.0309 - accuracy: 0.9891 - val_loss: 0.0214 - val_accuracy: 0.9937\n",
            "Epoch 93/1000\n",
            "14265/14265 - 326s - loss: 0.0311 - accuracy: 0.9889 - val_loss: 0.0224 - val_accuracy: 0.9932\n",
            "Epoch 94/1000\n",
            "14265/14265 - 326s - loss: 0.0311 - accuracy: 0.9890 - val_loss: 0.0229 - val_accuracy: 0.9932\n",
            "Epoch 95/1000\n",
            "14265/14265 - 326s - loss: 0.0307 - accuracy: 0.9890 - val_loss: 0.0222 - val_accuracy: 0.9934\n",
            "Epoch 96/1000\n",
            "14265/14265 - 327s - loss: 0.0308 - accuracy: 0.9891 - val_loss: 0.0212 - val_accuracy: 0.9928\n",
            "Epoch 97/1000\n",
            "14265/14265 - 327s - loss: 0.0306 - accuracy: 0.9890 - val_loss: 0.0243 - val_accuracy: 0.9939\n",
            "Epoch 98/1000\n",
            "14265/14265 - 326s - loss: 0.0309 - accuracy: 0.9889 - val_loss: 0.0214 - val_accuracy: 0.9930\n",
            "Epoch 99/1000\n",
            "14265/14265 - 327s - loss: 0.0308 - accuracy: 0.9891 - val_loss: 0.0227 - val_accuracy: 0.9936\n",
            "Epoch 100/1000\n",
            "14265/14265 - 326s - loss: 0.0305 - accuracy: 0.9891 - val_loss: 0.0219 - val_accuracy: 0.9936\n",
            "Epoch 101/1000\n",
            "14265/14265 - 326s - loss: 0.0311 - accuracy: 0.9889 - val_loss: 0.0194 - val_accuracy: 0.9941\n",
            "Epoch 102/1000\n",
            "14265/14265 - 327s - loss: 0.0304 - accuracy: 0.9892 - val_loss: 0.0217 - val_accuracy: 0.9941\n",
            "Epoch 103/1000\n",
            "14265/14265 - 327s - loss: 0.0307 - accuracy: 0.9892 - val_loss: 0.0217 - val_accuracy: 0.9932\n",
            "Epoch 104/1000\n",
            "14265/14265 - 326s - loss: 0.0306 - accuracy: 0.9891 - val_loss: 0.0232 - val_accuracy: 0.9928\n",
            "Epoch 105/1000\n",
            "14265/14265 - 327s - loss: 0.0307 - accuracy: 0.9891 - val_loss: 0.0219 - val_accuracy: 0.9939\n",
            "Epoch 106/1000\n",
            "14265/14265 - 327s - loss: 0.0306 - accuracy: 0.9891 - val_loss: 0.0218 - val_accuracy: 0.9937\n",
            "Epoch 107/1000\n",
            "14265/14265 - 331s - loss: 0.0306 - accuracy: 0.9892 - val_loss: 0.0229 - val_accuracy: 0.9939\n",
            "Epoch 108/1000\n",
            "14265/14265 - 327s - loss: 0.0307 - accuracy: 0.9890 - val_loss: 0.0220 - val_accuracy: 0.9934\n",
            "Epoch 109/1000\n",
            "14265/14265 - 327s - loss: 0.0308 - accuracy: 0.9890 - val_loss: 0.0212 - val_accuracy: 0.9941\n",
            "Epoch 110/1000\n",
            "14265/14265 - 327s - loss: 0.0306 - accuracy: 0.9890 - val_loss: 0.0215 - val_accuracy: 0.9939\n",
            "Epoch 111/1000\n",
            "14265/14265 - 326s - loss: 0.0305 - accuracy: 0.9891 - val_loss: 0.0217 - val_accuracy: 0.9941\n",
            "Epoch 112/1000\n",
            "14265/14265 - 326s - loss: 0.0305 - accuracy: 0.9891 - val_loss: 0.0214 - val_accuracy: 0.9943\n",
            "Epoch 113/1000\n",
            "14265/14265 - 326s - loss: 0.0304 - accuracy: 0.9892 - val_loss: 0.0229 - val_accuracy: 0.9936\n",
            "Epoch 114/1000\n",
            "14265/14265 - 327s - loss: 0.0305 - accuracy: 0.9890 - val_loss: 0.0222 - val_accuracy: 0.9936\n",
            "Epoch 115/1000\n",
            "14265/14265 - 326s - loss: 0.0305 - accuracy: 0.9891 - val_loss: 0.0209 - val_accuracy: 0.9943\n",
            "Epoch 116/1000\n",
            "14265/14265 - 326s - loss: 0.0304 - accuracy: 0.9892 - val_loss: 0.0221 - val_accuracy: 0.9936\n",
            "Epoch 117/1000\n",
            "14265/14265 - 326s - loss: 0.0305 - accuracy: 0.9891 - val_loss: 0.0198 - val_accuracy: 0.9947\n",
            "Epoch 118/1000\n",
            "14265/14265 - 326s - loss: 0.0304 - accuracy: 0.9892 - val_loss: 0.0213 - val_accuracy: 0.9941\n",
            "Epoch 119/1000\n",
            "14265/14265 - 326s - loss: 0.0303 - accuracy: 0.9892 - val_loss: 0.0228 - val_accuracy: 0.9939\n",
            "Epoch 120/1000\n",
            "14265/14265 - 326s - loss: 0.0303 - accuracy: 0.9891 - val_loss: 0.0215 - val_accuracy: 0.9934\n",
