Long_Model.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "id": "Q3YV78dzD-MD",
    "outputId": "c95a2c88-1ff7-43ec-9203-15e07b73b848"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Populating the interactive namespace from numpy and matplotlib\n"
     ]
    }
   ],
   "source": [
    "import os\n",
    "%pylab inline\n",
    "from skimage.io import imread\n",
    "from skimage import exposure\n",
    "from skimage.transform import resize\n",
    "from skimage.exposure import adjust_gamma\n",
    "import numpy as np\n",
    "import tensorflow as tf\n",
    "from skimage.measure import label, regionprops\n",
    "import numpy as np\n",
    "import sys\n",
    "from scipy.stats import mode\n",
    "from matplotlib import pyplot as plt\n",
    "import tensorflow as tf\n",
    "from scipy.stats import mode"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "iX1MPG92KLZ4"
   },
   "outputs": [],
   "source": [
    "# Data augmentation for batches\n",
    "class Data_Augmentation():\n",
    "    def __init__(self, batch_size, validation_size, directory, train_test='train'):\n",
    "        self.batch_size = batch_size\n",
    "        self.validation_size = validation_size\n",
    "        self.directory = directory\n",
    "\n",
    "        nPerSet = {'train': 85, 'testA': 60, 'testB': 20}\n",
    "        self.image_files = [os.path.join(self.directory, f'{train_test}_{i}.bmp') for i in range(1, nPerSet[train_test]+1)]\n",
    "        self.annotation_files = [os.path.join(self.directory, f'{train_test}_{i}_anno.bmp') for i in range(1, nPerSet[train_test]+1)]\n",
    "\n",
    "        # load images\n",
    "        self.full_images = [imread(f)/255 for f in self.image_files]\n",
    "        self.full_annotations = [imread(f) for f in self.annotation_files]\n",
    "\n",
    "        # Train-valid split\n",
    "        self.idxs = np.arange(len(self.image_files))\n",
    "        np.random.seed(1)\n",
    "        np.random.shuffle(self.idxs)\n",
    "        self.val_idxs = self.idxs[:self.validation_size]\n",
    "        self.train_idxs = self.idxs[self.validation_size:]\n",
    "\n",
    "    @staticmethod # Data Augmentation\n",
    "    def _augment(batch_x, batch_y):\n",
    "        # vertical and horizontal symmetry\n",
    "        if( np.random.random()<0.5 ):\n",
    "            batch_x = batch_x[:,::-1,:,:]\n",
    "            batch_y = batch_y[:,::-1,:]\n",
    "        if( np.random.random()<0.5 ):\n",
    "            batch_x = batch_x[:,:,::-1,:]\n",
    "            batch_y = batch_y[:,:,::-1]\n",
    "        # Gamma\n",
    "        gamma = (np.random.random()-0.5)*2\n",
    "        if gamma < 0:\n",
    "            gamma=1/(1-gamma)\n",
    "        else:\n",
    "            gamma=1+gamma\n",
    "        batch_x_ = batch_x.copy()\n",
    "        for i in range(len(batch_x)):\n",
    "            batch_x_[i] = adjust_gamma(batch_x[i], gamma=gamma)\n",
    "        # Random noise\n",
    "        batch_x_ += np.random.normal(0, 0.02, size=batch_x.shape)\n",
    "        return batch_x_,batch_y"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "opI808HfLB2p"
   },
   "outputs": [],
   "source": [
    "class imageGeneration(Data_Augmentation):\n",
    "    # read train and valid sets\n",
    "    def __init__(self, batch_size, validation_size, directory, image_size, train_test='train', **kwargs):\n",
    "        super().__init__(batch_size, validation_size, directory, train_test)\n",
    "\n",
    "        self.image_size = image_size\n",
    "        self.batches_per_epoch = len(self.train_idxs)//self.batch_size\n",
    "\n",
    "        self.val_x = np.array([self.preprocess_image(self.full_images[idx]) for idx in self.val_idxs])\n",
    "        self.val_y = np.array([self.preprocess_anno(self.full_annotations[idx]) for idx in self.val_idxs])\n",
    "\n",
    "        self.train_x = np.array([self.preprocess_image(self.full_images[idx]) for idx in self.train_idxs])\n",
    "        self.train_y = np.array([self.preprocess_anno(self.full_annotations[idx]) for idx in self.train_idxs])\n",
    "\n",
    "    # Random shuffle of images\n",
    "    def next_batch(self, n_epochs):\n",
    "        for e in range(n_epochs):\n",
    "            np.random.shuffle(self.train_idxs)\n",
    "            for i in range(self.batches_per_epoch):\n",
    "                batch_x = self.train_x[i*self.batch_size:(i+1)*self.batch_size]\n",
    "                batch_y = self.train_y[i*self.batch_size:(i+1)*self.batch_size]>0\n",
    "                yield self._augment(batch_x,batch_y)\n",
    "    def get_validation_data(self):\n",
    "        return self.val_x, self.val_y>0\n",
    "\n",
    "    def preprocess_image(self, im):\n",
    "        return resize(im, self.image_size)\n",
    "\n",
    "    def preprocess_anno(self, anno):\n",
    "        return resize(anno, self.image_size, preserve_range=True, order=0, anti_aliasing=False) "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "lLhdFZywONNN"
   },
   "outputs": [],
   "source": [
    "# Model training / Optimization / Post-processing\n",
    "class Model():\n",
    "    def __init__(self, image_size, clf_name, loadFrom=None, lr=1e-4, eps=1e-8):\n",
    "        self.image_size = image_size\n",
    "        self.clf_name = clf_name\n",
    "        if(loadFrom == None):\n",
    "            self.set_model()\n",
    "            opt = tf.keras.optimizers.Adam(learning_rate=lr, epsilon=eps, name='Adam')\n",
    "            self.model.compile(optimizer=opt, loss=tf.keras.losses.SparseCategoricalCrossentropy(), metrics=[tf.keras.losses.SparseCategoricalCrossentropy(name='crossentropy'), 'accuracy'])\n",
    "        else:\n",
    "            self.model = tf.keras.models.load_model(loadFrom, compile=False, custom_objects={'leaky_relu': tf.nn.leaky_relu})\n",
    "\n",
    "    def fit(self, n_epochs, dataset, patience=15):\n",
    "        validation_data = None\n",
    "        callbacks = []\n",
    "        if(len(dataset.val_idxs) > 0):\n",
    "            callbacks = [tf.keras.callbacks.ModelCheckpoint(f\"{self.clf_name}.hdf5\", save_best_only=True),tf.keras.callbacks.EarlyStopping(monitor='val_crossentropy', patience=patience)]\n",
    "            validation_data = dataset.get_validation_data()\n",
    "\n",
