{"cells":[{"metadata":{},"cell_type":"markdown","source":"# ResNet50","execution_count":null},{"metadata":{"_cell_guid":"b1076dfc-b9ad-4769-8c92-a6c4dae69d19","_uuid":"8f2839f25d086af736a60e9eeb907d3b93b6e0e5","trusted":true},"cell_type":"code","source":"# This Python 3 environment comes with many helpful analytics libraries installed\n# It is defined by the kaggle/python Docker image: https://github.com/kaggle/docker-python\n# For example, here's several helpful packages to load\n\nimport numpy as np # linear algebra\nimport pandas as pd # data processing, CSV file I/O (e.g. pd.read_csv)\n\n# Input data files are available in the read-only \"../input/\" directory\n# For example, running this (by clicking run or pressing Shift+Enter) will list all files under the input directory\n\nimport os\n# for dirname, _, filenames in os.walk('/kaggle/input'):\n#     for filename in filenames:\n#         print(os.path.join(dirname, filename))\n\n# You can write up to 5GB to the current directory (/kaggle/working/) that gets preserved as output when you create a version using \"Save & Run All\" \n# You can also write temporary files to /kaggle/temp/, but they won't be saved outside of the current session","execution_count":null,"outputs":[]},{"metadata":{"_cell_guid":"79c7e3d0-c299-4dcb-8224-4455121ee9b0","_uuid":"d629ff2d2480ee46fbb7e2d37f6b5fab8052498a","trusted":true},"cell_type":"code","source":"import json\nimport math\nimport os\n\nimport cv2\nfrom PIL import Image\nimport numpy as np\nfrom keras import layers\nfrom keras.applications import DenseNet121\nfrom keras.callbacks import Callback, ModelCheckpoint\nfrom keras.preprocessing.image import ImageDataGenerator\nfrom keras.models import Sequential\nfrom keras.optimizers import Adam\nimport matplotlib.pyplot as plt\nimport pandas as pd\nfrom sklearn.model_selection import train_test_split\nfrom sklearn.metrics import cohen_kappa_score, accuracy_score, auc, roc_auc_score, roc_curve\nimport sklearn\nimport scipy\nimport tensorflow as tf\nfrom tqdm import tqdm\nfrom keras.preprocessing import image\nfrom keras.models import Model\nfrom keras.layers import BatchNormalization, Dropout, Conv2D, MaxPooling2D, GlobalAveragePooling2D, Flatten, Dense\nimport keras\nfrom keras.models import Sequential\nfrom keras.layers import Dense, Activation, Dropout, Flatten, Conv2D, MaxPooling2D\nfrom keras.layers.normalization import BatchNormalization\n\n%matplotlib inline","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# !pip install git+https://github.com/qubvel/efficientnet","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# from efficientnet.tfkeras import EfficientNetB7 as effnetb7","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"np.random.seed(2019)\ntf.random.set_seed(2019)\nTEST_SIZE = 0.25\nSEED = 2019\nBATCH_SIZE = 8","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"train_df = pd.read_csv('../input/aptos2019-blindness-detection/train.csv')\ntest_df = pd.read_csv('../input/aptos2019-blindness-detection/test.csv')\ntrain_df.head(7)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"x_train = np.load('../input/four-fold-aptos/train_all_four.npy')\nx_test = np.load('../input/four-fold-aptos/test_all_four.npy')","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"y_train = train_df['diagnosis'].values\ny_train","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"y_train_one_hot = pd.get_dummies(train_df['diagnosis']).values\n\ny_train_multi = np.empty(y_train_one_hot.shape, dtype=y_train_one_hot.dtype)\ny_train_multi[:, 4] = y_train_one_hot[:, 4]\n\nfor i in range(3, -1, -1):\n    y_train_multi[:, i] = np.logical_or(y_train_one_hot[:, i], y_train_multi[:, i+1])\n\nprint(\"Original y_train:\", y_train_one_hot.sum(axis=0))\nprint(\"Multilabel version:\", y_train_multi.sum(axis=0))","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"y_train_one_hot","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"y_train_multi","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"x_train, x_val, y_train, y_val = train_test_split(\n    x_train, y_train_multi, \n    test_size=TEST_SIZE, \n    random_state=SEED\n)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"y_train_no_multi, y_val_no_multi = train_test_split(\n    y_train, \n    test_size=TEST_SIZE, \n    random_state=SEED\n)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def create_datagen():\n    return ImageDataGenerator(\n        zoom_range=0.15,  # set range for random zoom\n        # set mode for filling points outside the input boundaries\n        fill_mode='constant',\n        cval=0.,  # value used for fill_mode = \"constant\"\n        