            "Epoch 121/1000\n",
            "14265/14265 - 327s - loss: 0.0303 - accuracy: 0.9893 - val_loss: 0.0212 - val_accuracy: 0.9945\n",
            "Epoch 122/1000\n",
            "14265/14265 - 326s - loss: 0.0303 - accuracy: 0.9892 - val_loss: 0.0227 - val_accuracy: 0.9941\n",
            "Epoch 123/1000\n",
            "14265/14265 - 336s - loss: 0.0303 - accuracy: 0.9892 - val_loss: 0.0221 - val_accuracy: 0.9939\n",
            "Epoch 124/1000\n",
            "14265/14265 - 357s - loss: 0.0302 - accuracy: 0.9893 - val_loss: 0.0213 - val_accuracy: 0.9937\n",
            "Epoch 125/1000\n",
            "14265/14265 - 326s - loss: 0.0303 - accuracy: 0.9892 - val_loss: 0.0207 - val_accuracy: 0.9932\n",
            "Epoch 126/1000\n",
            "14265/14265 - 326s - loss: 0.0301 - accuracy: 0.9893 - val_loss: 0.0218 - val_accuracy: 0.9941\n",
            "Epoch 127/1000\n",
            "14265/14265 - 326s - loss: 0.0304 - accuracy: 0.9891 - val_loss: 0.0223 - val_accuracy: 0.9934\n",
            "Epoch 128/1000\n",
            "14265/14265 - 326s - loss: 0.0304 - accuracy: 0.9891 - val_loss: 0.0204 - val_accuracy: 0.9937\n",
            "Epoch 129/1000\n",
            "14265/14265 - 326s - loss: 0.0303 - accuracy: 0.9892 - val_loss: 0.0229 - val_accuracy: 0.9928\n",
            "Epoch 130/1000\n",
            "14265/14265 - 352s - loss: 0.0303 - accuracy: 0.9892 - val_loss: 0.0194 - val_accuracy: 0.9945\n",
            "Epoch 131/1000\n",
            "14265/14265 - 326s - loss: 0.0302 - accuracy: 0.9892 - val_loss: 0.0234 - val_accuracy: 0.9936\n",
            "Epoch 132/1000\n",
            "14265/14265 - 326s - loss: 0.0299 - accuracy: 0.9893 - val_loss: 0.0214 - val_accuracy: 0.9943\n",
            "Epoch 133/1000\n",
            "14265/14265 - 325s - loss: 0.0301 - accuracy: 0.9892 - val_loss: 0.0219 - val_accuracy: 0.9936\n",
            "Epoch 134/1000\n",
            "14265/14265 - 325s - loss: 0.0302 - accuracy: 0.9892 - val_loss: 0.0214 - val_accuracy: 0.9936\n",
            "Epoch 135/1000\n",
            "14265/14265 - 325s - loss: 0.0301 - accuracy: 0.9892 - val_loss: 0.0225 - val_accuracy: 0.9932\n",
            "Epoch 136/1000\n",
            "14265/14265 - 325s - loss: 0.0301 - accuracy: 0.9893 - val_loss: 0.0212 - val_accuracy: 0.9943\n",
            "Restoring model weights from the end of the best epoch.\n",
            "Epoch 00136: early stopping\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 313
        },
        "id": "2t2H8UlDWSBA",
        "outputId": "2f208f15-1c6c-491f-a41e-d9240ec1e9dc"
      },
      "source": [
        "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', 'Valid'], loc='upper right')"
      ],
      "execution_count": 13,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.legend.Legend at 0x7fcabd9dccf8>"
            ]
          },
          "metadata": {},
          "execution_count": 13
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 936x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 313
        },
        "id": "XLAAQk3KWTkq",
        "outputId": "3d0b7dad-76b2-4e06-d30e-7680a533a113"
      },
      "source": [
        "plt.plot(history.history['accuracy'])\n",
        "plt.plot(history.history['val_accuracy'])\n",
        "plt.title('Model accuracy')\n",
        "plt.ylabel('Accuracy')\n",
        "plt.xlabel('Epoch')\n",
        "plt.legend(['Train', 'Valid'], loc='lower right')"
      ],
      "execution_count": 14,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<matplotlib.legend.Legend at 0x7fcabd94e3c8>"
            ]
          },
          "metadata": {},
          "execution_count": 14
        },
        {
          "output_type": "display_data",
          "data": {
            "text/plain": [
              "<Figure size 936x288 with 1 Axes>"
            ],
            "image/png": "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\n"
          },
          "metadata": {
            "needs_background": "light"
          }
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "id": "3I8fIWqYsTTT",
        "outputId": "128e963e-8c8a-4e5f-8652-076a37f18ee4"
