actor.add(Dense(8))
actor.add(Activation('relu'))
actor.add(Dense(nb_actions))
actor.add(Activation('sigmoid'))
action_input = Input(shape=(nb_actions,), name='action_input')
observation_input = Input(shape=(1,) + (11,), name='observation_input')
flattened_observation = Flatten()(observation_input)
x = Concatenate()([action_input, flattened_observation])
x = Dense(32)(x)
x = Activation('relu')(x)
x = Dense(32)(x)
x = Activation('relu')(x)
x = Dense(32)(x)
x = Activation('relu')(x)
x = Dense(1)(x)
x = Activation('linear')(x)
critic = Model(inputs=[action_input, observation_input], outputs=x)
memory = SequentialMemory(limit=100000, window_length=1)
random_process = OrnsteinUhlenbeckProcess(size=nb_actions, theta=.15, mu=0., sigma=.1)
agent = DDPGAgent(nb_actions=nb_actions, actor=actor, critic=critic, critic_action_input=action_input,
memory=memory, nb_steps_warmup_critic=100, nb_steps_warmup_actor=10,
random_process=random_process, gamma=.995, target_model_update=1e-3)
agent.compile(Adam(lr=.0005, clipnorm=1.), metrics=['mae'])
agent.fit(env, nb_steps=10000, visualize=False, verbose=0, nb_max_episode_steps=95)
#agent.save_weights('weights/ddpg_{}_weights.h5f'.format("stormwater"), overwrite=True)
agent.test(env, nb_episodes=15, visualize=False, nb_max_episode_steps=95, plt="")
env=env,
memory=(memory, episode_mem),
critic_goal_input=goal_input,
delta_clip=1.0,
nb_steps_warmup_critic=1000,
nb_steps_warmup_actor=1000,
random_process=random_process,
gamma=0.98,
target_model_update=0.001,
val_goal=args.val_goal,
vectorized=args.vectorized)
agent.compile([Adam(lr=1e-4), Adam(lr=1e-3)], metrics=['mae'])
agent.fit(env,
nb_steps=200000,
folder_to_save=folder_to_save,
visualize=False,
verbose=1,
reward_type=reward_type,
nb_max_episode_steps=50,
max_tau=12)
agent.save_weights(os.path.join(
folder_to_save, 'ddpg_{}_weights_{}.h5f'.format(ENV_NAME,
reward_type)),
overwrite=True)
sys.exit()
# Finally, evaluate our algorithm for 5 episodes.
agent.test(env, nb_episodes=5, visualize=True, nb_max_episode_steps=50)