horizontal_flip=True,  # randomly flip images\n        vertical_flip=True,  # randomly flip images\n    )\n\n# Using original generator\ndata_generator = create_datagen().flow(x_train, y_train, batch_size=BATCH_SIZE, seed=SEED)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"\n\nclass Metrics(Callback):\n    def on_train_begin(self, logs={}):\n        self.val_kappas = []\n\n    def on_epoch_end(self, epoch, logs={}):\n        X_val, y_val = self.validation_data[:2]\n        y_val = y_val.sum(axis=1) - 1\n        \n        y_pred = self.model.predict(X_val) > 0.5\n        y_pred = y_pred.astype(int).sum(axis=1) - 1\n\n        _val_kappa = cohen_kappa_score(\n            y_val,\n            y_pred, \n            weights='quadratic'\n        )\n\n        self.val_kappas.append(_val_kappa)\n\n        print(f\"val_kappa: {_val_kappa:.4f}\")\n        \n        if _val_kappa == max(self.val_kappas):\n            print(\"Validation Kappa has improved. Saving model.\")\n            self.model.save('effnetb7_model.h5')\n\n        return\n\n","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"x_train[1].shape","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"## Building the model","execution_count":null},{"metadata":{},"cell_type":"markdown","source":"### Alexnet","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"def alex_net():\n    model = Sequential()\n\n    # 1st Convolutional Layer\n    model.add(Conv2D(filters=96, input_shape=(224,224,3), kernel_size=(11,11), strides=(4,4), padding='valid'))\n    model.add(Activation('relu'))\n    # Max Pooling\n    model.add(MaxPooling2D(pool_size=(2,2), strides=(2,2), padding='valid'))\n\n    # 2nd Convolutional Layer\n    model.add(Conv2D(filters=256, kernel_size=(11,11), strides=(1,1), padding='valid'))\n    model.add(Activation('relu'))\n    # Max Pooling\n    model.add(MaxPooling2D(pool_size=(2,2), strides=(2,2), padding='valid'))\n\n    # 3rd Convolutional Layer\n    model.add(Conv2D(filters=384, kernel_size=(3,3), strides=(1,1), padding='valid'))\n    model.add(Activation('relu'))\n\n    # 4th Convolutional Layer\n    model.add(Conv2D(filters=384, kernel_size=(3,3), strides=(1,1), padding='valid'))\n    model.add(Activation('relu'))\n\n    # 5th Convolutional Layer\n    model.add(Conv2D(filters=256, kernel_size=(3,3), strides=(1,1), padding='valid'))\n    model.add(Activation('relu'))\n    # Max Pooling\n    model.add(MaxPooling2D(pool_size=(2,2), strides=(2,2), padding='valid'))\n\n    # Passing it to a Fully Connected layer\n    model.add(Flatten())\n    # 1st Fully Connected Layer\n    model.add(Dense(4096, input_shape=(224*224*3,)))\n    model.add(Activation('relu'))\n    # Add Dropout to prevent overfitting\n    model.add(Dropout(0.4))\n\n    # 2nd Fully Connected Layer\n    model.add(Dense(4096))\n    model.add(Activation('relu'))\n    # Add Dropout\n    model.add(Dropout(0.4))\n\n    # 3rd Fully Connected Layer\n    model.add(Dense(1000))\n    model.add(Activation('relu'))\n    # Add Dropout\n    model.add(Dropout(0.4))\n\n    # Output Layer\n    model.add(Dense(5, activation = 'sigmoid'))\n#     model.add(Activation('softmax'))\n    return model","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"### VGG16","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"def vgg16():\n    base_model = tf.keras.applications.VGG16(include_top=False,\n                                            weights=\"imagenet\",\n                                            input_shape=x_train[0].shape)\n    x = base_model.output\n    batch_normal = BatchNormalization()(x)\n    global_avg_pooling = GlobalAveragePooling2D()(batch_normal)\n    drop_out = Dropout(0.5)(global_avg_pooling)\n    dense1 = Dense(1024, activation='relu')(drop_out)\n    dense2 = Dense(5, activation = 'sigmoid')(dense1)\n    model = Model(inputs = base_model.input, outputs = dense2)\n    for layer in model.layers:\n        layer.trainable = True\n    return model","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"### VGG19","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"def vgg19():\n    base_model = tf.keras.applications.VGG19(include_top=False,\n                                            weights=\"imagenet\",\n                                            input_shape=x_train[0].shape)\n    x = base_model.output\n    batch_normal = BatchNormalization()(x)\n    global_avg_pooling = GlobalAveragePooling2D()(batch_normal)\n    drop_out = Dropout(0.5)(global_avg_pooling)\n    dense1 = Dense(1024, activation='relu')(drop_out)\n    dense2 = Dense(5, activation = 'sigmoid')(dense1)\n    model = Model(inputs = base_model.input, outputs = dense2)\n    for layer in model.layers:\n        layer.trainable = True\n    return model","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"### InceptionV3","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"def inceptionv3():\n    base_model = tf.keras.applications.InceptionV3(include_top=False,\n                                            weights=\"imagenet\",\n                                            input_shape=x_train[0].shape)\n    x = base_model.output\n    batch_normal = BatchNormalization()(x)\n    global_avg_pooling = GlobalAveragePooling2D()(batch_normal)\n    drop_out = Dropout(0.5)(global_avg_pooling)\n    dense1 = Dense(1024, activation='relu')(drop_out)\n    dense2 = Dense(5, activation = 'sigmoid')(dense1)\n    model = Model(inputs = base_model.input, outputs = dense2)\n    for layer in model.layers:\n        layer.trainable = True\n    return model","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"### ResNet","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"def resnet50():\n    base_model = tf.keras.applications.ResNet50(include_top=False,\n                                            weights=\"imagenet\",\n                                            input_shape=x_train[0].shape)\n    x = base_model.output\n    batch_normal = BatchNormalization()(x)\n    global_avg_pooling = GlobalAveragePooling2D()(batch_normal)\n    drop_out = Dropout(0.5)(global_avg_pooling)\n    dense1 = Dense(1024, activation='relu')(drop_out)\n    dense2 = Dense(5, activation = 'sigmoid')(dense1)\n    model = Model(inputs = base_model.input, outputs = dense2)\n    for layer in model.layers:\n        layer.trainable = True\n    return model","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# base_model = alex_net()\n# base_model = vgg19()\n# base_model = inceptionv3()\nbase_model = resnet50()\nbase_model.compile(loss=keras.losses.categorical_crossentropy, optimizer='adam', metrics=[\"accuracy\"])","execution_count":null,"outputs":[]},{"metadata":{"_kg_hide-output":true,"trusted":true},"cell_type":"code","source":"base_model.summary()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"model = base_model","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"for layer in base_model.layers:\n    layer.trainable = False\n\nfor layer in model.layers:\n    layer.trainable = True\n    \n# model.summary()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"from keras.callbacks import EarlyStopping, ReduceLROnPlateau\n\nearly_stop = EarlyStopping(monitor='val_loss', min_delta=0.0001, patience=3, verbose=1, mode='auto')\n# Reducing the Learning Rate if result is not improving. \nreduce_lr = ReduceLROnPlateau(monitor='val_loss', min_delta=0.0004, patience=2, factor=0.1, min_lr=1e-6, mode='auto',\n                              verbose=1)\n\nkappa_metrics = Metrics()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# model.compile(loss='categorical_crossentropy', optimizer=sgd, metrics=['accuracy'])\nmodel.compile(loss='binary_crossentropy', optimizer=Adam(lr=0.00005), metrics=['accuracy','AUC'])","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# model.load_weights('../input/effnet-preprocessed/weights/efficientnet-b7_noisy_student_notop_aptos.h5')","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"history = model.fit_generator(\n    data_generator,\n    steps_per_epoch=x_train.shape[0] / BATCH_SIZE,\n    epochs=20,\n    validation_data=(x_val, y_val),\n    callbacks=[early_stop, reduce_lr]\n)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"model.save('./resnet50_aptos.h5')","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"history.history.keys()","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"with open('./resnet50_aptos.json', 'w') as fp:\n    json.dump(str(history.history), fp)","execution_count":null,"outputs":[]},{"metadata":{},"cell_type":"markdown","source":"## To load previous Model","execution_count":null},{"metadata":{"trusted":true},"cell_type":"code","source":"# base_model = effnetb7(include_top=False,\n#                      weights = None,\n#                      input_shape=(224,224,3))\n\n\n# x = base_model.output\n# batch_normal = BatchNormalization()(x)\n# global_avg_pooling = GlobalAveragePooling2D()(batch_normal)\n# drop_out = Dropout(0.5)(global_avg_pooling)\n# dense1 = Dense(1024, activation='relu')(drop_out)\n# dense2 = Dense(5, activation = 'sigmoid')(dense1)\n# model = Model(inputs = base_model.input, outputs = dense2)\n\n# model.load_weights('../input/effnet-preprocessed/weights/efficientnet-b7_noisy_student_notop_aptos.h5')\n\n# model.compile(loss='binary_crossentropy', optimizer=Adam(lr=0.00005), metrics=['accuracy','AUC'])","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"y_predict = model.predict(x_val)\ny_predict","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"val_y = y_predict > 0.5\nval_y = val_y.astype(int).sum(axis=1) - 1\nval_y","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"y_real = [4 if (list(i)[4]==1) else list(i).index(0)-1 for i in y_val]\n# y_real","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"from sklearn.metrics import confusion_matrix \nfrom sklearn.metrics import accuracy_score \nfrom sklearn.metrics import classification_report \n  \nactual = y_real\npredicted = val_y\nresults = confusion_matrix(actual, predicted) \n  \nprint ('Confusion Matrix :')\nprint(results)\nprint ('Accuracy Score :',accuracy_score(actual, predicted) )\nprint ('Report : ')\nprint (classification_report(actual, predicted))","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"predict_probab = [y_predict[i][val_y[i]] for i in range(len(val_y))]\n# predict_probab","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"y_val_one_hot = []\nfor i in range(len(y_real)):\n    y_val_one_hot.append(list(np.zeros(5, dtype = 'uint8')))\ny_val_one_hot = np.array(y_val_one_hot)\nfor i in range(y_val_one_hot.shape[0]):\n    y_val_one_hot[i][y_real[i]] = 1","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"def plot_roc(label):\n    y_probab = y_predict[:, label]\n    y_label = y_val_one_hot[:,label]\n    fpr, tpr, thresholds = roc_curve(y_label, y_probab)\n    auc = sklearn.metrics.auc(fpr, tpr)\n    plt.plot([0,1],[0,1], 'k--')\n    plt.plot(fpr,tpr, label = 'AUC SCORE : {:.3f}'.format(auc))\n    plt.title('AUC ROC Curve of class '+str(label))\n    plt.xlabel('False Positive rate')\n    plt.ylabel('True Positive rate')\n    plt.legend(loc = 'best')","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"plot_roc(0)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"plot_roc(1)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"plot_roc(2)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"plot_roc(3)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"plot_roc(4)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"model.evaluate(x_val,y_val)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"y_test_p = model.predict(x_test)","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"test_y = y_test_p > 0.5\ntest_y = test_y.astype(int).sum(axis=1) - 1\ntest_y","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"cohen_kappa_score(\n            y_real,\n            val_y, \n            weights='quadratic'\n        )","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"# import json\n\n# f = open('../input/effnet-preprocessed/efficientnet-b7_noisy_student_notop.json')\n\n# data = json.load(f)\n# data = data.replace(\"\\'\", \"\\\"\")\n# f.close()\n\n# hist = json.loads(data)\nhist = history.history","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"plt.figure(figsize=(8, 8))\nplt.title(\"Learning curve\")\nplt.plot(hist[\"loss\"], label=\"loss\")\nplt.plot(hist[\"val_loss\"], label=\"val_loss\")\n# plt.plot(np.argmin(hist[\"val_loss\"]), np.min(hist[\"val_loss\"]), marker=\"x\", color=\"r\",\n#          label=\"best model\")\nplt.xlabel(\"Epochs\")\nplt.ylabel(\"log_loss\")\nplt.legend();","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"plt.figure(figsize=(8, 8))\nplt.title(\"Learning curve\")\nplt.plot(hist[\"accuracy\"], label=\"loss\")\nplt.plot(hist[\"val_accuracy\"], label=\"val_loss\")\n# plt.plot(np.argmin(hist[\"val_loss\"]), np.min(hist[\"val_loss\"]), marker=\"x\", color=\"r\",\n#          label=\"best model\")\nplt.xlabel(\"Epochs\")\nplt.ylabel(\"accuracy\")\nplt.legend();","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"plt.figure(figsize=(8, 8))\nplt.title(\"Learning curve\")\nplt.plot(hist[\"auc\"], label=\"loss\")\nplt.plot(hist[\"val_auc\"], label=\"val_loss\")\n# plt.plot(np.argmin(hist[\"val_loss\"]), np.min(hist[\"val_loss\"]), marker=\"x\", color=\"r\",\n#          label=\"best model\")\nplt.xlabel(\"Epochs\")\nplt.ylabel(\"auc\")\nplt.legend();","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"hist","execution_count":null,"outputs":[]},{"metadata":{"trusted":true},"cell_type":"code","source":"","execution_count":null,"outputs":[]}],"metadata":{"kernelspec":{"language":"python","display_name":"Python 3","name":"python3"},"language_info":{"pygments_lexer":"ipython3","nbconvert_exporter":"python","version":"3.6.4","file_extension":".py","codemirror_mode":{"name":"ipython","version":3},"name":"python","mimetype":"text/x-python"}},"nbformat":4,"nbformat_minor":4}