    "        return self.model.fit(dataset.next_batch(n_epochs), epochs=n_epochs, steps_per_epoch=dataset.batches_per_epoch, validation_data=validation_data, callbacks=callbacks)\n",
    "\n",
    "    def predict(self, data):\n",
    "        return self.model.predict(data)\n",
    "\n",
    "    # post-processing\n",
    "    @staticmethod\n",
    "    def post_process(pred, min_area=400):\n",
    "        pred_mask = np.argmax(pred, axis=2)\n",
    "        # labeling\n",
    "        lab = label(pred_mask)\n",
    "        # remove small objects\n",
    "        for obj in regionprops(lab):\n",
    "            if( obj.area < min_area ):\n",
    "                lab[lab==obj.label] = 0\n",
    "            # hole filling\n",
    "            else:\n",
    "                region = lab[obj.bbox[0]:obj.bbox[2],obj.bbox[1]:obj.bbox[3]]\n",
    "                region[obj.filled_image] = obj.label\n",
    "        return lab"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "ukzqIwcpPIZj"
   },
   "outputs": [],
   "source": [
    "class evaluation:\n",
    "\n",
    "    @classmethod\n",
    "    def evaluate(cls, model, data, train_val, overlap='minimum', min_area=40):\n",
    "\n",
    "        metrics = []\n",
    "        if train_val == 'train':\n",
    "            data_x = [data.full_images[idx] for idx in data.train_idxs]\n",
    "            data_y = [data.full_annotations[idx] for idx in data.train_idxs]\n",
    "        elif train_val == 'val':\n",
    "            data_x = [data.full_images[idx] for idx in data.val_idxs]\n",
    "            data_y = [data.full_annotations[idx] for idx in data.val_idxs]\n",
    "\n",
    "        # predict\n",
    "        for im, anno in zip(data_x, data_y):\n",
    "            im_ = data.preprocess_image(im)\n",
    "            pred_image = resize(model.predict(np.array([im_]))[0], im.shape[:2])\n",
    "            pred_labels = model.post_process(pred_image, min_area)\n",
    "            metrics += [cls._get_metrics(anno, pred_labels)]\n",
    "        return np.array(metrics)\n",
    "\n",
    "    @staticmethod\n",
    "    def _get_metrics(gt_labels, pred_labels):\n",
    "        # remove pixels with area smaller than 5\n",
    "        for i in range(1, gt_labels.max()+1):\n",
    "            if( (gt_labels==i).sum() < 5 ):\n",
    "                gt_labels[gt_labels==i] = 0\n",
    "\n",
    "        trueLabels = np.unique(gt_labels)\n",
    "        trueLabels = trueLabels[trueLabels>0].astype('int')\n",
    "        predLabels = np.unique(pred_labels)\n",
    "        predLabels = predLabels[predLabels>0].astype('int')\n",
    "\n",
    "        best_matches = np.zeros((len(predLabels),3))\n",
    "        best_matches[:,0] = predLabels\n",
    "        for i in range(len(predLabels)):\n",
    "            predObject = pred_labels==predLabels[i]\n",
    "            corrRegionInGT = gt_labels[predObject]  \n",
    "            if corrRegionInGT.max() > 0:\n",
    "                bestMatch = mode(corrRegionInGT[corrRegionInGT>0])[0][0] \n",
    "                matchInGT = gt_labels==bestMatch   \n",
    "                best_matches[i,1] = bestMatch       \n",
    "                overlap = predObject*matchInGT \n",
    "                best_matches[i,2] = (overlap.sum()/matchInGT.sum())>0.2 # % of overlap \n",
    "        \n",
    "        TP = int(best_matches[:,2].sum())\n",
    "        FP = int((best_matches[:,2]==0).sum())\n",
    "        FN = int(len(trueLabels)-TP)\n",
    "        \n",
    "        # precision & recall\n",
    "        precision = TP/(TP+FP)\n",
    "        recall = TP/(TP+FN)\n",
    "        \n",
    "        gt_mask = gt_labels>0\n",
    "        pred_mask = pred_labels>0\n",
    "        TP = float(((gt_mask==True)*(pred_mask==True)).sum())\n",
    "        FP = float(((gt_mask==False)*(pred_mask==True)).sum())\n",
    "        FN = float(((gt_mask==True)*(pred_mask==False)).sum())\n",
    "        TN = float(((gt_mask==False)*(pred_mask==False)).sum())\n",
    "        # segmentation MCC\n",
    "        MCC = ((TP*TN)-(FP*FN))/np.sqrt((TP+FP)*(TP+FN)*(TN+FP)*(TN+FN))\n",
    "        return precision, recall, MCC"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "5q21ifb9lTKS"
   },
   "outputs": [],
   "source": [
    "class Long_Model(Model):\n",
    "    def __init__(self, image_size, clf_name, loadFrom=None, lr=1e-4, eps=1e-8):\n",
    "        super().__init__(image_size, clf_name, loadFrom=loadFrom, lr=lr, eps=eps)\n",
    "\n",
    "    def set_model(self):\n",
    "        inputs = tf.keras.Input(shape=self.image_size+(3,))\n",
    "        x = tf.keras.layers.Conv2D(32, 3, activation=tf.nn.leaky_relu, padding='same')(inputs)\n",
    "        A = tf.keras.layers.Conv2D(32, 3, activation=tf.nn.leaky_relu, padding='same')(x)\n",
    "        x_ = tf.keras.layers.MaxPool2D(2)(A)\n",
    "        x = tf.keras.layers.Conv2D(64, 3, activation=tf.nn.leaky_relu, padding='same')(x_)\n",
    "        x = tf.keras.layers.Conv2D(64, 3, activation=tf.nn.leaky_relu, padding='same')(x)\n",
    "        B = tf.concat([x_, x], axis=3)\n",
    "        x_ = tf.keras.layers.MaxPool2D(2)(B)\n",
    "        x = tf.keras.layers.Conv2D(128, 3, activation=tf.nn.leaky_relu, padding='same')(x_)\n",
    "        x = tf.keras.layers.Conv2D(128, 3, activation=tf.nn.leaky_relu, padding='same')(x)\n",
    "        x = tf.concat([x_, x], axis=3)\n",
    "        x = tf.keras.layers.UpSampling2D(2)(x)\n",
    "        x = tf.concat([B, x], axis=3)\n",
    "        x = tf.keras.layers.Conv2D(64, 3, activation=tf.nn.leaky_relu, padding='same')(x)\n",
    "        x = tf.keras.layers.UpSampling2D(2)(x)\n",
    "        x = tf.concat([A, x], axis=3)\n",
    "        x = tf.keras.layers.Conv2D(32, 3, activation=tf.nn.leaky_relu, padding='same')(x)\n",
    "        outputs = tf.keras.layers.Conv2D(2, 1, activation=tf.nn.softmax)(x)\n",
    "\n",
    "        self.model = tf.keras.Model(inputs=inputs, outputs=outputs)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 1000
    },
    "id": "W2vTEA2BzRy-",
    "outputId": "a040129d-f4ba-4a74-da35-67855d07ea23"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "15/15 [==============================] - 4s 224ms/step - loss: 0.6914 - crossentropy: 0.6914 - accuracy: 0.5309 - val_loss: 0.6981 - val_crossentropy: 0.6981 - val_accuracy: 0.4838\n",
