best_model.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-21T10:27:48.547421Z",
     "iopub.status.busy": "2022-11-21T10:27:48.546800Z",
     "iopub.status.idle": "2022-11-21T10:27:48.687300Z",
     "shell.execute_reply": "2022-11-21T10:27:48.683539Z",
     "shell.execute_reply.started": "2022-11-21T10:27:48.547337Z"
    }
   },
   "source": [
    "## Directory definition"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_dir = \"./train\"\n",
    "train_oversampled_dir = \"./train_oversampled\"\n",
    "val_dir = \"./val\""
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Important:\n",
    "ConvexNeXt large is an extremly heavy model, thus we were not able to perform Transfer learning and Fine tuning on the same session without getting kaggle's RAM exploding.\n",
    "For this reason, we opened a first session to perform Transfer learning and saved the model.\n",
    "Finally, we opened a second session running the Fine tuning cells, loading the model got through Transfer learning."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-23T15:09:47.772891Z",
     "iopub.status.busy": "2022-11-23T15:09:47.771696Z",
     "iopub.status.idle": "2022-11-23T15:09:58.082926Z",
     "shell.execute_reply": "2022-11-23T15:09:58.081907Z",
     "shell.execute_reply.started": "2022-11-23T15:09:47.772739Z"
    },
    "id": "I2s82BSfvsKW",
    "outputId": "c778c0a8-9d08-40b2-ff01-328c15fc144f"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Default GPU Device: /device:GPU:0\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2022-11-23 15:09:54.410521: I tensorflow/core/platform/cpu_feature_guard.cc:142] This TensorFlow binary is optimized with oneAPI Deep Neural Network Library (oneDNN) to use the following CPU instructions in performance-critical operations:  AVX2 AVX512F FMA\n",
      "To enable them in other operations, rebuild TensorFlow with the appropriate compiler flags.\n",
      "2022-11-23 15:09:54.785879: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:54.786732: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:54.932098: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:54.933079: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:54.933909: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:54.934691: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:58.038030: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:58.039387: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:58.040483: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:58.041554: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:58.042589: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:58.043663: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1510] Created device /device:GPU:0 with 13789 MB memory:  -> device: 0, name: Tesla T4, pci bus id: 0000:00:04.0, compute capability: 7.5\n",
      "2022-11-23 15:09:58.047573: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:58.048624: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1510] Created device /device:GPU:1 with 13789 MB memory:  -> device: 1, name: Tesla T4, pci bus id: 0000:00:05.0, compute capability: 7.5\n",
      "2022-11-23 15:09:58.061732: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:58.062848: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:58.064061: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:58.065769: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:58.066863: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:58.067953: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:58.069262: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:58.070308: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:58.071376: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:58.072549: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1510] Created device /device:GPU:0 with 13789 MB memory:  -> device: 0, name: Tesla T4, pci bus id: 0000:00:04.0, compute capability: 7.5\n",
      "2022-11-23 15:09:58.072716: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:09:58.073700: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1510] Created device /device:GPU:1 with 13789 MB memory:  -> device: 1, name: Tesla T4, pci bus id: 0000:00:05.0, compute capability: 7.5\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "\n",
    "\n",
    "import keras\n",
    "import numpy as np\n",
    "import os\n",
    "import random\n",
    "import pandas as pd\n",
    "import seaborn as sns\n",
    "import matplotlib as mpl\n",
    "import matplotlib.pyplot as plt\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import accuracy_score, f1_score, precision_score, recall_score\n",
    "from sklearn.metrics import confusion_matrix\n",
    "from PIL import Image\n",
    "import time\n",
    "\n",
    "\n",
    "\n",
    "tfk = tf.keras\n",
    "tfkl = tf.keras.layers\n",
    "\n",
    "\n",
    "# Random seed for reproducibility\n",
    "seed = 42\n",
    "\n",
    "random.seed(seed)\n",
    "os.environ['PYTHONHASHSEED'] = str(seed)\n",
    "np.random.seed(seed)\n",
    "tf.random.set_seed(seed)\n",
    "tf.compat.v1.set_random_seed(seed)\n",
    "\n",
    "if tf.test.gpu_device_name(): \n",
    "    print('Default GPU Device: {}'.format(tf.test.gpu_device_name()))\n",
    "else:\n",
    "    print(\"Warning: no GPU is running\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# COMPUTING WEIGHTS\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "computing weights for the weighted loss function. Less frequent are samples from a given category, the higher will be the penalty on missing the correct label.\n",
    "We could not see any improvement using these weights in all of our experiments, thus we do not use this method in this notebook."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-20T23:23:59.090032Z",
     "iopub.status.busy": "2022-11-20T23:23:59.088618Z",
     "iopub.status.idle": "2022-11-20T23:23:59.113469Z",
     "shell.execute_reply": "2022-11-20T23:23:59.112601Z",
     "shell.execute_reply.started": "2022-11-20T23:23:59.089990Z"
    }
   },
   "outputs": [],
   "source": [
    "labels = os.listdir(train_dir)\n",
    "labels.sort()\n",
    "counts = []\n",
    "weights = {}\n",
    "for label in labels:\n",
    "    counts.append(len(os.listdir(f\"{train_dir}/{label}\")))\n",
    "total = sum(counts)\n",
    "\n",
    "for index, c in enumerate(counts):\n",
    "    weights[index] = (1 / c) * (total / 8.0)\n",
    "weights"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# SELF ENSEMBLING"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Methods used for self ensembling.\n",
    "- tune_inference_augmentation is responsible of turning off the infference mode to the augmentation layers\n",
    "- self_ensembling_predict() method perform multiple predictions on the same validation set (that is augmented by the augmentation layer) allowing to perform self-ensembling.\n",
    "Unfortunally, we were not able to increase the performances with self-enseble."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-23T15:10:05.862271Z",
     "iopub.status.busy": "2022-11-23T15:10:05.861557Z",
     "iopub.status.idle": "2022-11-23T15:10:05.870110Z",
     "shell.execute_reply": "2022-11-23T15:10:05.868859Z",
     "shell.execute_reply.started": "2022-11-23T15:10:05.862232Z"
    }
   },
   "outputs": [],
   "source": [
    "def tune_inference_augmentation(model, activate):\n",
    "    \"\"\"\n",
    "    This method is responsible for turning on or off the inference mode to the augmentation layers of\n",
    "    a given model.\n",
    "    :param model: model whose augmentation layers must be turn\n",
    "    :param activate: boolean value. If True, activate inference mode, otherwise, \n",
    "    turn off inference mode.\n",
    "    :retrun: None\n",
    "    \"\"\"\n",
    "    for l in model.layers:\n",
    "        if type(l) in [tfkl.RandomFlip, tfkl.RandomContrast, tfkl.RandomRotation]:\n",
    "            l.inference=activate\n",
    "\n",
    "def self_ensembling_predict(model, validation, voters=5):\n",
    "    \"\"\"\n",
    "    This method perform a self ensemble prediction. First, it disactivates the inference mode\n",
    "    for the augmentation layers of the model. Given a number of voters, it runs \"voters\" predictions.\n",
    "    Finally averages the answers and gives the argmax for each prediction.\n",
    "    :param model: model for the predictions.\n",
    "    :param validation: dataset object to use for the prediction.\n",
    "    :param voters: number of predictions to run.\n",
    "    :return a tensor containing all the predictions over validation.\n",
