x.add(Flatten()) x.add(Dense(16, activation='relu')) x.add(Dropout(0.5)) x.add(BatchNormalization()) # predictions = Dense(num_classes, activation = 'softmax')(x) x.add(Dense(len(classes), activation="softmax")) # Compile x.compile(optimizer=Adam(lr=0.0001), loss='categorical_crossentropy', metrics=['accuracy']) # Let's keep on csv file... for c in batches.class_indices: classes[batches.class_indices[c]] = c x.classes = classes filename = 'model_train_new.csv' csv_log = keras.callbacks.CSVLogger(filename, separator=',', append=False) # Early stopping if loss doesnt improve early_stopping = EarlyStopping(patience=10) checkpointer = ModelCheckpoint('resnet_best.h5', verbose=1, save_best_only=True) tensorboard_callback = keras.callbacks.TensorBoard(log_dir='./logs', histogram_freq=0, batch_size=BATCH_SIZE, write_graph=True, write_grads=False, write_images=False,