      "Epoch 2/100\n",
      "15/15 [==============================] - 3s 216ms/step - loss: 0.6938 - crossentropy: 0.6938 - accuracy: 0.5062 - val_loss: 0.6897 - val_crossentropy: 0.6897 - val_accuracy: 0.5136\n",
      "Epoch 3/100\n",
      "15/15 [==============================] - 3s 211ms/step - loss: 0.6867 - crossentropy: 0.6867 - accuracy: 0.5604 - val_loss: 0.6867 - val_crossentropy: 0.6867 - val_accuracy: 0.5267\n",
      "Epoch 4/100\n",
      "15/15 [==============================] - 3s 216ms/step - loss: 0.6831 - crossentropy: 0.6831 - accuracy: 0.5755 - val_loss: 0.7009 - val_crossentropy: 0.7009 - val_accuracy: 0.5073\n",
      "Epoch 5/100\n",
      "15/15 [==============================] - 3s 214ms/step - loss: 0.6852 - crossentropy: 0.6852 - accuracy: 0.5560 - val_loss: 0.6900 - val_crossentropy: 0.6900 - val_accuracy: 0.5245\n",
      "Epoch 6/100\n",
      "15/15 [==============================] - 3s 217ms/step - loss: 0.6853 - crossentropy: 0.6853 - accuracy: 0.5608 - val_loss: 0.6781 - val_crossentropy: 0.6781 - val_accuracy: 0.5858\n",
      "Epoch 7/100\n",
      "15/15 [==============================] - 3s 212ms/step - loss: 0.6675 - crossentropy: 0.6675 - accuracy: 0.6202 - val_loss: 0.6727 - val_crossentropy: 0.6727 - val_accuracy: 0.6072\n",
      "Epoch 8/100\n",
      "15/15 [==============================] - 3s 220ms/step - loss: 0.6619 - crossentropy: 0.6619 - accuracy: 0.6006 - val_loss: 0.6812 - val_crossentropy: 0.6812 - val_accuracy: 0.5409\n",
      "Epoch 9/100\n",
      "15/15 [==============================] - 3s 210ms/step - loss: 0.6608 - crossentropy: 0.6608 - accuracy: 0.6200 - val_loss: 0.6650 - val_crossentropy: 0.6650 - val_accuracy: 0.5977\n",
      "Epoch 10/100\n",
      "15/15 [==============================] - 3s 214ms/step - loss: 0.6580 - crossentropy: 0.6580 - accuracy: 0.6136 - val_loss: 0.6557 - val_crossentropy: 0.6557 - val_accuracy: 0.6165\n",
      "Epoch 11/100\n",
      "15/15 [==============================] - 3s 214ms/step - loss: 0.6485 - crossentropy: 0.6485 - accuracy: 0.6201 - val_loss: 0.6602 - val_crossentropy: 0.6602 - val_accuracy: 0.5907\n",
      "Epoch 12/100\n",
      "15/15 [==============================] - 3s 218ms/step - loss: 0.6370 - crossentropy: 0.6370 - accuracy: 0.6403 - val_loss: 0.6463 - val_crossentropy: 0.6463 - val_accuracy: 0.6222\n",
      "Epoch 13/100\n",
      "15/15 [==============================] - 3s 215ms/step - loss: 0.6316 - crossentropy: 0.6316 - accuracy: 0.6362 - val_loss: 0.6469 - val_crossentropy: 0.6469 - val_accuracy: 0.6246\n",
      "Epoch 14/100\n",
      "15/15 [==============================] - 3s 214ms/step - loss: 0.6258 - crossentropy: 0.6258 - accuracy: 0.6454 - val_loss: 0.6259 - val_crossentropy: 0.6259 - val_accuracy: 0.6628\n",
      "Epoch 15/100\n",
      "15/15 [==============================] - 3s 215ms/step - loss: 0.6149 - crossentropy: 0.6149 - accuracy: 0.6513 - val_loss: 0.6425 - val_crossentropy: 0.6425 - val_accuracy: 0.6095\n",
      "Epoch 16/100\n",
      "15/15 [==============================] - 3s 213ms/step - loss: 0.6479 - crossentropy: 0.6479 - accuracy: 0.6191 - val_loss: 0.6325 - val_crossentropy: 0.6325 - val_accuracy: 0.6210\n",
      "Epoch 17/100\n",
      "15/15 [==============================] - 3s 216ms/step - loss: 0.6157 - crossentropy: 0.6157 - accuracy: 0.6605 - val_loss: 0.6124 - val_crossentropy: 0.6124 - val_accuracy: 0.6680\n",
      "Epoch 18/100\n",
      "15/15 [==============================] - 3s 216ms/step - loss: 0.6195 - crossentropy: 0.6195 - accuracy: 0.6506 - val_loss: 0.6512 - val_crossentropy: 0.6512 - val_accuracy: 0.6085\n",
      "Epoch 19/100\n",
      "15/15 [==============================] - 3s 215ms/step - loss: 0.6008 - crossentropy: 0.6008 - accuracy: 0.6645 - val_loss: 0.6398 - val_crossentropy: 0.6398 - val_accuracy: 0.6107\n",
      "Epoch 20/100\n",
      "15/15 [==============================] - 3s 211ms/step - loss: 0.5897 - crossentropy: 0.5897 - accuracy: 0.6763 - val_loss: 0.6154 - val_crossentropy: 0.6154 - val_accuracy: 0.6555\n",
      "Epoch 21/100\n",
      "15/15 [==============================] - 3s 212ms/step - loss: 0.5842 - crossentropy: 0.5842 - accuracy: 0.6863 - val_loss: 0.6177 - val_crossentropy: 0.6177 - val_accuracy: 0.6402\n",
      "Epoch 22/100\n",
      "15/15 [==============================] - 3s 214ms/step - loss: 0.5834 - crossentropy: 0.5834 - accuracy: 0.6793 - val_loss: 0.5864 - val_crossentropy: 0.5864 - val_accuracy: 0.6879\n",
      "Epoch 23/100\n",
      "15/15 [==============================] - 3s 213ms/step - loss: 0.5766 - crossentropy: 0.5766 - accuracy: 0.6848 - val_loss: 0.6060 - val_crossentropy: 0.6060 - val_accuracy: 0.6753\n",
      "Epoch 24/100\n",
      "15/15 [==============================] - 3s 215ms/step - loss: 0.5843 - crossentropy: 0.5843 - accuracy: 0.6803 - val_loss: 0.6141 - val_crossentropy: 0.6141 - val_accuracy: 0.6402\n",
      "Epoch 25/100\n",
      "15/15 [==============================] - 3s 214ms/step - loss: 0.6040 - crossentropy: 0.6040 - accuracy: 0.6663 - val_loss: 0.6270 - val_crossentropy: 0.6270 - val_accuracy: 0.6212\n",
      "Epoch 26/100\n",
      "15/15 [==============================] - 3s 213ms/step - loss: 0.5854 - crossentropy: 0.5854 - accuracy: 0.6881 - val_loss: 0.6122 - val_crossentropy: 0.6122 - val_accuracy: 0.6572\n",
      "Epoch 27/100\n",
      "15/15 [==============================] - 3s 214ms/step - loss: 0.5882 - crossentropy: 0.5882 - accuracy: 0.6797 - val_loss: 0.6200 - val_crossentropy: 0.6200 - val_accuracy: 0.6354\n",
      "Epoch 28/100\n",
      "15/15 [==============================] - 3s 216ms/step - loss: 0.5737 - crossentropy: 0.5737 - accuracy: 0.6885 - val_loss: 0.5860 - val_crossentropy: 0.5860 - val_accuracy: 0.6805\n",
      "Epoch 29/100\n",
      "15/15 [==============================] - 3s 210ms/step - loss: 0.5617 - crossentropy: 0.5617 - accuracy: 0.7060 - val_loss: 0.5778 - val_crossentropy: 0.5778 - val_accuracy: 0.7014\n",
      "Epoch 30/100\n",
      "15/15 [==============================] - 3s 210ms/step - loss: 0.5606 - crossentropy: 0.5606 - accuracy: 0.7071 - val_loss: 0.5644 - val_crossentropy: 0.5644 - val_accuracy: 0.6956\n",
      "Epoch 31/100\n",