    "    \"\"\"\n",
    "    # activating augmentation at inference\n",
    "    tune_inference_augmentation(model, activate=False)\n",
    "    y_predicted = []\n",
    "    for _ in range(voters):\n",
    "        y_predicted.append(model.predict(validation))\n",
    "    y_predicted = np.array(y_predicted)\n",
    "    y_predicted = np.mean(y_predicted, axis=0)\n",
    "    y_predicted = tf.argmax(y_predicted, axis=1)\n",
    "    tune_inference_augmentation(model, activate=True)\n",
    "    return y_predicted\n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-23T15:10:11.172731Z",
     "iopub.status.busy": "2022-11-23T15:10:11.172252Z",
     "iopub.status.idle": "2022-11-23T15:10:11.187340Z",
     "shell.execute_reply": "2022-11-23T15:10:11.186195Z",
     "shell.execute_reply.started": "2022-11-23T15:10:11.172687Z"
    }
   },
   "outputs": [],
   "source": [
    "def build_heatmap(model, validation_generator, ensemble=False):\n",
    "    \"\"\"\n",
    "    This method generates an heatmap used by seaborn to create a confusion matrix.\n",
    "    :param model: model for which the method computes the confusion matrix.\n",
    "    :param validation_generator: dataset object over which the model computes its predictions.\n",
    "    :param ensemble: Optional argument. If ensemble is True, it computes the confusion matrix using\n",
    "    a selfe-ensemble method with 5 voters.\n",
    "    :return pd.DataFrame object containing the confusion matrix.\n",
    "    \"\"\"\n",
    "    if ensemble:\n",
    "        y_predicted = self_ensembling_predict(model, validation_generator, voters=5)\n",
    "    else:\n",
    "        y_predicted = model.predict(validation_generator)\n",
    "        y_predicted = tf.argmax(y_predicted, axis=1)\n",
    "    y_test_labels = np.concatenate([y for x, y in validation_generator], axis=0)\n",
    "    y_test_labels = tf.argmax(y_test_labels, axis=1)\n",
    "    confusion_matrix = tf.math.confusion_matrix(\n",
    "        y_test_labels, \n",
    "        y_predicted,\n",
    "        num_classes=8\n",
    "    )\n",
    "    c = []\n",
    "    for item in confusion_matrix:\n",
    "        c.append(np.around(item / np.sum(item), decimals=3))\n",
    "    df_heatmap = pd.DataFrame(c)\n",
    "    return df_heatmap\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# EFFICIENT PIPELINE"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The following function is used to load the images in dataset objects. Keras recommend to use Dataset objects to avoid data loading being the bottleneck while training. This implementation allows us to exploit the maximum performances from GPUs.\n",
    "\n",
    "Furthermore, to increase the training speed, we use:\n",
    "- prefetch() method to read the next batch while processing the current batch;\n",
    "- cache() to keep the maximum amount of data in cache to limit the number of times we have to read files."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-23T15:10:16.003807Z",
     "iopub.status.busy": "2022-11-23T15:10:16.003414Z",
     "iopub.status.idle": "2022-11-23T15:10:16.013533Z",
     "shell.execute_reply": "2022-11-23T15:10:16.012528Z",
     "shell.execute_reply.started": "2022-11-23T15:10:16.003759Z"
    }
   },
   "outputs": [],
   "source": [
    "\n",
    "def prepare_batches(\n",
    "    TRAINING_DATA_DIR, VALID_DATA_DIR, \n",
    "    IMAGE_SHAPE, BATCH_SIZE\n",
    "):\n",
    "    autotune = tf.data.AUTOTUNE\n",
    "    train = tf.keras.utils.image_dataset_from_directory(TRAINING_DATA_DIR,\n",
    "                                                               labels=\"inferred\",\n",
    "                                                               label_mode=\"categorical\",\n",
    "                                                               batch_size=BATCH_SIZE,\n",
    "                                                               image_size=IMAGE_SHAPE,\n",
    "                                                               shuffle=True\n",
    "                                                               ).prefetch(buffer_size=autotune).cache()\n",
    "    valid =  tf.keras.utils.image_dataset_from_directory(VALID_DATA_DIR,\n",
    "                                                               labels=\"inferred\",\n",
    "                                                               label_mode=\"categorical\",\n",
    "                                                               batch_size=BATCH_SIZE,\n",
    "                                                               image_size=IMAGE_SHAPE,\n",
    "                                                               shuffle=True\n",
    "                                                               ).prefetch(buffer_size=autotune).cache()\n",
    "    return train, valid\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-23T13:44:59.146209Z",
     "iopub.status.busy": "2022-11-23T13:44:59.145530Z",
     "iopub.status.idle": "2022-11-23T13:44:59.162423Z",
     "shell.execute_reply": "2022-11-23T13:44:59.159042Z",
     "shell.execute_reply.started": "2022-11-23T13:44:59.146171Z"
    }
   },
   "outputs": [],
   "source": [
    "def build_model(num_classes, input_shape, n_layers=1, augmentation=True, n_neurons=10, dropout=0, activation=\"relu\"):\n",
    "    model = tfk.models.Sequential()\n",
    "\n",
    "    # preprocessing layers:  \n",
    "    if augmentation:\n",
    "        model.add(tfkl.RandomFlip(mode='horizontal_and_vertical'))\n",
    "        model.add(tfkl.RandomRotation(0.5))\n",
    "        #model.add(tfkl.RandomZoom(0.1))\n",
    "        #model.add(tfkl.RandomContrast(factor=0.2))\n",
    "        model.add(tfkl.RandomTranslation(height_factor=0.2, width_factor=0.2, fill_mode=\"reflect\"))\n",
    "    \n",
    "    # since convexnext large was not implemented in tensorflow 2.6 (used by kaggle) and by upgrading tf to the latest version we\n",
    "    # could not use GPUs anymore, we downloaded the model locally (with tensorflow latest version) und uploaded it on kaggle.\n",
    "    # feature_extractor = tf.keras.models.load_model(\"../input/convexnet/convexnet/large\") \n",
    "    feature_extractor = tf.keras.applications.convnext.ConvNeXtXLarge(include_top=False,\n",
    "                                                                      include_preprocessing=True,\n",
    "                                                                      input_shape=(96, 96, 3))\n",
    "    feature_extractor.trainable = False\n",
    "    \n",
    "    # fully connected stack\n",
    "    model.add(feature_extractor)\n",
    "    model.add(tfkl.Flatten())\n",
    "\n",
    "    model.add(tfkl.Dropout(dropout))\n",
    "    model.add(tfkl.Dense(256, activation=activation, kernel_initializer=tfk.initializers.HeUniform(seed)))\n",
    "    model.add(tfkl.Dropout(dropout))\n",
    "    model.add(tfkl.Dense(128, activation=activation, kernel_initializer=tfk.initializers.HeUniform(seed)))\n",
    "    model.add(tfkl.Dropout(dropout))\n",
    "    model.add(tfkl.Dense(64, activation=activation, kernel_initializer=tfk.initializers.HeUniform(seed)))\n",
    "    model.add(tfkl.Dropout(dropout))\n",
    "    model.add(tfkl.Dense(32, activation=activation, kernel_initializer=tfk.initializers.HeUniform(seed)))\n",
    "    \n",
    "    model.add(tfkl.Dropout(dropout))\n",
    "    model.add(tfkl.Dense(num_classes, activation=\"softmax\", kernel_initializer=tfk.initializers.GlorotUniform(seed)))\n",
    "    return model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Hyperparametes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-23T14:24:45.376268Z",
     "iopub.status.busy": "2022-11-23T14:24:45.375914Z",
     "iopub.status.idle": "2022-11-23T14:24:45.381333Z",
     "shell.execute_reply": "2022-11-23T14:24:45.380358Z",
     "shell.execute_reply.started": "2022-11-23T14:24:45.376237Z"
    }
   },
   "outputs": [],
   "source": [
    "input_shape = (96, 96, 3)\n",
    "IMAGE_SHAPE = (96, 96) \n",
    "TRAINING_DATA_DIR = train_oversampled_dir\n",
    "VALID_DATA_DIR = val_dir\n",
    "BATCH_SIZE = 512\n",
    "EPOCHS = 300\n",
    "lr = 1e-3"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Transfer learning with ConvexNext Large"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We used kaggle to perform our training process. Since Kaggle provides us with 2 GPUs, we exploit both of them using the mirrored strategy."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-23T15:10:20.892471Z",
     "iopub.status.busy": "2022-11-23T15:10:20.891981Z",
