ddpg.py 文件源码

def create_actor_network(self, state_size, action_dim):
        """Create actor network."""
        print ("[MESSAGE] Build actor network.""")
        S = Input(shape=state_size)
        h_0 = Conv2D(32, (3, 3), padding="same",
                     kernel_regularizer=l2(0.0001),
                     activation="relu")(S)
        h_1 = Conv2D(32, (3, 3), padding="same",
                     kernel_regularizer=l2(0.0001),
                     activation="relu")(h_0)
        h_1 = AveragePooling2D(2, 2)(h_1)
        h_1 = Flatten()(h_1)
        h_1 = Dense(600, activation="relu")(h_1)
        A = Dense(action_dim, activation="softmax")(h_1)

        model = Model(inputs=S, outputs=A)

        return model, model.trainable_weights, S