      "15/15 [==============================] - 3s 217ms/step - loss: 0.5515 - crossentropy: 0.5515 - accuracy: 0.7081 - val_loss: 0.6079 - val_crossentropy: 0.6079 - val_accuracy: 0.6481\n",
      "Epoch 32/100\n",
      "15/15 [==============================] - 3s 218ms/step - loss: 0.5628 - crossentropy: 0.5628 - accuracy: 0.7048 - val_loss: 0.5976 - val_crossentropy: 0.5976 - val_accuracy: 0.6589\n",
      "Epoch 33/100\n",
      "15/15 [==============================] - 3s 214ms/step - loss: 0.5517 - crossentropy: 0.5517 - accuracy: 0.7123 - val_loss: 0.5914 - val_crossentropy: 0.5914 - val_accuracy: 0.6672\n",
      "Epoch 34/100\n",
      "15/15 [==============================] - 3s 217ms/step - loss: 0.5509 - crossentropy: 0.5509 - accuracy: 0.7139 - val_loss: 0.5631 - val_crossentropy: 0.5631 - val_accuracy: 0.7134\n",
      "Epoch 35/100\n",
      "15/15 [==============================] - 3s 218ms/step - loss: 0.5469 - crossentropy: 0.5469 - accuracy: 0.7217 - val_loss: 0.5481 - val_crossentropy: 0.5481 - val_accuracy: 0.7266\n",
      "Epoch 36/100\n",
      "15/15 [==============================] - 3s 218ms/step - loss: 0.5545 - crossentropy: 0.5545 - accuracy: 0.7143 - val_loss: 0.5818 - val_crossentropy: 0.5818 - val_accuracy: 0.6948\n",
      "Epoch 37/100\n",
      "15/15 [==============================] - 3s 215ms/step - loss: 0.5723 - crossentropy: 0.5723 - accuracy: 0.6857 - val_loss: 0.5909 - val_crossentropy: 0.5909 - val_accuracy: 0.6651\n",
      "Epoch 38/100\n",
      "15/15 [==============================] - 3s 213ms/step - loss: 0.5654 - crossentropy: 0.5654 - accuracy: 0.7060 - val_loss: 0.5747 - val_crossentropy: 0.5747 - val_accuracy: 0.7153\n",
      "Epoch 39/100\n",
      "15/15 [==============================] - 3s 216ms/step - loss: 0.5461 - crossentropy: 0.5461 - accuracy: 0.7257 - val_loss: 0.5916 - val_crossentropy: 0.5916 - val_accuracy: 0.6666\n",
      "Epoch 40/100\n",
      "15/15 [==============================] - 3s 215ms/step - loss: 0.5529 - crossentropy: 0.5529 - accuracy: 0.7084 - val_loss: 0.5797 - val_crossentropy: 0.5797 - val_accuracy: 0.6823\n",
      "Epoch 41/100\n",
      "15/15 [==============================] - 3s 211ms/step - loss: 0.5367 - crossentropy: 0.5367 - accuracy: 0.7272 - val_loss: 0.5746 - val_crossentropy: 0.5746 - val_accuracy: 0.6979\n",
      "Epoch 42/100\n",
      "15/15 [==============================] - 3s 213ms/step - loss: 0.5280 - crossentropy: 0.5280 - accuracy: 0.7315 - val_loss: 0.5539 - val_crossentropy: 0.5539 - val_accuracy: 0.7143\n",
      "Epoch 43/100\n",
      "15/15 [==============================] - 3s 217ms/step - loss: 0.5414 - crossentropy: 0.5414 - accuracy: 0.7237 - val_loss: 0.5446 - val_crossentropy: 0.5446 - val_accuracy: 0.7280\n",
      "Epoch 44/100\n",
      "15/15 [==============================] - 3s 215ms/step - loss: 0.5313 - crossentropy: 0.5313 - accuracy: 0.7338 - val_loss: 0.5824 - val_crossentropy: 0.5824 - val_accuracy: 0.6798\n",
      "Epoch 45/100\n",
      "15/15 [==============================] - 3s 214ms/step - loss: 0.5503 - crossentropy: 0.5503 - accuracy: 0.7180 - val_loss: 0.5641 - val_crossentropy: 0.5641 - val_accuracy: 0.7007\n",
      "Epoch 46/100\n",
      "15/15 [==============================] - 3s 218ms/step - loss: 0.5357 - crossentropy: 0.5357 - accuracy: 0.7267 - val_loss: 0.5634 - val_crossentropy: 0.5634 - val_accuracy: 0.7119\n",
      "Epoch 47/100\n",
      "15/15 [==============================] - 3s 215ms/step - loss: 0.5376 - crossentropy: 0.5376 - accuracy: 0.7325 - val_loss: 0.5468 - val_crossentropy: 0.5468 - val_accuracy: 0.7228\n",
      "Epoch 48/100\n",
      "15/15 [==============================] - 3s 213ms/step - loss: 0.5396 - crossentropy: 0.5396 - accuracy: 0.7254 - val_loss: 0.5753 - val_crossentropy: 0.5753 - val_accuracy: 0.6935\n",
      "Epoch 49/100\n",
      "15/15 [==============================] - 3s 216ms/step - loss: 0.5255 - crossentropy: 0.5255 - accuracy: 0.7369 - val_loss: 0.5568 - val_crossentropy: 0.5568 - val_accuracy: 0.7089\n",
      "Epoch 50/100\n",
      "15/15 [==============================] - 3s 213ms/step - loss: 0.5273 - crossentropy: 0.5273 - accuracy: 0.7296 - val_loss: 0.5763 - val_crossentropy: 0.5763 - val_accuracy: 0.6907\n",
      "Epoch 51/100\n",
      "15/15 [==============================] - 3s 212ms/step - loss: 0.5315 - crossentropy: 0.5315 - accuracy: 0.7356 - val_loss: 0.5627 - val_crossentropy: 0.5627 - val_accuracy: 0.7111\n",
      "Epoch 52/100\n",
      "15/15 [==============================] - 3s 214ms/step - loss: 0.5187 - crossentropy: 0.5187 - accuracy: 0.7421 - val_loss: 0.5527 - val_crossentropy: 0.5527 - val_accuracy: 0.7135\n",
      "Epoch 53/100\n",
      "15/15 [==============================] - 3s 212ms/step - loss: 0.5135 - crossentropy: 0.5135 - accuracy: 0.7479 - val_loss: 0.5550 - val_crossentropy: 0.5550 - val_accuracy: 0.7113\n",
      "Epoch 54/100\n",
      "15/15 [==============================] - 3s 215ms/step - loss: 0.5394 - crossentropy: 0.5394 - accuracy: 0.7288 - val_loss: 0.5704 - val_crossentropy: 0.5704 - val_accuracy: 0.7002\n",
      "Epoch 55/100\n",
      "15/15 [==============================] - 3s 215ms/step - loss: 0.5227 - crossentropy: 0.5227 - accuracy: 0.7398 - val_loss: 0.5597 - val_crossentropy: 0.5597 - val_accuracy: 0.7064\n",
      "Epoch 56/100\n",
      "15/15 [==============================] - 3s 214ms/step - loss: 0.5336 - crossentropy: 0.5336 - accuracy: 0.7262 - val_loss: 0.5450 - val_crossentropy: 0.5450 - val_accuracy: 0.7187\n",
      "Epoch 57/100\n",
      "15/15 [==============================] - 3s 218ms/step - loss: 0.5341 - crossentropy: 0.5341 - accuracy: 0.7312 - val_loss: 0.5513 - val_crossentropy: 0.5513 - val_accuracy: 0.7218\n",
      "Epoch 58/100\n",
      "15/15 [==============================] - 3s 214ms/step - loss: 0.5097 - crossentropy: 0.5097 - accuracy: 0.7463 - val_loss: 0.5591 - val_crossentropy: 0.5591 - val_accuracy: 0.7124\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light",
      "tags": []
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# parameters\n",
    "lr = 1e-3\n",
    "eps = 1e-7\n",
    "overlap = 'minimum'\n",
    "max_epochs = 100\n",
    "patience = 20\n",
    "directory = \"/content/drive/MyDrive/Spring 21/Spring Courses/COMP 548/Project/Warwick_Dataset/\"\n",
    "image_size = (256,384)\n",