     "iopub.status.idle": "2022-11-23T15:10:21.684842Z",
     "shell.execute_reply": "2022-11-23T15:10:21.683841Z",
     "shell.execute_reply.started": "2022-11-23T15:10:20.892430Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2022-11-23 15:10:20.895163: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:10:20.896483: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:10:20.897628: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:10:20.898745: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:10:20.899920: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:10:20.901057: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:10:20.905186: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:10:20.907062: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:10:20.908211: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:10:20.909508: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:10:20.910956: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:10:20.912318: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:10:20.913623: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:10:20.914869: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:10:20.916145: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:10:20.917226: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1510] Created device /job:localhost/replica:0/task:0/device:GPU:0 with 13789 MB memory:  -> device: 0, name: Tesla T4, pci bus id: 0000:00:04.0, compute capability: 7.5\n",
      "2022-11-23 15:10:20.917401: I tensorflow/stream_executor/cuda/cuda_gpu_executor.cc:937] successful NUMA node read from SysFS had negative value (-1), but there must be at least one NUMA node, so returning NUMA node zero\n",
      "2022-11-23 15:10:20.918474: I tensorflow/core/common_runtime/gpu/gpu_device.cc:1510] Created device /job:localhost/replica:0/task:0/device:GPU:1 with 13789 MB memory:  -> device: 1, name: Tesla T4, pci bus id: 0000:00:05.0, compute capability: 7.5\n"
     ]
    }
   ],
   "source": [
    "mirrored_strategy = tf.distribute.MirroredStrategy(devices=[\"/gpu:0\", \"/gpu:1\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-20T23:23:59.204384Z",
     "iopub.status.busy": "2022-11-20T23:23:59.204132Z",
     "iopub.status.idle": "2022-11-20T23:46:00.019991Z",
     "shell.execute_reply": "2022-11-20T23:46:00.018999Z",
     "shell.execute_reply.started": "2022-11-20T23:23:59.204361Z"
    }
   },
   "outputs": [],
   "source": [
    "\n",
    "with mirrored_strategy.scope():\n",
    "    # Prepare the data batches.\n",
    "    train_batches, valid_batches = prepare_batches(\n",
    "        TRAINING_DATA_DIR, VALID_DATA_DIR, IMAGE_SHAPE, BATCH_SIZE\n",
    "    )\n",
    "    # Prepare and compile the model.\n",
    "    model = build_model(num_classes=8,\n",
    "                        input_shape=input_shape,\n",
    "                        augmentation=True,\n",
    "                        n_layers=1,\n",
    "                        n_neurons=256,\n",
    "                        dropout=0.1,\n",
    "                        activation=\"swish\"\n",
    "                       )\n",
    "    model.compile(\n",
    "        optimizer=tf.keras.optimizers.Adam(learning_rate=lr),\n",
    "        loss=tf.keras.losses.CategoricalCrossentropy(),\n",
    "        metrics=['accuracy']\n",
    "    )\n",
    "    # TRAINING PHASE\n",
    "    early_stopping = tf.keras.callbacks.EarlyStopping(patience=30, monitor=\"val_accuracy\", restore_best_weights=True)\n",
    "        \n",
    "    history = model.fit(\n",
    "        train_batches,\n",
    "        epochs=EPOCHS,\n",
    "        validation_data=valid_batches,\n",
    "        verbose=1,\n",
    "        workers=4,\n",
    "        use_multiprocessing=True,\n",
    "        callbacks=[early_stopping]\n",
    "    )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-20T23:46:00.021999Z",
     "iopub.status.busy": "2022-11-20T23:46:00.021653Z",
     "iopub.status.idle": "2022-11-20T23:46:07.005759Z",
     "shell.execute_reply": "2022-11-20T23:46:07.004756Z",
     "shell.execute_reply.started": "2022-11-20T23:46:00.021971Z"
    }
   },
   "outputs": [],
   "source": [
    "x_axis = len(history.history[\"accuracy\"])\n",
    "print(max(history.history[\"val_accuracy\"]), max(history.history[\"accuracy\"]))\n",
    "history.history.keys()\n",
    "plt.figure(figsize=(12, 6))\n",
    "plt.plot([i for i in range(x_axis)], history.history[\"accuracy\"],  color=\"blue\", label=\"train accuracy\")\n",
    "plt.plot([i for i in range(x_axis)], history.history[\"val_accuracy\"], color=\"red\", label=\"validation accuracy\")\n",
    "plt.plot([np.argmax(history.history[\"val_accuracy\"])],\n",
    "         [max(history.history[\"val_accuracy\"])],\n",
    "         marker=\"o\",\n",
    "         markeredgecolor=\"red\",\n",
    "         markerfacecolor=\"red\",\n",
    "         label=\"best model\",\n",
    "         markersize=10\n",
    "        )\n",
    "plt.legend()\n",
    "plt.grid()\n",
    "plt.xlabel(\"Epochs\")\n",
    "plt.ylabel(\"Accuracy\")\n",
    "plt.ylim([0, 1])\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-20T23:46:07.007623Z",
     "iopub.status.busy": "2022-11-20T23:46:07.007263Z",
     "iopub.status.idle": "2022-11-20T23:46:07.237583Z",
     "shell.execute_reply": "2022-11-20T23:46:07.236662Z",
     "shell.execute_reply.started": "2022-11-20T23:46:07.007586Z"
    }
   },
   "outputs": [],
   "source": [
    "plt.figure(figsize=(12, 6))\n",
    "plt.title(\"Cross entropy\")\n",
    "plt.plot([i for i in range(x_axis)], history.history[\"loss\"],  color=\"blue\", label=\"train cross entropy\")\n",
    "plt.plot([i for i in range(x_axis)], history.history[\"val_loss\"], color=\"red\", label=\"validation cross entropy\")\n",
    "plt.plot([np.argmin(history.history[\"val_loss\"])],\n",
    "         [min(history.history[\"val_loss\"])],\n",
    "         marker=\"o\",\n",
    "         markeredgecolor=\"red\",\n",
    "         markerfacecolor=\"red\",\n",
    "         label=\"best model\",\n",
    "         markersize=10\n",
    "        )\n",
    "plt.legend()\n",
    "plt.grid()\n",
    "plt.xlabel(\"Epochs\")\n",
    "plt.ylabel(\"Cross entropy\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-20T23:46:07.239382Z",
     "iopub.status.busy": "2022-11-20T23:46:07.238907Z",
     "iopub.status.idle": "2022-11-20T23:46:18.406367Z",
     "shell.execute_reply": "2022-11-20T23:46:18.405442Z",
     "shell.execute_reply.started": "2022-11-20T23:46:07.239343Z"
    }
   },
   "outputs": [],
   "source": [
    "plt.figure(figsize=(8, 7))\n",
    "df = build_heatmap(model, valid_batches)\n",
    "sns.heatmap(df, annot=True, cmap=\"crest\")\n",
    "plt.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-20T23:46:18.409595Z",
     "iopub.status.busy": "2022-11-20T23:46:18.408527Z",
     "iopub.status.idle": "2022-11-20T23:46:30.621600Z",
     "shell.execute_reply": "2022-11-20T23:46:30.620640Z",
     "shell.execute_reply.started": "2022-11-20T23:46:18.409555Z"
    }
   },
   "outputs": [],
   "source": [
    "plt.figure(figsize=(8, 7))\n",
    "df = build_heatmap(model, valid_batches, ensemble=True)\n",
    "sns.heatmap(df, annot=True, cmap=\"crest\")\n",
    "plt.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-20T23:46:30.623631Z",
     "iopub.status.busy": "2022-11-20T23:46:30.623253Z",
     "iopub.status.idle": "2022-11-20T23:47:15.317026Z",
     "shell.execute_reply": "2022-11-20T23:47:15.315945Z",
     "shell.execute_reply.started": "2022-11-20T23:46:30.623594Z"
    }
   },
   "outputs": [],
   "source": [
    "model.save(f\"models/ConvexNet_transfer_learning_Best\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Fine tuning\n",
    "Since there are too much abstract features that may not fit with our images we want to drop the last layers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-20T23:48:09.233798Z",
     "iopub.status.busy": "2022-11-20T23:48:09.232909Z",
     "iopub.status.idle": "2022-11-20T23:48:10.399815Z",
     "shell.execute_reply": "2022-11-20T23:48:10.398789Z",
     "shell.execute_reply.started": "2022-11-20T23:48:09.233752Z"
    }
   },
   "outputs": [],
   "source": [
    "del model; gc.collect()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-23T15:10:49.111706Z",
     "iopub.status.busy": "2022-11-23T15:10:49.111335Z",
     "iopub.status.idle": "2022-11-23T15:10:49.117182Z",
     "shell.execute_reply": "2022-11-23T15:10:49.116017Z",
     "shell.execute_reply.started": "2022-11-23T15:10:49.111675Z"
    }
   },
   "outputs": [],
   "source": [
