def LSTM_model(self, conf, arm_shape):
road_num = arm_shape[0]
input_x = Input((road_num, conf.observe_length, 1))
output = MyReshape(conf.batch_size)(input_x)
output = LSTM(conf.observe_length)(output)
output = Dense(conf.predict_length)(output)
output = MyInverseReshape(conf.batch_size)(output)
model = Model(inputs=input_x, outputs=output)
return model
def LCRNNBN_model(self, conf, arm_shape):
road_num = arm_shape[0]
A = arm_shape[1]
input_x = Input((road_num, conf.observe_length, 1))
input_ram = Input(arm_shape)
output = Lookup(conf.batch_size)([input_x, input_ram])
output = Conv3D(16, (1, A, 2), activation="relu")(output)
output = BatchNormalization()(output)
output = LookUpSqueeze()(output)
output = Lookup(conf.batch_size)([output, input_ram])
output = Conv3D(16, (1, A, 2), activation="relu")(output)
output = BatchNormalization()(output)
output = LookUpSqueeze()(output)
output = Lookup(conf.batch_size)([output, input_ram])
output = Conv3D(16, (1, A, 2), activation="relu")(output)
output = BatchNormalization()(output)
output = LookUpSqueeze()(output)
output = MyReshape(conf.batch_size)(output)
output = SimpleRNN(5)(output)
inputs = [input_x, input_ram]
if conf.use_externel:
output = Dense(conf.predict_length, activation="relu")(output)
output = MyInverseReshape(conf.batch_size)(output)
input_e, output_e = self.__E_input_output(conf, arm_shape)
if isinstance(input_e, list):
inputs += input_e
else:
inputs += [input_e]
if conf.use_matrix_fuse:
outputs = [matrixLayer()(output)]
outputs.append(matrixLayer()(output_e))
output = Add()(outputs)
else:
output = Add()([output, output_e])
output = Activation("tanh")(output)
else:
output = Dense(conf.predict_length, activation="tanh")(output)
output = MyInverseReshape(conf.batch_size)(output)
model = Model(inputs=inputs, outputs=output)
return model
def RNN_model(self, conf, arm_shape):
road_num = arm_shape[0]
input_x = Input((road_num, conf.observe_length, 1))
output = MyReshape(conf.batch_size)(input_x)
# output = SimpleRNN(32, return_sequences=True)(output)
output = SimpleRNN(conf.observe_length)(output)
# output = Dropout(0.1)(output)
output = Dense(conf.predict_length, activation="tanh")(output)
output = MyInverseReshape(conf.batch_size)(output)
model = Model(inputs=input_x, outputs=output)
return model
def __E_input_output(self, conf, arm_shape, activation="tanh"):
road_num = arm_shape[0]
if conf.observe_p != 0:
input_x1 = Input((road_num, conf.observe_p))
output1 = MyReshape(conf.batch_size)(input_x1)
output1 = Dense(conf.observe_p + 1, activation="relu")(output1)
if conf.observe_t != 0:
input_x2 = Input((road_num, conf.observe_t))
output2 = MyReshape(conf.batch_size)(input_x2)
output2 = Dense(conf.observe_t + 1, activation="relu")(output2)
if conf.observe_p != 0:
if conf.observe_t != 0:
output = Concatenate()([output1, output2])
input_x = [input_x1, input_x2]
else:
output = output1
input_x = input_x1
else:
output = output2
input_x = input_x2
output = Dense(conf.predict_length, activation=activation)(output)
output = MyInverseReshape(conf.batch_size)(output)
input_x3 = Input((conf.predict_length, 37)) # 37 is externel dim
if isinstance(input_x, list):
input_x += [input_x3]
else:
input_x = [input_x, input_x3]
output_3 = MyReshape(conf.batch_size)(input_x3)
output_3 = Dense(road_num, activation=activation)(output_3)
output_3 = MyInverseReshape(conf.batch_size)(output_3)
output_3 = Reshape((road_num, conf.predict_length))(output_3)
output = Add()([output, output_3])
return input_x, output
def CRNN_model(self, conf, arm_shape):
road_num = arm_shape[0]
input_x = Input((road_num, conf.observe_length, 1))
output = Conv2D(32, (2, 2), strides=(1, 1), padding="same")(input_x)
output = MaxPooling2D(pool_size=(1, 2))(output)
output = Activation(activation="relu")(output)
output = Conv2D(16, (2, 2), strides=(1, 1), padding="same")(output)
# pool2 = AveragePooling2D(pool_size=(1,2))(conv2)
# pool2 = Activation(activation="sigmoid")(conv2)
# conv3 = Conv2D(4, (2, 2), strides=(1, 1), padding="same")(pool2)
# pool3 = AveragePooling2D(pool_size=(1, 2))(conv3)
output = Activation(activation="relu")(output)
output = MyReshape(conf.batch_size)(output)
output = SimpleRNN(5)(output)
output = Dense(conf.predict_length)(output)
output = MyInverseReshape(conf.batch_size)(output)
# f = Flatten()(pool3)
# output = Dense(road_num * conf.predict_length, activation="sigmoid")(f)
# output = Reshape((road_num, conf.predict_length))(output)
model = Model(inputs=input_x, outputs=output)
return model