    "batch_size = 5\n",
    "validation_size = 10\n",
    "min_area = 40\n",
    "clf_name = 'long_model_image'\n",
    "\n",
    "tf.keras.backend.clear_session()\n",
    "\n",
    "# Training\n",
    "generator = imageGeneration(batch_size, validation_size, directory, image_size)\n",
    "model = Long_Model((256,384),clf_name = 'long_model_image',loadFrom = None)\n",
    "history = model.fit(max_epochs, generator, patience=patience)\n",
    "\n",
    "# plots\n",
    "plt.figure()\n",
    "plt.subplot(2,1,1)\n",
    "plt.plot(history.history['val_loss'], 'r-')\n",
    "plt.plot(history.history['loss'], 'b-')\n",
    "plt.subplot(2,1,2)\n",
    "plt.plot(history.history['val_accuracy'], 'r-')\n",
    "plt.plot(history.history['accuracy'], 'b-')\n",
    "plt.savefig(f'{clf_name}_history.png')\n",
    "\n",
    "\n",
    "train_metrics = evaluation.evaluate(model, generator, 'train', overlap, min_area)\n",
    "val_metrics = evaluation.evaluate(model, generator, 'val', overlap, min_area)\n",
    "\n",
    "with open(f\"{clf_name}_metrics.txt\", 'w') as fp:\n",
    "    print(\"Training perfomance:\", file=fp)\n",
    "    print(\"Precision\\tRecall\\tMCC\", file=fp)\n",
    "    print(train_metrics.mean(axis=0), file=fp)\n",
    "    print(np.median(train_metrics,axis=0), file=fp)\n",
    "    print(\" ---- \", file=fp)\n",
    "    print(\"Validation perfomance:\", file=fp)\n",
    "    print(\"Precision\\tRecall\\tMCC\", file=fp)\n",
    "    print(val_metrics.mean(axis=0), file=fp)\n",
    "    print(np.median(val_metrics,axis=0), file=fp)\n",
    "\n",
    "np.save(f\"{'long_model_image'}_metrics_train.npy\", train_metrics)\n",
    "np.save(f\"{'long_model_image'}_metrics_val.npy\", val_metrics)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 1000
    },
    "id": "fOflyLV53VTx",
    "outputId": "f9d2ad3d-be80-4762-bce7-3df378aecbf2"
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/100\n",
      "10/10 [==============================] - 8s 398ms/step - loss: 0.6974 - crossentropy: 0.6974 - accuracy: 0.5034 - val_loss: 0.6982 - val_crossentropy: 0.6982 - val_accuracy: 0.4500\n",
      "Epoch 2/100\n",
      "10/10 [==============================] - 2s 217ms/step - loss: 0.6921 - crossentropy: 0.6921 - accuracy: 0.5267 - val_loss: 0.6924 - val_crossentropy: 0.6924 - val_accuracy: 0.5050\n",
      "Epoch 3/100\n",
      "10/10 [==============================] - 2s 220ms/step - loss: 0.6890 - crossentropy: 0.6890 - accuracy: 0.5193 - val_loss: 0.6892 - val_crossentropy: 0.6892 - val_accuracy: 0.5698\n",
      "Epoch 4/100\n",
      "10/10 [==============================] - 2s 218ms/step - loss: 0.6839 - crossentropy: 0.6839 - accuracy: 0.5845 - val_loss: 0.6863 - val_crossentropy: 0.6863 - val_accuracy: 0.5770\n",
      "Epoch 5/100\n",
      "10/10 [==============================] - 2s 223ms/step - loss: 0.6799 - crossentropy: 0.6799 - accuracy: 0.5758 - val_loss: 0.6761 - val_crossentropy: 0.6761 - val_accuracy: 0.6573\n",
      "Epoch 6/100\n",
      "10/10 [==============================] - 2s 226ms/step - loss: 0.6681 - crossentropy: 0.6681 - accuracy: 0.5902 - val_loss: 0.6811 - val_crossentropy: 0.6811 - val_accuracy: 0.5756\n",
      "Epoch 7/100\n",
      "10/10 [==============================] - 2s 220ms/step - loss: 0.6718 - crossentropy: 0.6718 - accuracy: 0.5689 - val_loss: 0.6815 - val_crossentropy: 0.6815 - val_accuracy: 0.5781\n",
      "Epoch 8/100\n",
      "10/10 [==============================] - 2s 220ms/step - loss: 0.6604 - crossentropy: 0.6604 - accuracy: 0.6232 - val_loss: 0.6832 - val_crossentropy: 0.6832 - val_accuracy: 0.5654\n",
      "Epoch 9/100\n",
      "10/10 [==============================] - 2s 218ms/step - loss: 0.6467 - crossentropy: 0.6467 - accuracy: 0.6416 - val_loss: 0.6365 - val_crossentropy: 0.6365 - val_accuracy: 0.6885\n",
      "Epoch 10/100\n",
      "10/10 [==============================] - 2s 217ms/step - loss: 0.6402 - crossentropy: 0.6402 - accuracy: 0.6211 - val_loss: 0.7191 - val_crossentropy: 0.7191 - val_accuracy: 0.4853\n",
      "Epoch 11/100\n",
      "10/10 [==============================] - 2s 224ms/step - loss: 0.6507 - crossentropy: 0.6507 - accuracy: 0.6317 - val_loss: 0.6992 - val_crossentropy: 0.6992 - val_accuracy: 0.5137\n",
      "Epoch 12/100\n",
      "10/10 [==============================] - 2s 218ms/step - loss: 0.6706 - crossentropy: 0.6706 - accuracy: 0.5913 - val_loss: 0.6667 - val_crossentropy: 0.6667 - val_accuracy: 0.5973\n",
      "Epoch 13/100\n",
      "10/10 [==============================] - 2s 217ms/step - loss: 0.6732 - crossentropy: 0.6732 - accuracy: 0.5899 - val_loss: 0.7048 - val_crossentropy: 0.7048 - val_accuracy: 0.4624\n",
      "Epoch 14/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.6729 - crossentropy: 0.6729 - accuracy: 0.5647 - val_loss: 0.6540 - val_crossentropy: 0.6540 - val_accuracy: 0.6638\n",
      "Epoch 15/100\n",
      "10/10 [==============================] - 2s 217ms/step - loss: 0.6497 - crossentropy: 0.6497 - accuracy: 0.6506 - val_loss: 0.6830 - val_crossentropy: 0.6830 - val_accuracy: 0.5270\n",
      "Epoch 16/100\n",
      "10/10 [==============================] - 2s 219ms/step - loss: 0.6427 - crossentropy: 0.6427 - accuracy: 0.6383 - val_loss: 0.6262 - val_crossentropy: 0.6262 - val_accuracy: 0.7017\n",
      "Epoch 17/100\n",
      "10/10 [==============================] - 2s 223ms/step - loss: 0.6255 - crossentropy: 0.6255 - accuracy: 0.6617 - val_loss: 0.6689 - val_crossentropy: 0.6689 - val_accuracy: 0.5924\n",
      "Epoch 18/100\n",
      "10/10 [==============================] - 2s 218ms/step - loss: 0.6832 - crossentropy: 0.6832 - accuracy: 0.6065 - val_loss: 0.7383 - val_crossentropy: 0.7383 - val_accuracy: 0.4681\n",
      "Epoch 19/100\n",
      "10/10 [==============================] - 2s 217ms/step - loss: 0.6619 - crossentropy: 0.6619 - accuracy: 0.5990 - val_loss: 0.6254 - val_crossentropy: 0.6254 - val_accuracy: 0.7004\n",
      "Epoch 20/100\n",
      "10/10 [==============================] - 2s 216ms/step - loss: 0.6388 - crossentropy: 0.6388 - accuracy: 0.6454 - val_loss: 0.6380 - val_crossentropy: 0.6380 - val_accuracy: 0.6879\n",
      "Epoch 21/100\n",
      "10/10 [==============================] - 2s 222ms/step - loss: 0.6274 - crossentropy: 0.6274 - accuracy: 0.6566 - val_loss: 0.6511 - val_crossentropy: 0.6511 - val_accuracy: 0.6357\n",