    "input_shape = (96, 96, 3)\n",
    "IMAGE_SHAPE = (96, 96) # For resizing\n",
    "TRAINING_DATA_DIR = train_oversampled_dir\n",
    "VALID_DATA_DIR = val_dir\n",
    "BATCH_SIZE = 512\n",
    "EPOCHS = 300"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-23T15:14:40.813434Z",
     "iopub.status.busy": "2022-11-23T15:14:40.813016Z",
     "iopub.status.idle": "2022-11-23T16:45:40.175557Z",
     "shell.execute_reply": "2022-11-23T16:45:40.174471Z",
     "shell.execute_reply.started": "2022-11-23T15:14:40.813400Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 3362 files belonging to 8 classes.\n",
      "Found 705 files belonging to 8 classes.\n",
      "freeze from 150\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2022-11-23 15:15:03.664864: W tensorflow/core/grappler/optimizers/data/auto_shard.cc:695] AUTO sharding policy will apply DATA sharding policy as it failed to apply FILE sharding policy because of the following reason: Found an unshardable source dataset: name: \"TensorSliceDataset/_1\"\n",
      "op: \"TensorSliceDataset\"\n",
      "input: \"Placeholder/_0\"\n",
      "attr {\n",
      "  key: \"Toutput_types\"\n",
      "  value {\n",
      "    list {\n",
      "      type: DT_STRING\n",
      "    }\n",
      "  }\n",
      "}\n",
      "attr {\n",
      "  key: \"output_shapes\"\n",
      "  value {\n",
      "    list {\n",
      "      shape {\n",
      "      }\n",
      "    }\n",
      "  }\n",
      "}\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/300\n",
      "7/7 [==============================] - ETA: 0s - loss: 0.3002 - accuracy: 0.8968"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2022-11-23 15:16:17.093237: W tensorflow/core/grappler/optimizers/data/auto_shard.cc:695] AUTO sharding policy will apply DATA sharding policy as it failed to apply FILE sharding policy because of the following reason: Found an unshardable source dataset: name: \"TensorSliceDataset/_1\"\n",
      "op: \"TensorSliceDataset\"\n",
      "input: \"Placeholder/_0\"\n",
      "attr {\n",
      "  key: \"Toutput_types\"\n",
      "  value {\n",
      "    list {\n",
      "      type: DT_STRING\n",
      "    }\n",
      "  }\n",
      "}\n",
      "attr {\n",
      "  key: \"output_shapes\"\n",
      "  value {\n",
      "    list {\n",
      "      shape {\n",
      "      }\n",
      "    }\n",
      "  }\n",
      "}\n",
      "\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "7/7 [==============================] - 88s 6s/step - loss: 0.3002 - accuracy: 0.8968 - val_loss: 0.3883 - val_accuracy: 0.8809\n",
      "Epoch 2/300\n",
      "7/7 [==============================] - 22s 3s/step - loss: 0.2579 - accuracy: 0.9149 - val_loss: 0.3741 - val_accuracy: 0.8865\n",
      "Epoch 3/300\n",
      "7/7 [==============================] - 23s 3s/step - loss: 0.2212 - accuracy: 0.9280 - val_loss: 0.3640 - val_accuracy: 0.8879\n",
      "Epoch 4/300\n",
      "7/7 [==============================] - 23s 3s/step - loss: 0.2203 - accuracy: 0.9295 - val_loss: 0.3674 - val_accuracy: 0.8922\n",
      "Epoch 5/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.1875 - accuracy: 0.9346 - val_loss: 0.3745 - val_accuracy: 0.8965\n",
      "Epoch 6/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.2082 - accuracy: 0.9313 - val_loss: 0.3632 - val_accuracy: 0.8879\n",
      "Epoch 7/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.1757 - accuracy: 0.9402 - val_loss: 0.3687 - val_accuracy: 0.8894\n",
      "Epoch 8/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.1721 - accuracy: 0.9420 - val_loss: 0.3657 - val_accuracy: 0.8936\n",
      "Epoch 9/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.1605 - accuracy: 0.9459 - val_loss: 0.3758 - val_accuracy: 0.8879\n",
      "Epoch 10/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.1696 - accuracy: 0.9441 - val_loss: 0.3627 - val_accuracy: 0.8936\n",
      "Epoch 11/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.1599 - accuracy: 0.9494 - val_loss: 0.3362 - val_accuracy: 0.9064\n",
      "Epoch 12/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.1766 - accuracy: 0.9390 - val_loss: 0.3621 - val_accuracy: 0.8851\n",
      "Epoch 13/300\n",
      "7/7 [==============================] - 24s 4s/step - loss: 0.1514 - accuracy: 0.9506 - val_loss: 0.3335 - val_accuracy: 0.9007\n",
      "Epoch 14/300\n",
      "7/7 [==============================] - 25s 4s/step - loss: 0.1405 - accuracy: 0.9506 - val_loss: 0.3566 - val_accuracy: 0.8950\n",
      "Epoch 15/300\n",
      "7/7 [==============================] - 23s 3s/step - loss: 0.1319 - accuracy: 0.9554 - val_loss: 0.3349 - val_accuracy: 0.9021\n",
      "Epoch 16/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.1312 - accuracy: 0.9566 - val_loss: 0.3728 - val_accuracy: 0.8950\n",
      "Epoch 17/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.1204 - accuracy: 0.9613 - val_loss: 0.3447 - val_accuracy: 0.9035\n",
      "Epoch 18/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.1171 - accuracy: 0.9595 - val_loss: 0.3636 - val_accuracy: 0.8979\n",
      "Epoch 19/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.1050 - accuracy: 0.9643 - val_loss: 0.3502 - val_accuracy: 0.9035\n",
      "Epoch 20/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.1144 - accuracy: 0.9637 - val_loss: 0.4195 - val_accuracy: 0.8979\n",
      "Epoch 21/300\n",
      "7/7 [==============================] - 24s 4s/step - loss: 0.1090 - accuracy: 0.9649 - val_loss: 0.3838 - val_accuracy: 0.8936\n",
      "Epoch 22/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.1041 - accuracy: 0.9694 - val_loss: 0.4290 - val_accuracy: 0.8879\n",
      "Epoch 23/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0988 - accuracy: 0.9685 - val_loss: 0.3647 - val_accuracy: 0.9035\n",
      "Epoch 24/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.1009 - accuracy: 0.9667 - val_loss: 0.3844 - val_accuracy: 0.8965\n",
      "Epoch 25/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0988 - accuracy: 0.9679 - val_loss: 0.3783 - val_accuracy: 0.8979\n",
      "Epoch 26/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0958 - accuracy: 0.9691 - val_loss: 0.3862 - val_accuracy: 0.9021\n",
      "Epoch 27/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0945 - accuracy: 0.9723 - val_loss: 0.3910 - val_accuracy: 0.8993\n",
      "Epoch 28/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0820 - accuracy: 0.9723 - val_loss: 0.3934 - val_accuracy: 0.8993\n",
      "Epoch 29/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0818 - accuracy: 0.9735 - val_loss: 0.4033 - val_accuracy: 0.8979\n",
      "Epoch 30/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0844 - accuracy: 0.9738 - val_loss: 0.3984 - val_accuracy: 0.8879\n",
      "Epoch 31/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0870 - accuracy: 0.9714 - val_loss: 0.3828 - val_accuracy: 0.9050\n",
      "Epoch 32/300\n",
      "7/7 [==============================] - 24s 4s/step - loss: 0.0672 - accuracy: 0.9765 - val_loss: 0.4041 - val_accuracy: 0.8965\n",
      "Epoch 33/300\n",
      "7/7 [==============================] - 25s 4s/step - loss: 0.0810 - accuracy: 0.9771 - val_loss: 0.3909 - val_accuracy: 0.9078\n",
      "Epoch 34/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0719 - accuracy: 0.9801 - val_loss: 0.3984 - val_accuracy: 0.8993\n",
      "Epoch 35/300\n",
      "7/7 [==============================] - 25s 4s/step - loss: 0.0578 - accuracy: 0.9801 - val_loss: 0.4420 - val_accuracy: 0.8837\n",
      "Epoch 36/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0631 - accuracy: 0.9792 - val_loss: 0.4338 - val_accuracy: 0.8879\n",
      "Epoch 37/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0667 - accuracy: 0.9768 - val_loss: 0.4485 - val_accuracy: 0.8908\n",
      "Epoch 38/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0582 - accuracy: 0.9810 - val_loss: 0.4616 - val_accuracy: 0.8979\n",
      "Epoch 39/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0554 - accuracy: 0.9827 - val_loss: 0.4523 - val_accuracy: 0.8965\n",
      "Epoch 40/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0528 - accuracy: 0.9825 - val_loss: 0.4565 - val_accuracy: 0.8950\n",
      "Epoch 41/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0474 - accuracy: 0.9860 - val_loss: 0.4597 - val_accuracy: 0.8965\n",
      "Epoch 42/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0529 - accuracy: 0.9830 - val_loss: 0.4497 - val_accuracy: 0.9035\n",
      "Epoch 43/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0653 - accuracy: 0.9801 - val_loss: 0.4283 - val_accuracy: 0.9050\n",
