model = (TimeDistributed(Conv2D(256, (3, 3), strides=(1, 1), activation='relu', padding='same')))(model)
model = (TimeDistributed(Conv2D(256, (3,3), activation='relu')))(model)
model = (TimeDistributed(MaxPooling2D((2, 2), strides=(1, 1))))(model)
model = (Dropout(0.25))(model)
model = (TimeDistributed(Flatten()))(model)
model = (Dropout(0.5))(model)
#AFTER LSTM
model = LSTM(128, return_sequences=True)(model)
model = attention_3d_block(model)
model = Flatten()(model)
model = (Dense(6, activation='softmax'))(model)
model = Model(input = x, output = model)
opt = keras.optimizers.Adam(lr=0.0005)
print(model.summary())
model.compile(loss='categorical_crossentropy', optimizer=opt, metrics=['accuracy'])
## filepath="weights-improvement-{epoch:02d}-{val_acc:.2f}.hdf5"
## checkpoint = ModelCheckpoint(filepath, monitor='val_acc', verbose=1, save_best_only=True, mode='auto')
## callbacks_list = [checkpoint]
#mm = ModelCheckpoint('/user1/temp/cscr/pooja_t/', monitor='loss', verbose=0, save_best_only=True, mode='auto', period=1)
checkpoint = EarlyStopping(monitor='loss', min_delta=0, patience=10, verbose=0, mode='auto')
callbacks_list = [checkpoint]
model.fit(X_train, y_train, batch_size=32, epochs=70, verbose=1, callbacks=callbacks_list, validation_split=0.1)
#model.fit(X_train, y_train, batch_size=32, epochs=100, verbose=1)
#model.load_weights('C:\\Users\\Sanmoy\\Desktop\\pooja\\paper read\\sports\\dataset\\UIUC2\\B128-256.h5')
score = (model.evaluate(X_test,y_test))
print("%s: %.2f%%" % (model.metrics_names[1], score[1]*100))
model.save('/user1/temp/cscr/pooja_t/after64.h5')