      "Epoch 22/100\n",
      "10/10 [==============================] - 2s 220ms/step - loss: 0.6206 - crossentropy: 0.6206 - accuracy: 0.6615 - val_loss: 0.6809 - val_crossentropy: 0.6809 - val_accuracy: 0.5865\n",
      "Epoch 23/100\n",
      "10/10 [==============================] - 2s 217ms/step - loss: 0.6154 - crossentropy: 0.6154 - accuracy: 0.6796 - val_loss: 0.6711 - val_crossentropy: 0.6711 - val_accuracy: 0.6110\n",
      "Epoch 24/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.6197 - crossentropy: 0.6197 - accuracy: 0.6589 - val_loss: 0.7483 - val_crossentropy: 0.7483 - val_accuracy: 0.5039\n",
      "Epoch 25/100\n",
      "10/10 [==============================] - 2s 222ms/step - loss: 0.6774 - crossentropy: 0.6774 - accuracy: 0.6045 - val_loss: 0.6028 - val_crossentropy: 0.6028 - val_accuracy: 0.7070\n",
      "Epoch 26/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.6037 - crossentropy: 0.6037 - accuracy: 0.6972 - val_loss: 0.6544 - val_crossentropy: 0.6544 - val_accuracy: 0.6131\n",
      "Epoch 27/100\n",
      "10/10 [==============================] - 2s 225ms/step - loss: 0.6206 - crossentropy: 0.6206 - accuracy: 0.6446 - val_loss: 0.6535 - val_crossentropy: 0.6535 - val_accuracy: 0.6206\n",
      "Epoch 28/100\n",
      "10/10 [==============================] - 2s 222ms/step - loss: 0.6166 - crossentropy: 0.6166 - accuracy: 0.6635 - val_loss: 0.7406 - val_crossentropy: 0.7406 - val_accuracy: 0.5467\n",
      "Epoch 29/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.6279 - crossentropy: 0.6279 - accuracy: 0.6417 - val_loss: 0.5961 - val_crossentropy: 0.5961 - val_accuracy: 0.7236\n",
      "Epoch 30/100\n",
      "10/10 [==============================] - 2s 227ms/step - loss: 0.6338 - crossentropy: 0.6338 - accuracy: 0.6314 - val_loss: 0.6001 - val_crossentropy: 0.6001 - val_accuracy: 0.7171\n",
      "Epoch 31/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.6022 - crossentropy: 0.6022 - accuracy: 0.6872 - val_loss: 0.6495 - val_crossentropy: 0.6495 - val_accuracy: 0.6332\n",
      "Epoch 32/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.5959 - crossentropy: 0.5959 - accuracy: 0.6762 - val_loss: 0.6021 - val_crossentropy: 0.6021 - val_accuracy: 0.6857\n",
      "Epoch 33/100\n",
      "10/10 [==============================] - 2s 218ms/step - loss: 0.6161 - crossentropy: 0.6161 - accuracy: 0.6685 - val_loss: 0.5708 - val_crossentropy: 0.5708 - val_accuracy: 0.7096\n",
      "Epoch 34/100\n",
      "10/10 [==============================] - 2s 219ms/step - loss: 0.5853 - crossentropy: 0.5853 - accuracy: 0.6883 - val_loss: 0.5987 - val_crossentropy: 0.5987 - val_accuracy: 0.6913\n",
      "Epoch 35/100\n",
      "10/10 [==============================] - 2s 222ms/step - loss: 0.5762 - crossentropy: 0.5762 - accuracy: 0.6955 - val_loss: 0.6491 - val_crossentropy: 0.6491 - val_accuracy: 0.6358\n",
      "Epoch 36/100\n",
      "10/10 [==============================] - 2s 224ms/step - loss: 0.5752 - crossentropy: 0.5752 - accuracy: 0.6943 - val_loss: 0.6295 - val_crossentropy: 0.6295 - val_accuracy: 0.6476\n",
      "Epoch 37/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.5823 - crossentropy: 0.5823 - accuracy: 0.6874 - val_loss: 0.6188 - val_crossentropy: 0.6188 - val_accuracy: 0.6565\n",
      "Epoch 38/100\n",
      "10/10 [==============================] - 2s 226ms/step - loss: 0.5702 - crossentropy: 0.5702 - accuracy: 0.6993 - val_loss: 0.5670 - val_crossentropy: 0.5670 - val_accuracy: 0.7007\n",
      "Epoch 39/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.5664 - crossentropy: 0.5664 - accuracy: 0.7100 - val_loss: 0.6995 - val_crossentropy: 0.6995 - val_accuracy: 0.6267\n",
      "Epoch 40/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.6209 - crossentropy: 0.6209 - accuracy: 0.6727 - val_loss: 0.7517 - val_crossentropy: 0.7517 - val_accuracy: 0.5197\n",
      "Epoch 41/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.6171 - crossentropy: 0.6171 - accuracy: 0.6400 - val_loss: 0.5680 - val_crossentropy: 0.5680 - val_accuracy: 0.7423\n",
      "Epoch 42/100\n",
      "10/10 [==============================] - 2s 230ms/step - loss: 0.5994 - crossentropy: 0.5994 - accuracy: 0.6753 - val_loss: 0.6082 - val_crossentropy: 0.6082 - val_accuracy: 0.6867\n",
      "Epoch 43/100\n",
      "10/10 [==============================] - 2s 223ms/step - loss: 0.5855 - crossentropy: 0.5855 - accuracy: 0.6833 - val_loss: 0.5539 - val_crossentropy: 0.5539 - val_accuracy: 0.7209\n",
      "Epoch 44/100\n",
      "10/10 [==============================] - 2s 223ms/step - loss: 0.5760 - crossentropy: 0.5760 - accuracy: 0.6978 - val_loss: 0.5382 - val_crossentropy: 0.5382 - val_accuracy: 0.7284\n",
      "Epoch 45/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.5621 - crossentropy: 0.5621 - accuracy: 0.7014 - val_loss: 0.6286 - val_crossentropy: 0.6286 - val_accuracy: 0.6689\n",
      "Epoch 46/100\n",
      "10/10 [==============================] - 2s 223ms/step - loss: 0.5753 - crossentropy: 0.5753 - accuracy: 0.6956 - val_loss: 0.6954 - val_crossentropy: 0.6954 - val_accuracy: 0.6180\n",
      "Epoch 47/100\n",
      "10/10 [==============================] - 2s 217ms/step - loss: 0.5620 - crossentropy: 0.5620 - accuracy: 0.7094 - val_loss: 0.5569 - val_crossentropy: 0.5569 - val_accuracy: 0.7470\n",
      "Epoch 48/100\n",
      "10/10 [==============================] - 2s 225ms/step - loss: 0.6098 - crossentropy: 0.6098 - accuracy: 0.6593 - val_loss: 0.5641 - val_crossentropy: 0.5641 - val_accuracy: 0.7225\n",
      "Epoch 49/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.5540 - crossentropy: 0.5540 - accuracy: 0.7137 - val_loss: 0.5863 - val_crossentropy: 0.5863 - val_accuracy: 0.6938\n",
      "Epoch 50/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.5618 - crossentropy: 0.5618 - accuracy: 0.7095 - val_loss: 0.7246 - val_crossentropy: 0.7246 - val_accuracy: 0.6152\n",
      "Epoch 51/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.5508 - crossentropy: 0.5508 - accuracy: 0.7201 - val_loss: 0.6105 - val_crossentropy: 0.6105 - val_accuracy: 0.6835\n",
      "Epoch 52/100\n",
      "10/10 [==============================] - 2s 218ms/step - loss: 0.5812 - crossentropy: 0.5812 - accuracy: 0.6755 - val_loss: 0.5826 - val_crossentropy: 0.5826 - val_accuracy: 0.7099\n",