      "Epoch 44/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0708 - accuracy: 0.9786 - val_loss: 0.4228 - val_accuracy: 0.9078\n",
      "Epoch 45/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0609 - accuracy: 0.9819 - val_loss: 0.4579 - val_accuracy: 0.8965\n",
      "Epoch 46/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0543 - accuracy: 0.9819 - val_loss: 0.4422 - val_accuracy: 0.9064\n",
      "Epoch 47/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0506 - accuracy: 0.9869 - val_loss: 0.4326 - val_accuracy: 0.9064\n",
      "Epoch 48/300\n",
      "7/7 [==============================] - 24s 4s/step - loss: 0.0490 - accuracy: 0.9857 - val_loss: 0.4371 - val_accuracy: 0.9078\n",
      "Epoch 49/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0596 - accuracy: 0.9813 - val_loss: 0.4045 - val_accuracy: 0.9021\n",
      "Epoch 50/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0450 - accuracy: 0.9860 - val_loss: 0.4136 - val_accuracy: 0.9149\n",
      "Epoch 51/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0515 - accuracy: 0.9842 - val_loss: 0.3910 - val_accuracy: 0.9163\n",
      "Epoch 52/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0549 - accuracy: 0.9845 - val_loss: 0.4220 - val_accuracy: 0.9135\n",
      "Epoch 53/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0442 - accuracy: 0.9851 - val_loss: 0.4305 - val_accuracy: 0.9121\n",
      "Epoch 54/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0556 - accuracy: 0.9833 - val_loss: 0.4119 - val_accuracy: 0.9149\n",
      "Epoch 55/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0465 - accuracy: 0.9839 - val_loss: 0.4188 - val_accuracy: 0.9177\n",
      "Epoch 56/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0392 - accuracy: 0.9863 - val_loss: 0.4473 - val_accuracy: 0.8965\n",
      "Epoch 57/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0378 - accuracy: 0.9893 - val_loss: 0.4041 - val_accuracy: 0.9135\n",
      "Epoch 58/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0435 - accuracy: 0.9839 - val_loss: 0.4448 - val_accuracy: 0.9078\n",
      "Epoch 59/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0412 - accuracy: 0.9878 - val_loss: 0.4383 - val_accuracy: 0.9149\n",
      "Epoch 60/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0395 - accuracy: 0.9866 - val_loss: 0.4204 - val_accuracy: 0.9149\n",
      "Epoch 61/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0380 - accuracy: 0.9893 - val_loss: 0.4726 - val_accuracy: 0.9050\n",
      "Epoch 62/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0405 - accuracy: 0.9875 - val_loss: 0.4512 - val_accuracy: 0.9092\n",
      "Epoch 63/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0409 - accuracy: 0.9872 - val_loss: 0.5130 - val_accuracy: 0.9007\n",
      "Epoch 64/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0340 - accuracy: 0.9893 - val_loss: 0.4850 - val_accuracy: 0.9163\n",
      "Epoch 65/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0365 - accuracy: 0.9890 - val_loss: 0.4497 - val_accuracy: 0.9106\n",
      "Epoch 66/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0351 - accuracy: 0.9884 - val_loss: 0.4809 - val_accuracy: 0.9035\n",
      "Epoch 67/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0420 - accuracy: 0.9875 - val_loss: 0.4392 - val_accuracy: 0.9092\n",
      "Epoch 68/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0363 - accuracy: 0.9905 - val_loss: 0.4684 - val_accuracy: 0.9078\n",
      "Epoch 69/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0361 - accuracy: 0.9902 - val_loss: 0.4559 - val_accuracy: 0.9050\n",
      "Epoch 70/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0308 - accuracy: 0.9899 - val_loss: 0.4248 - val_accuracy: 0.9135\n",
      "Epoch 71/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0298 - accuracy: 0.9914 - val_loss: 0.4692 - val_accuracy: 0.9007\n",
      "Epoch 72/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0304 - accuracy: 0.9914 - val_loss: 0.4442 - val_accuracy: 0.9007\n",
      "Epoch 73/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0335 - accuracy: 0.9890 - val_loss: 0.4421 - val_accuracy: 0.9092\n",
      "Epoch 74/300\n",
      "7/7 [==============================] - 25s 4s/step - loss: 0.0266 - accuracy: 0.9926 - val_loss: 0.4493 - val_accuracy: 0.9163\n",
      "Epoch 75/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0308 - accuracy: 0.9908 - val_loss: 0.4286 - val_accuracy: 0.9092\n",
      "Epoch 76/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0302 - accuracy: 0.9917 - val_loss: 0.4679 - val_accuracy: 0.9121\n",
      "Epoch 77/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0314 - accuracy: 0.9923 - val_loss: 0.4688 - val_accuracy: 0.9064\n",
      "Epoch 78/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0331 - accuracy: 0.9887 - val_loss: 0.4644 - val_accuracy: 0.9121\n",
      "Epoch 79/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0219 - accuracy: 0.9938 - val_loss: 0.4905 - val_accuracy: 0.9078\n",
      "Epoch 80/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0256 - accuracy: 0.9917 - val_loss: 0.4640 - val_accuracy: 0.9092\n",
      "Epoch 81/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0288 - accuracy: 0.9914 - val_loss: 0.4943 - val_accuracy: 0.9092\n",
      "Epoch 82/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0312 - accuracy: 0.9899 - val_loss: 0.4814 - val_accuracy: 0.9121\n",
      "Epoch 83/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0241 - accuracy: 0.9935 - val_loss: 0.5019 - val_accuracy: 0.9064\n",
      "Epoch 84/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0195 - accuracy: 0.9932 - val_loss: 0.4809 - val_accuracy: 0.9078\n",
      "Epoch 85/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0248 - accuracy: 0.9917 - val_loss: 0.4720 - val_accuracy: 0.9121\n",
      "Epoch 86/300\n",
      "7/7 [==============================] - 25s 4s/step - loss: 0.0200 - accuracy: 0.9949 - val_loss: 0.4284 - val_accuracy: 0.9064\n",
      "Epoch 87/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0244 - accuracy: 0.9920 - val_loss: 0.5142 - val_accuracy: 0.9007\n",
      "Epoch 88/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0299 - accuracy: 0.9914 - val_loss: 0.4624 - val_accuracy: 0.9121\n",
      "Epoch 89/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0267 - accuracy: 0.9920 - val_loss: 0.4657 - val_accuracy: 0.9106\n",
      "Epoch 90/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0289 - accuracy: 0.9929 - val_loss: 0.4972 - val_accuracy: 0.9035\n",
      "Epoch 91/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0265 - accuracy: 0.9926 - val_loss: 0.4722 - val_accuracy: 0.9050\n",
      "Epoch 92/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0221 - accuracy: 0.9952 - val_loss: 0.4329 - val_accuracy: 0.9135\n",
      "Epoch 93/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0157 - accuracy: 0.9955 - val_loss: 0.4749 - val_accuracy: 0.9050\n",
      "Epoch 94/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0253 - accuracy: 0.9923 - val_loss: 0.4477 - val_accuracy: 0.9149\n",
      "Epoch 95/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0199 - accuracy: 0.9935 - val_loss: 0.5175 - val_accuracy: 0.9050\n",
      "Epoch 96/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0187 - accuracy: 0.9938 - val_loss: 0.4699 - val_accuracy: 0.9177\n",
      "Epoch 97/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0293 - accuracy: 0.9920 - val_loss: 0.4641 - val_accuracy: 0.9106\n",
      "Epoch 98/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0245 - accuracy: 0.9920 - val_loss: 0.5170 - val_accuracy: 0.9106\n",
      "Epoch 99/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0173 - accuracy: 0.9946 - val_loss: 0.5105 - val_accuracy: 0.9092\n",
      "Epoch 100/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0162 - accuracy: 0.9938 - val_loss: 0.4903 - val_accuracy: 0.9177\n",
      "Epoch 101/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0249 - accuracy: 0.9955 - val_loss: 0.5153 - val_accuracy: 0.9191\n",
      "Epoch 102/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0280 - accuracy: 0.9917 - val_loss: 0.4811 - val_accuracy: 0.9220\n",
      "Epoch 103/300\n",
      "7/7 [==============================] - 24s 4s/step - loss: 0.0195 - accuracy: 0.9943 - val_loss: 0.5071 - val_accuracy: 0.9163\n",
      "Epoch 104/300\n",
      "7/7 [==============================] - 25s 4s/step - loss: 0.0119 - accuracy: 0.9958 - val_loss: 0.4879 - val_accuracy: 0.9106\n",