      },
      "source": [
        "results = model.evaluate(train_dataset.batch(8192), verbose=0)\n",
        "print('train loss, train acc:', results)\n",
        "results = model.evaluate(test_dataset.batch(512), verbose=0)\n",
        "print('test loss, test acc:', results)\n",
        "results = model.evaluate(valid_dataset.batch(512), verbose=0)\n",
        "print('valid loss, valid acc:', results)"
      ],
      "execution_count": 15,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "train loss, train acc: [0.016179345548152924, 0.9948329329490662]\n",
            "test loss, test acc: [0.0193141158670187, 0.994413435459137]\n",
            "valid loss, valid acc: [0.0191042460501194, 0.9935557246208191]\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "RSIMDvP7XL0z",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "64bc1bd6-9873-40cd-fb63-ed42a76fcbc7"
      },
      "source": [
        "pred = model.predict(test_dataset.batch(512)).squeeze(-1) > 0\n",
        "print(np.sum(pred != list(test_dataset.map(lambda x, y: y))), test_dataset.cardinality().numpy())\n",
        "pred = model.predict(valid_dataset.batch(512)).squeeze(-1) > 0\n",
        "print(np.sum(pred != list(valid_dataset.map(lambda x, y: y))), valid_dataset.cardinality().numpy())"
      ],
      "execution_count": 16,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "38 6086\n",
            "33 5276\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "chpjeHQVWazQ",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "2ee58c3e-1819-4f10-9cde-9947d9e18e30"
      },
      "source": [
        "stream_model = build_model(True)\n",
        "for layer in stream_model.layers:\n",
        "  if layer.get_weights():\n",
        "    print(\"Transfer weights for layer {}\".format(layer.name))\n",
        "    layer.set_weights(model.get_layer(name=layer.name).get_weights())"
      ],
      "execution_count": 17,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "Transfer weights for layer dense\n",
            "Transfer weights for layer batch_normalization_5\n",
            "Transfer weights for layer conv2d\n",
            "Transfer weights for layer separable_conv2d\n",
            "Transfer weights for layer separable_conv2d_1\n",
            "Transfer weights for layer separable_conv2d_2\n",
            "Transfer weights for layer separable_conv2d_3\n",
            "Transfer weights for layer batch_normalization\n",
            "Transfer weights for layer batch_normalization_1\n",
            "Transfer weights for layer batch_normalization_2\n",
            "Transfer weights for layer batch_normalization_3\n",
            "Transfer weights for layer batch_normalization_4\n",
            "Transfer weights for layer dense_1\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "w_UWOOb72I06",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "241d4606-5870-41ce-df00-d94455277bcb"
      },
      "source": [
        "converter = tf.lite.TFLiteConverter.from_keras_model(stream_model)\n",
        "tflite_model = converter.convert()\n",
        "with open(\"dcnn13.bin.tflite\", \"wb\") as f:\n",
        "  f.write(tflite_model)"
      ],
      "execution_count": 18,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "WARNING:tensorflow:Compiled the loaded model, but the compiled metrics have yet to be built. `model.compile_metrics` will be empty until you train or evaluate the model.\n",
            "INFO:tensorflow:Assets written to: /tmp/tmpyjmof62x/assets\n"
          ]
        }
      ]
    },
    {
      "cell_type": "code",
      "metadata": {
        "id": "sN7FZnLM2d8V",
        "colab": {
          "base_uri": "https://localhost:8080/"
        },
        "outputId": "d2e16253-dd17-435a-ad35-0cb5b0d4ea31"
      },
      "source": [
        "!md5sum 'dcnn13.bin.tflite'"
      ],
      "execution_count": 19,
      "outputs": [
        {
          "output_type": "stream",
          "name": "stdout",
          "text": [
            "1a9f3dd322377511b6f15f4070c627c3  dcnn13.bin.tflite\r\n"
          ]
        }
      ]
    }
  ]
}