      "Epoch 53/100\n",
      "10/10 [==============================] - 2s 223ms/step - loss: 0.5539 - crossentropy: 0.5539 - accuracy: 0.7112 - val_loss: 0.5809 - val_crossentropy: 0.5809 - val_accuracy: 0.7084\n",
      "Epoch 54/100\n",
      "10/10 [==============================] - 2s 222ms/step - loss: 0.5419 - crossentropy: 0.5419 - accuracy: 0.7200 - val_loss: 0.5836 - val_crossentropy: 0.5836 - val_accuracy: 0.7121\n",
      "Epoch 55/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.5536 - crossentropy: 0.5536 - accuracy: 0.7170 - val_loss: 0.4876 - val_crossentropy: 0.4876 - val_accuracy: 0.7671\n",
      "Epoch 56/100\n",
      "10/10 [==============================] - 2s 217ms/step - loss: 0.5304 - crossentropy: 0.5304 - accuracy: 0.7291 - val_loss: 0.5469 - val_crossentropy: 0.5469 - val_accuracy: 0.7334\n",
      "Epoch 57/100\n",
      "10/10 [==============================] - 2s 220ms/step - loss: 0.5282 - crossentropy: 0.5282 - accuracy: 0.7368 - val_loss: 0.5081 - val_crossentropy: 0.5081 - val_accuracy: 0.7590\n",
      "Epoch 58/100\n",
      "10/10 [==============================] - 2s 219ms/step - loss: 0.5137 - crossentropy: 0.5137 - accuracy: 0.7480 - val_loss: 0.6043 - val_crossentropy: 0.6043 - val_accuracy: 0.6980\n",
      "Epoch 59/100\n",
      "10/10 [==============================] - 2s 222ms/step - loss: 0.5237 - crossentropy: 0.5237 - accuracy: 0.7333 - val_loss: 0.6476 - val_crossentropy: 0.6476 - val_accuracy: 0.6567\n",
      "Epoch 60/100\n",
      "10/10 [==============================] - 2s 219ms/step - loss: 0.5653 - crossentropy: 0.5653 - accuracy: 0.7100 - val_loss: 0.6153 - val_crossentropy: 0.6153 - val_accuracy: 0.6731\n",
      "Epoch 61/100\n",
      "10/10 [==============================] - 2s 223ms/step - loss: 0.5538 - crossentropy: 0.5538 - accuracy: 0.7118 - val_loss: 0.5091 - val_crossentropy: 0.5091 - val_accuracy: 0.7590\n",
      "Epoch 62/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.5314 - crossentropy: 0.5314 - accuracy: 0.7406 - val_loss: 0.5132 - val_crossentropy: 0.5132 - val_accuracy: 0.7543\n",
      "Epoch 63/100\n",
      "10/10 [==============================] - 2s 222ms/step - loss: 0.5114 - crossentropy: 0.5114 - accuracy: 0.7466 - val_loss: 0.5895 - val_crossentropy: 0.5895 - val_accuracy: 0.7065\n",
      "Epoch 64/100\n",
      "10/10 [==============================] - 2s 225ms/step - loss: 0.5132 - crossentropy: 0.5132 - accuracy: 0.7509 - val_loss: 0.5861 - val_crossentropy: 0.5861 - val_accuracy: 0.7136\n",
      "Epoch 65/100\n",
      "10/10 [==============================] - 2s 224ms/step - loss: 0.5408 - crossentropy: 0.5408 - accuracy: 0.7249 - val_loss: 0.5659 - val_crossentropy: 0.5659 - val_accuracy: 0.7105\n",
      "Epoch 66/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.5397 - crossentropy: 0.5397 - accuracy: 0.7315 - val_loss: 0.5146 - val_crossentropy: 0.5146 - val_accuracy: 0.7524\n",
      "Epoch 67/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.5089 - crossentropy: 0.5089 - accuracy: 0.7490 - val_loss: 0.6259 - val_crossentropy: 0.6259 - val_accuracy: 0.6795\n",
      "Epoch 68/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.5373 - crossentropy: 0.5373 - accuracy: 0.7243 - val_loss: 0.5974 - val_crossentropy: 0.5974 - val_accuracy: 0.6866\n",
      "Epoch 69/100\n",
      "10/10 [==============================] - 2s 224ms/step - loss: 0.5485 - crossentropy: 0.5485 - accuracy: 0.7213 - val_loss: 0.4650 - val_crossentropy: 0.4650 - val_accuracy: 0.7884\n",
      "Epoch 70/100\n",
      "10/10 [==============================] - 2s 220ms/step - loss: 0.5055 - crossentropy: 0.5055 - accuracy: 0.7496 - val_loss: 0.6406 - val_crossentropy: 0.6406 - val_accuracy: 0.6658\n",
      "Epoch 71/100\n",
      "10/10 [==============================] - 2s 224ms/step - loss: 0.5147 - crossentropy: 0.5147 - accuracy: 0.7520 - val_loss: 0.4885 - val_crossentropy: 0.4885 - val_accuracy: 0.7761\n",
      "Epoch 72/100\n",
      "10/10 [==============================] - 2s 224ms/step - loss: 0.4985 - crossentropy: 0.4985 - accuracy: 0.7565 - val_loss: 0.5875 - val_crossentropy: 0.5875 - val_accuracy: 0.7054\n",
      "Epoch 73/100\n",
      "10/10 [==============================] - 2s 224ms/step - loss: 0.5128 - crossentropy: 0.5128 - accuracy: 0.7485 - val_loss: 0.5147 - val_crossentropy: 0.5147 - val_accuracy: 0.7559\n",
      "Epoch 74/100\n",
      "10/10 [==============================] - 2s 225ms/step - loss: 0.5530 - crossentropy: 0.5530 - accuracy: 0.7162 - val_loss: 0.4545 - val_crossentropy: 0.4545 - val_accuracy: 0.7937\n",
      "Epoch 75/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.5074 - crossentropy: 0.5074 - accuracy: 0.7489 - val_loss: 0.5509 - val_crossentropy: 0.5509 - val_accuracy: 0.7267\n",
      "Epoch 76/100\n",
      "10/10 [==============================] - 2s 225ms/step - loss: 0.5117 - crossentropy: 0.5117 - accuracy: 0.7414 - val_loss: 0.5345 - val_crossentropy: 0.5345 - val_accuracy: 0.7398\n",
      "Epoch 77/100\n",
      "10/10 [==============================] - 2s 219ms/step - loss: 0.5260 - crossentropy: 0.5260 - accuracy: 0.7424 - val_loss: 0.4482 - val_crossentropy: 0.4482 - val_accuracy: 0.7909\n",
      "Epoch 78/100\n",
      "10/10 [==============================] - 2s 224ms/step - loss: 0.5124 - crossentropy: 0.5124 - accuracy: 0.7530 - val_loss: 0.5565 - val_crossentropy: 0.5565 - val_accuracy: 0.7260\n",
      "Epoch 79/100\n",
      "10/10 [==============================] - 2s 223ms/step - loss: 0.5218 - crossentropy: 0.5218 - accuracy: 0.7396 - val_loss: 0.4449 - val_crossentropy: 0.4449 - val_accuracy: 0.7954\n",
      "Epoch 80/100\n",
      "10/10 [==============================] - 2s 228ms/step - loss: 0.5140 - crossentropy: 0.5140 - accuracy: 0.7479 - val_loss: 0.5108 - val_crossentropy: 0.5108 - val_accuracy: 0.7571\n",
      "Epoch 81/100\n",
      "10/10 [==============================] - 2s 219ms/step - loss: 0.4893 - crossentropy: 0.4893 - accuracy: 0.7702 - val_loss: 0.5251 - val_crossentropy: 0.5251 - val_accuracy: 0.7460\n",
      "Epoch 82/100\n",
      "10/10 [==============================] - 2s 227ms/step - loss: 0.5168 - crossentropy: 0.5168 - accuracy: 0.7423 - val_loss: 0.5276 - val_crossentropy: 0.5276 - val_accuracy: 0.7464\n",