      "Epoch 105/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0245 - accuracy: 0.9923 - val_loss: 0.4861 - val_accuracy: 0.9106\n",
      "Epoch 106/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0161 - accuracy: 0.9961 - val_loss: 0.4824 - val_accuracy: 0.9135\n",
      "Epoch 107/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0208 - accuracy: 0.9949 - val_loss: 0.4837 - val_accuracy: 0.9149\n",
      "Epoch 108/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0258 - accuracy: 0.9932 - val_loss: 0.5159 - val_accuracy: 0.9121\n",
      "Epoch 109/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0163 - accuracy: 0.9952 - val_loss: 0.5541 - val_accuracy: 0.9149\n",
      "Epoch 110/300\n",
      "7/7 [==============================] - 24s 4s/step - loss: 0.0180 - accuracy: 0.9946 - val_loss: 0.5205 - val_accuracy: 0.9135\n",
      "Epoch 111/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0258 - accuracy: 0.9929 - val_loss: 0.5278 - val_accuracy: 0.9121\n",
      "Epoch 112/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0169 - accuracy: 0.9943 - val_loss: 0.5129 - val_accuracy: 0.9106\n",
      "Epoch 113/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0244 - accuracy: 0.9935 - val_loss: 0.4940 - val_accuracy: 0.9050\n",
      "Epoch 114/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0213 - accuracy: 0.9958 - val_loss: 0.4796 - val_accuracy: 0.9135\n",
      "Epoch 115/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0220 - accuracy: 0.9938 - val_loss: 0.5079 - val_accuracy: 0.9149\n",
      "Epoch 116/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0170 - accuracy: 0.9935 - val_loss: 0.4975 - val_accuracy: 0.9149\n",
      "Epoch 117/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0175 - accuracy: 0.9949 - val_loss: 0.4702 - val_accuracy: 0.9191\n",
      "Epoch 118/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0170 - accuracy: 0.9946 - val_loss: 0.4747 - val_accuracy: 0.9206\n",
      "Epoch 119/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0177 - accuracy: 0.9938 - val_loss: 0.5195 - val_accuracy: 0.9177\n",
      "Epoch 120/300\n",
      "7/7 [==============================] - 25s 4s/step - loss: 0.0160 - accuracy: 0.9946 - val_loss: 0.5044 - val_accuracy: 0.9135\n",
      "Epoch 121/300\n",
      "7/7 [==============================] - 23s 3s/step - loss: 0.0188 - accuracy: 0.9938 - val_loss: 0.5293 - val_accuracy: 0.9106\n",
      "Epoch 122/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0174 - accuracy: 0.9943 - val_loss: 0.5263 - val_accuracy: 0.9177\n",
      "Epoch 123/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0297 - accuracy: 0.9923 - val_loss: 0.4720 - val_accuracy: 0.9092\n",
      "Epoch 124/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0210 - accuracy: 0.9929 - val_loss: 0.5230 - val_accuracy: 0.9064\n",
      "Epoch 125/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0148 - accuracy: 0.9955 - val_loss: 0.5034 - val_accuracy: 0.9092\n",
      "Epoch 126/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0185 - accuracy: 0.9958 - val_loss: 0.4450 - val_accuracy: 0.9121\n",
      "Epoch 127/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0172 - accuracy: 0.9949 - val_loss: 0.4366 - val_accuracy: 0.9177\n",
      "Epoch 128/300\n",
      "7/7 [==============================] - 24s 4s/step - loss: 0.0105 - accuracy: 0.9979 - val_loss: 0.4535 - val_accuracy: 0.9163\n",
      "Epoch 129/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0129 - accuracy: 0.9961 - val_loss: 0.4728 - val_accuracy: 0.9149\n",
      "Epoch 130/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0093 - accuracy: 0.9964 - val_loss: 0.5221 - val_accuracy: 0.9149\n",
      "Epoch 131/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0183 - accuracy: 0.9946 - val_loss: 0.4805 - val_accuracy: 0.9206\n",
      "Epoch 132/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0147 - accuracy: 0.9958 - val_loss: 0.4950 - val_accuracy: 0.9234\n",
      "Epoch 133/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0139 - accuracy: 0.9958 - val_loss: 0.5593 - val_accuracy: 0.9021\n",
      "Epoch 134/300\n",
      "7/7 [==============================] - 24s 4s/step - loss: 0.0219 - accuracy: 0.9946 - val_loss: 0.5667 - val_accuracy: 0.9092\n",
      "Epoch 135/300\n",
      "7/7 [==============================] - 25s 4s/step - loss: 0.0156 - accuracy: 0.9946 - val_loss: 0.5133 - val_accuracy: 0.9177\n",
      "Epoch 136/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0203 - accuracy: 0.9943 - val_loss: 0.5065 - val_accuracy: 0.9177\n",
      "Epoch 137/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0216 - accuracy: 0.9926 - val_loss: 0.5373 - val_accuracy: 0.9078\n",
      "Epoch 138/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0223 - accuracy: 0.9923 - val_loss: 0.4906 - val_accuracy: 0.9149\n",
      "Epoch 139/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0119 - accuracy: 0.9961 - val_loss: 0.5110 - val_accuracy: 0.9121\n",
      "Epoch 140/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0105 - accuracy: 0.9982 - val_loss: 0.5124 - val_accuracy: 0.9206\n",
      "Epoch 141/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0112 - accuracy: 0.9967 - val_loss: 0.4766 - val_accuracy: 0.9106\n",
      "Epoch 142/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0150 - accuracy: 0.9958 - val_loss: 0.4349 - val_accuracy: 0.9149\n",
      "Epoch 143/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0128 - accuracy: 0.9964 - val_loss: 0.4479 - val_accuracy: 0.9149\n",
      "Epoch 144/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0143 - accuracy: 0.9958 - val_loss: 0.5223 - val_accuracy: 0.9064\n",
      "Epoch 145/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0193 - accuracy: 0.9955 - val_loss: 0.4978 - val_accuracy: 0.9149\n",
      "Epoch 146/300\n",
      "7/7 [==============================] - 24s 4s/step - loss: 0.0123 - accuracy: 0.9952 - val_loss: 0.5278 - val_accuracy: 0.9007\n",
      "Epoch 147/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0118 - accuracy: 0.9949 - val_loss: 0.5355 - val_accuracy: 0.9106\n",
      "Epoch 148/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0167 - accuracy: 0.9955 - val_loss: 0.5118 - val_accuracy: 0.9106\n",
      "Epoch 149/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0172 - accuracy: 0.9943 - val_loss: 0.4759 - val_accuracy: 0.9121\n",
      "Epoch 150/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0147 - accuracy: 0.9958 - val_loss: 0.4772 - val_accuracy: 0.9135\n",
      "Epoch 151/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0116 - accuracy: 0.9958 - val_loss: 0.4887 - val_accuracy: 0.9163\n",
      "Epoch 152/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0106 - accuracy: 0.9970 - val_loss: 0.4613 - val_accuracy: 0.9191\n",
      "Epoch 153/300\n",
      "7/7 [==============================] - 25s 4s/step - loss: 0.0102 - accuracy: 0.9976 - val_loss: 0.4864 - val_accuracy: 0.9149\n",
      "Epoch 154/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0145 - accuracy: 0.9958 - val_loss: 0.5188 - val_accuracy: 0.9163\n",
      "Epoch 155/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0105 - accuracy: 0.9964 - val_loss: 0.5421 - val_accuracy: 0.9135\n",
      "Epoch 156/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0104 - accuracy: 0.9970 - val_loss: 0.4545 - val_accuracy: 0.9234\n",
      "Epoch 157/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0155 - accuracy: 0.9958 - val_loss: 0.4658 - val_accuracy: 0.9177\n",
      "Epoch 158/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0189 - accuracy: 0.9967 - val_loss: 0.5015 - val_accuracy: 0.9106\n",
      "Epoch 159/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0199 - accuracy: 0.9946 - val_loss: 0.4748 - val_accuracy: 0.9234\n",
      "Epoch 160/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0123 - accuracy: 0.9964 - val_loss: 0.4502 - val_accuracy: 0.9191\n",
      "Epoch 161/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0111 - accuracy: 0.9970 - val_loss: 0.4647 - val_accuracy: 0.9121\n",
      "Epoch 162/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0088 - accuracy: 0.9982 - val_loss: 0.4679 - val_accuracy: 0.9262\n",
      "Epoch 163/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0074 - accuracy: 0.9976 - val_loss: 0.4794 - val_accuracy: 0.9262\n",
      "Epoch 164/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0093 - accuracy: 0.9970 - val_loss: 0.4674 - val_accuracy: 0.9291\n",
      "Epoch 165/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0114 - accuracy: 0.9976 - val_loss: 0.4737 - val_accuracy: 0.9149\n",
      "Epoch 166/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0089 - accuracy: 0.9967 - val_loss: 0.5125 - val_accuracy: 0.9163\n",
      "Epoch 167/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0055 - accuracy: 0.9973 - val_loss: 0.5026 - val_accuracy: 0.9177\n",
      "Epoch 168/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0083 - accuracy: 0.9976 - val_loss: 0.5073 - val_accuracy: 0.9177\n",
      "Epoch 169/300\n",
      "7/7 [==============================] - 24s 4s/step - loss: 0.0067 - accuracy: 0.9985 - val_loss: 0.4897 - val_accuracy: 0.9121\n",
      "Epoch 170/300\n",
      "7/7 [==============================] - 25s 4s/step - loss: 0.0129 - accuracy: 0.9961 - val_loss: 0.5076 - val_accuracy: 0.9078\n",
      "Epoch 171/300\n",
      "7/7 [==============================] - 23s 3s/step - loss: 0.0060 - accuracy: 0.9982 - val_loss: 0.5164 - val_accuracy: 0.9177\n",
      "Epoch 172/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0102 - accuracy: 0.9982 - val_loss: 0.5431 - val_accuracy: 0.9149\n",
      "Epoch 173/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0083 - accuracy: 0.9973 - val_loss: 0.5278 - val_accuracy: 0.9149\n",
      "Epoch 174/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0135 - accuracy: 0.9964 - val_loss: 0.5275 - val_accuracy: 0.9220\n",
      "Epoch 175/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0059 - accuracy: 0.9982 - val_loss: 0.5365 - val_accuracy: 0.9135\n",
      "Epoch 176/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0097 - accuracy: 0.9967 - val_loss: 0.4914 - val_accuracy: 0.9234\n",
      "Epoch 177/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0099 - accuracy: 0.9964 - val_loss: 0.5082 - val_accuracy: 0.9206\n",
      "Epoch 178/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0159 - accuracy: 0.9952 - val_loss: 0.5084 - val_accuracy: 0.9177\n",
      "Epoch 179/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0149 - accuracy: 0.9955 - val_loss: 0.5635 - val_accuracy: 0.9078\n",
      "Epoch 180/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0136 - accuracy: 0.9967 - val_loss: 0.5332 - val_accuracy: 0.9149\n",
      "Epoch 181/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0135 - accuracy: 0.9970 - val_loss: 0.5182 - val_accuracy: 0.9121\n",
      "Epoch 182/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0175 - accuracy: 0.9955 - val_loss: 0.5628 - val_accuracy: 0.9135\n",
      "Epoch 183/300\n",
      "7/7 [==============================] - 25s 4s/step - loss: 0.0135 - accuracy: 0.9955 - val_loss: 0.5264 - val_accuracy: 0.9121\n",
      "Epoch 184/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0095 - accuracy: 0.9973 - val_loss: 0.5336 - val_accuracy: 0.9078\n",
      "Epoch 185/300\n",
      "7/7 [==============================] - 25s 4s/step - loss: 0.0156 - accuracy: 0.9958 - val_loss: 0.5609 - val_accuracy: 0.9121\n",
      "Epoch 186/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0141 - accuracy: 0.9964 - val_loss: 0.5483 - val_accuracy: 0.9135\n",
      "Epoch 187/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0127 - accuracy: 0.9964 - val_loss: 0.5410 - val_accuracy: 0.9078\n",
      "Epoch 188/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0134 - accuracy: 0.9961 - val_loss: 0.5678 - val_accuracy: 0.9092\n",
      "Epoch 189/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0064 - accuracy: 0.9979 - val_loss: 0.5407 - val_accuracy: 0.9149\n",
      "Epoch 190/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0044 - accuracy: 0.9985 - val_loss: 0.5332 - val_accuracy: 0.9163\n",
      "Epoch 191/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0099 - accuracy: 0.9973 - val_loss: 0.5606 - val_accuracy: 0.9064\n",
      "Epoch 192/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0107 - accuracy: 0.9976 - val_loss: 0.5347 - val_accuracy: 0.9092\n",
      "Epoch 193/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0049 - accuracy: 0.9988 - val_loss: 0.5483 - val_accuracy: 0.9106\n",
      "Epoch 194/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0083 - accuracy: 0.9976 - val_loss: 0.5443 - val_accuracy: 0.9092\n",
      "Epoch 195/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0078 - accuracy: 0.9970 - val_loss: 0.5270 - val_accuracy: 0.9092\n",
      "Epoch 196/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0095 - accuracy: 0.9970 - val_loss: 0.5124 - val_accuracy: 0.9106\n",
      "Epoch 197/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0118 - accuracy: 0.9967 - val_loss: 0.5190 - val_accuracy: 0.9135\n",
      "Epoch 198/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0083 - accuracy: 0.9973 - val_loss: 0.5167 - val_accuracy: 0.9149\n",
      "Epoch 199/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0056 - accuracy: 0.9982 - val_loss: 0.5052 - val_accuracy: 0.9149\n",
      "Epoch 200/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0082 - accuracy: 0.9979 - val_loss: 0.5076 - val_accuracy: 0.9177\n",
      "Epoch 201/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0037 - accuracy: 0.9988 - val_loss: 0.5100 - val_accuracy: 0.9177\n",
      "Epoch 202/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0035 - accuracy: 0.9991 - val_loss: 0.5481 - val_accuracy: 0.9177\n",
      "Epoch 203/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0077 - accuracy: 0.9982 - val_loss: 0.5854 - val_accuracy: 0.9121\n",
      "Epoch 204/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0094 - accuracy: 0.9976 - val_loss: 0.5626 - val_accuracy: 0.9135\n",
      "Epoch 205/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0158 - accuracy: 0.9967 - val_loss: 0.5745 - val_accuracy: 0.9191\n",
      "Epoch 206/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0075 - accuracy: 0.9985 - val_loss: 0.5612 - val_accuracy: 0.9149\n",
      "Epoch 207/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0042 - accuracy: 0.9988 - val_loss: 0.5546 - val_accuracy: 0.9149\n",
      "Epoch 208/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0146 - accuracy: 0.9958 - val_loss: 0.5210 - val_accuracy: 0.9149\n",
      "Epoch 209/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0108 - accuracy: 0.9976 - val_loss: 0.5134 - val_accuracy: 0.9163\n",
      "Epoch 210/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0101 - accuracy: 0.9979 - val_loss: 0.5659 - val_accuracy: 0.9149\n",
      "Epoch 211/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0158 - accuracy: 0.9964 - val_loss: 0.5104 - val_accuracy: 0.9135\n",
      "Epoch 212/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0047 - accuracy: 0.9988 - val_loss: 0.5047 - val_accuracy: 0.9163\n",
      "Epoch 213/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0136 - accuracy: 0.9976 - val_loss: 0.5667 - val_accuracy: 0.9092\n",
      "Epoch 214/300\n",
      "7/7 [==============================] - 24s 3s/step - loss: 0.0158 - accuracy: 0.9949 - val_loss: 0.6003 - val_accuracy: 0.9050\n"
     ]
    }
   ],
   "source": [
    "EPOCHS = 300\n",
    "\n",
    "\n",
    "with mirrored_strategy.scope():\n",
    "    # TRAINING PHASE \n",
    "    train_batches, valid_batches = prepare_batches(\n",
    "        TRAINING_DATA_DIR, VALID_DATA_DIR, IMAGE_SHAPE, 512\n",
    "    )    \n",
    "    \n",
    "    \n",
    "    model_fined = tf.keras.models.load_model(f\"../input/convexnet90/models/ConvexNet_transfer_learning_Best\")\n",
    "    \n",
    "    # find the index in the model referring to the convexnext part.\n",
    "    fe_index = None\n",
    "    for ind, l in enumerate(model_fined.layers):\n",
    "        if type(l) == keras.engine.functional.Functional:\n",
    "            fe_index = ind\n",
    "\n",
    "    freeze_from = 150\n",
    "    print(f\"freeze from {freeze_from}\")\n",
    "    \n",
    "    # unfreeze layers\n",
    "    model_fined.layers[fe_index].trainable=True\n",
    "    \n",
    "    # freeze remainig layers \n",
    "    for l in model_fined.layers[fe_index].layers[:freeze_from]:\n",
    "        l.trainable = False\n",
    "        \n",
    "    # freeze all batch normalization layers\n",
    "    for l in model_fined.layers[fe_index].layers:\n",
    "        if type(l) == tfkl.BatchNormalization:\n",
    "            l.trainable = False\n",
    "    \n",
    "    # Compile the model\n",
    "    model_fined.compile(loss=tfk.losses.CategoricalCrossentropy(), optimizer=tfk.optimizers.Adam(1e-4), metrics='accuracy')\n",
    "    \n",