      "Epoch 83/100\n",
      "10/10 [==============================] - 2s 222ms/step - loss: 0.5049 - crossentropy: 0.5049 - accuracy: 0.7486 - val_loss: 0.4923 - val_crossentropy: 0.4923 - val_accuracy: 0.7722\n",
      "Epoch 84/100\n",
      "10/10 [==============================] - 2s 225ms/step - loss: 0.4987 - crossentropy: 0.4987 - accuracy: 0.7567 - val_loss: 0.4828 - val_crossentropy: 0.4828 - val_accuracy: 0.7758\n",
      "Epoch 85/100\n",
      "10/10 [==============================] - 2s 224ms/step - loss: 0.4825 - crossentropy: 0.4825 - accuracy: 0.7719 - val_loss: 0.4397 - val_crossentropy: 0.4397 - val_accuracy: 0.8013\n",
      "Epoch 86/100\n",
      "10/10 [==============================] - 2s 220ms/step - loss: 0.4588 - crossentropy: 0.4588 - accuracy: 0.7824 - val_loss: 0.4613 - val_crossentropy: 0.4613 - val_accuracy: 0.7873\n",
      "Epoch 87/100\n",
      "10/10 [==============================] - 2s 222ms/step - loss: 0.4621 - crossentropy: 0.4621 - accuracy: 0.7818 - val_loss: 0.5353 - val_crossentropy: 0.5353 - val_accuracy: 0.7400\n",
      "Epoch 88/100\n",
      "10/10 [==============================] - 2s 227ms/step - loss: 0.5659 - crossentropy: 0.5659 - accuracy: 0.7159 - val_loss: 0.4435 - val_crossentropy: 0.4435 - val_accuracy: 0.7980\n",
      "Epoch 89/100\n",
      "10/10 [==============================] - 2s 230ms/step - loss: 0.4901 - crossentropy: 0.4901 - accuracy: 0.7651 - val_loss: 0.4947 - val_crossentropy: 0.4947 - val_accuracy: 0.7691\n",
      "Epoch 90/100\n",
      "10/10 [==============================] - 2s 226ms/step - loss: 0.4548 - crossentropy: 0.4548 - accuracy: 0.7917 - val_loss: 0.4896 - val_crossentropy: 0.4896 - val_accuracy: 0.7689\n",
      "Epoch 91/100\n",
      "10/10 [==============================] - 2s 224ms/step - loss: 0.4578 - crossentropy: 0.4578 - accuracy: 0.7857 - val_loss: 0.5493 - val_crossentropy: 0.5493 - val_accuracy: 0.7278\n",
      "Epoch 92/100\n",
      "10/10 [==============================] - 2s 225ms/step - loss: 0.4759 - crossentropy: 0.4759 - accuracy: 0.7764 - val_loss: 0.4482 - val_crossentropy: 0.4482 - val_accuracy: 0.7946\n",
      "Epoch 93/100\n",
      "10/10 [==============================] - 2s 225ms/step - loss: 0.4660 - crossentropy: 0.4660 - accuracy: 0.7822 - val_loss: 0.4554 - val_crossentropy: 0.4554 - val_accuracy: 0.7903\n",
      "Epoch 94/100\n",
      "10/10 [==============================] - 2s 225ms/step - loss: 0.4871 - crossentropy: 0.4871 - accuracy: 0.7694 - val_loss: 0.4932 - val_crossentropy: 0.4932 - val_accuracy: 0.7669\n",
      "Epoch 95/100\n",
      "10/10 [==============================] - 2s 225ms/step - loss: 0.4773 - crossentropy: 0.4773 - accuracy: 0.7713 - val_loss: 0.4731 - val_crossentropy: 0.4731 - val_accuracy: 0.7785\n",
      "Epoch 96/100\n",
      "10/10 [==============================] - 2s 225ms/step - loss: 0.4723 - crossentropy: 0.4723 - accuracy: 0.7786 - val_loss: 0.4474 - val_crossentropy: 0.4474 - val_accuracy: 0.7937\n",
      "Epoch 97/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.4797 - crossentropy: 0.4797 - accuracy: 0.7698 - val_loss: 0.5019 - val_crossentropy: 0.5019 - val_accuracy: 0.7575\n",
      "Epoch 98/100\n",
      "10/10 [==============================] - 2s 222ms/step - loss: 0.4848 - crossentropy: 0.4848 - accuracy: 0.7635 - val_loss: 0.4763 - val_crossentropy: 0.4763 - val_accuracy: 0.7759\n",
      "Epoch 99/100\n",
      "10/10 [==============================] - 2s 225ms/step - loss: 0.4827 - crossentropy: 0.4827 - accuracy: 0.7719 - val_loss: 0.4271 - val_crossentropy: 0.4271 - val_accuracy: 0.8048\n",
      "Epoch 100/100\n",
      "10/10 [==============================] - 2s 221ms/step - loss: 0.4762 - crossentropy: 0.4762 - accuracy: 0.7676 - val_loss: 0.4753 - val_crossentropy: 0.4753 - val_accuracy: 0.7736\n",
      "Test A\n",
      "Test B\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light",
      "tags": []
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# parameters\n",
    "lr = 1e-3\n",
    "eps = 1e-7\n",
    "overlap = 'minimum'\n",
    "max_epochs = 100\n",
    "patience = 20\n",
    "directory = \"/content/drive/MyDrive/Spring 21/Spring Courses/COMP 548/Project/Warwick_Dataset/\"\n",
    "image_size = (256,384)\n",
    "batch_size = 5\n",
    "validation_size = 10\n",
    "min_area = 40\n",
    "clf_name = 'long_model_image'\n",
    "\n",
    "tf.keras.backend.clear_session()\n",
    "\n",
    "# model = Model(image_size, clf_name, loadFrom=f'{clf_name}.hdf5')\n",
    "model = Long_Model((256,384),clf_name = 'long_model_image',loadFrom = None)\n",
    "\n",
    "# Compute & save metrics\n",
    "generator = imageGeneration(batch_size, validation_size, directory, image_size, train_test='testA')\n",
    "metrics = evaluation.evaluate(model, generator, 'train', overlap, min_area)\n",
    "history = model.fit(max_epochs, generator, patience=patience)\n",
    "\n",
    "plt.figure()\n",
    "plt.subplot(2,1,1)\n",
    "plt.plot(history.history['loss'], 'b-')\n",
    "plt.subplot(2,1,2)\n",
    "plt.plot(history.history['accuracy'], 'b-')\n",
    "plt.savefig(f'{clf_name}_history.png')\n",
    "\n",
    "\n",
    "with open(f\"{clf_name}_testA_metrics.txt\", 'w') as fp:\n",
    "    print(\"Test A\")\n",
    "    print(\"Precision\\tRecall\\tMCC\", file=fp)\n",
    "    print(metrics.mean(axis=0), file=fp)\n",
    "    print(np.median(metrics,axis=0), file=fp)\n",
    "    \n",
    "np.save(f\"{clf_name}_testA_metrics.npy\", metrics)\n",
    "\n",
    "generator = imageGeneration(batch_size, validation_size, directory, image_size, train_test='testB')\n",
    "metrics = evaluation.evaluate(model, generator, 'train', overlap, min_area)\n",
    "\n",
    "with open(f\"{clf_name}_testB_metrics.txt\", 'w') as fp:\n",
    "    print(\"Test B\")\n",
    "    print(\"Precision\\tRecall\\tMCC\", file=fp)\n",
    "    print(metrics.mean(axis=0), file=fp)\n",
    "    print(np.median(metrics,axis=0), file=fp)\n",
    "    \n",
    "np.save(f\"{clf_name}_testB_metrics.npy\", metrics)\n"
   ]
  }
 ],
 "metadata": {
  "accelerator": "GPU",
  "colab": {
   "name": "Main_LongSkip_Model.ipynb",
   "provenance": []
  },
  "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": "3.9.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}