    "    early_stopping = tf.keras.callbacks.EarlyStopping(patience=50, monitor=\"val_accuracy\", restore_best_weights=True)\n",
    "    history = model_fined.fit(\n",
    "        train_batches,\n",
    "        epochs=EPOCHS,\n",
    "        validation_data=valid_batches,\n",
    "        verbose=1,\n",
    "        workers=4,\n",
    "        use_multiprocessing=True,\n",
    "        callbacks=[early_stopping]\n",
    "    )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-23T16:53:04.969540Z",
     "iopub.status.busy": "2022-11-23T16:53:04.968770Z",
     "iopub.status.idle": "2022-11-23T16:53:05.351769Z",
     "shell.execute_reply": "2022-11-23T16:53:05.350748Z",
     "shell.execute_reply.started": "2022-11-23T16:53:04.969494Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.9290780425071716 0.9991076588630676\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 864x432 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "x_axis = len(history.history[\"accuracy\"])\n",
    "print(max(history.history[\"val_accuracy\"]), max(history.history[\"accuracy\"]))\n",
    "history.history.keys()\n",
    "plt.figure(figsize=(12, 6))\n",
    "plt.plot([i for i in range(x_axis)], history.history[\"accuracy\"],  color=\"blue\", label=\"train accuracy\")\n",
    "plt.plot([i for i in range(x_axis)], history.history[\"val_accuracy\"], color=\"red\", label=\"validation accuracy\")\n",
    "plt.plot([np.argmax(history.history[\"val_accuracy\"])],\n",
    "         [max(history.history[\"val_accuracy\"])],\n",
    "         marker=\"o\",\n",
    "         markeredgecolor=\"red\",\n",
    "         markerfacecolor=\"red\",\n",
    "         label=\"best model\",\n",
    "         markersize=10\n",
    "        )\n",
    "plt.axhline(y=0.88, color='green', linestyle='-')\n",
    "plt.legend()\n",
    "plt.grid()\n",
    "plt.xlabel(\"Epochs\")\n",
    "plt.ylabel(\"Accuracy\")\n",
    "minimum = min(min(history.history[\"accuracy\"]), min(history.history[\"val_accuracy\"]))\n",
    "plt.ylim([minimum, 1])\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-23T16:53:15.345977Z",
     "iopub.status.busy": "2022-11-23T16:53:15.345370Z",
     "iopub.status.idle": "2022-11-23T16:53:29.033841Z",
     "shell.execute_reply": "2022-11-23T16:53:29.032698Z",
     "shell.execute_reply.started": "2022-11-23T16:53:15.345939Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2022-11-23 16:53:15.366781: W tensorflow/core/grappler/optimizers/data/auto_shard.cc:695] AUTO sharding policy will apply DATA sharding policy as it failed to apply FILE sharding policy because of the following reason: Found an unshardable source dataset: name: \"TensorSliceDataset/_1\"\n",
      "op: \"TensorSliceDataset\"\n",
      "input: \"Placeholder/_0\"\n",
      "attr {\n",
      "  key: \"Toutput_types\"\n",
      "  value {\n",
      "    list {\n",
      "      type: DT_STRING\n",
      "    }\n",
      "  }\n",
      "}\n",
      "attr {\n",
      "  key: \"output_shapes\"\n",
      "  value {\n",
      "    list {\n",
      "      shape {\n",
      "      }\n",
      "    }\n",
      "  }\n",
      "}\n",
      "\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[]"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x504 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(8, 7))\n",
    "df = build_heatmap(model_fined, valid_batches)\n",
    "sns.heatmap(df, annot=True, cmap=\"crest\")\n",
    "plt.plot()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-23T17:10:00.018663Z",
     "iopub.status.busy": "2022-11-23T17:10:00.018281Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2022-11-23 17:10:12.703328: W tensorflow/python/util/util.cc:348] Sets are not currently considered sequences, but this may change in the future, so consider avoiding using them.\n"
     ]
    }
   ],
   "source": [
    "model_fined.save(f\"models/ConvexNet_fined_Best\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Confusion matrix for self-ensemble method.\n",
    "It does not change anything..."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "execution": {
     "iopub.execute_input": "2022-11-23T16:53:43.425557Z",
     "iopub.status.busy": "2022-11-23T16:53:43.424474Z",
     "iopub.status.idle": "2022-11-23T16:53:55.801203Z",
     "shell.execute_reply": "2022-11-23T16:53:55.800005Z",
     "shell.execute_reply.started": "2022-11-23T16:53:43.425482Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2022-11-23 16:53:43.458328: W tensorflow/core/grappler/optimizers/data/auto_shard.cc:695] AUTO sharding policy will apply DATA sharding policy as it failed to apply FILE sharding policy because of the following reason: Found an unshardable source dataset: name: \"TensorSliceDataset/_1\"\n",
      "op: \"TensorSliceDataset\"\n",
      "input: \"Placeholder/_0\"\n",
      "attr {\n",
      "  key: \"Toutput_types\"\n",
      "  value {\n",
      "    list {\n",
      "      type: DT_STRING\n",
      "    }\n",
      "  }\n",
      "}\n",
      "attr {\n",
      "  key: \"output_shapes\"\n",
      "  value {\n",
      "    list {\n",
      "      shape {\n",
      "      }\n",
      "    }\n",
      "  }\n",
      "}\n",
      "\n",
      "2022-11-23 16:53:45.930910: W tensorflow/core/grappler/optimizers/data/auto_shard.cc:695] AUTO sharding policy will apply DATA sharding policy as it failed to apply FILE sharding policy because of the following reason: Found an unshardable source dataset: name: \"TensorSliceDataset/_1\"\n",
      "op: \"TensorSliceDataset\"\n",
      "input: \"Placeholder/_0\"\n",
      "attr {\n",
      "  key: \"Toutput_types\"\n",
      "  value {\n",
      "    list {\n",
      "      type: DT_STRING\n",
      "    }\n",
      "  }\n",
      "}\n",
      "attr {\n",
      "  key: \"output_shapes\"\n",
      "  value {\n",
      "    list {\n",
      "      shape {\n",
      "      }\n",
      "    }\n",
      "  }\n",
      "}\n",
      "\n",
      "2022-11-23 16:53:48.240621: W tensorflow/core/grappler/optimizers/data/auto_shard.cc:695] AUTO sharding policy will apply DATA sharding policy as it failed to apply FILE sharding policy because of the following reason: Found an unshardable source dataset: name: \"TensorSliceDataset/_1\"\n",
      "op: \"TensorSliceDataset\"\n",
      "input: \"Placeholder/_0\"\n",
      "attr {\n",
      "  key: \"Toutput_types\"\n",
      "  value {\n",
      "    list {\n",
      "      type: DT_STRING\n",
      "    }\n",
      "  }\n",
      "}\n",
      "attr {\n",
      "  key: \"output_shapes\"\n",
      "  value {\n",
      "    list {\n",
      "      shape {\n",
      "      }\n",
      "    }\n",
      "  }\n",
      "}\n",
      "\n",
      "2022-11-23 16:53:50.571250: W tensorflow/core/grappler/optimizers/data/auto_shard.cc:695] AUTO sharding policy will apply DATA sharding policy as it failed to apply FILE sharding policy because of the following reason: Found an unshardable source dataset: name: \"TensorSliceDataset/_1\"\n",
      "op: \"TensorSliceDataset\"\n",
      "input: \"Placeholder/_0\"\n",
      "attr {\n",
      "  key: \"Toutput_types\"\n",
      "  value {\n",
      "    list {\n",
      "      type: DT_STRING\n",
      "    }\n",
      "  }\n",
      "}\n",
      "attr {\n",
      "  key: \"output_shapes\"\n",
      "  value {\n",
      "    list {\n",
      "      shape {\n",
      "      }\n",
      "    }\n",
      "  }\n",
      "}\n",
      "\n",
      "2022-11-23 16:53:52.875252: W tensorflow/core/grappler/optimizers/data/auto_shard.cc:695] AUTO sharding policy will apply DATA sharding policy as it failed to apply FILE sharding policy because of the following reason: Found an unshardable source dataset: name: \"TensorSliceDataset/_1\"\n",
      "op: \"TensorSliceDataset\"\n",
      "input: \"Placeholder/_0\"\n",
      "attr {\n",
      "  key: \"Toutput_types\"\n",
      "  value {\n",
      "    list {\n",
      "      type: DT_STRING\n",
      "    }\n",
      "  }\n",
      "}\n",
      "attr {\n",
      "  key: \"output_shapes\"\n",
      "  value {\n",
      "    list {\n",
      "      shape {\n",
      "      }\n",
      "    }\n",
      "  }\n",
      "}\n",
      "\n",
      "2022-11-23 16:53:55.216932: W tensorflow/core/data/root_dataset.cc:167] Optimization loop failed: Cancelled: Operation was cancelled\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "[]"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x504 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(8, 7))\n",
    "df = build_heatmap(model_fined, valid_batches, ensemble=True)\n",
    "sns.heatmap(df, annot=True, cmap=\"crest\")\n",
    "plt.plot()"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}