def finalLayer(self, y, n_iters=1, learner_size=200):
print "Final Layer"
sigmoid = Layers.SigmoidLayer(self.X.shape[1],
learner_size,
noise=Noise.GaussianNoise(0.1))
softmax = Layers.SoftmaxLayer(learner_size, y.shape[1])
trainer = Trainer()
sigmoid, softmax = trainer.train([sigmoid, softmax], self.X, y,
n_iters)
self.Layers.append(sigmoid)
self.Layers.append(softmax)
if done:
print(i, "---", ep_reward)
break
if i % 20 == 0:
pass
#video_saver.release()
if __name__ == '__main__':
batch_size = 64
tf_config = tf.ConfigProto()
#tf_config.gpu_options.per_process_gpu_memory_fraction = 0.6
tf_config.gpu_options.allow_growth = True
saver = tf.train.Saver()
env = gym.make('InvertedPendulum-v2')
#print(env.action_space.high)
with tf.Session(config=tf_config) as sess:
actor = Actor.Actor(sess, [4], 1, 0.0001, 0.001, batch_size)
critic = Critic.Critic(sess, [4], 1, 0.001, 0.001, 0.99,
actor.get_num_trainable_vars())
actor_noise = Noise.GaussianNoise()
train_feature(sess, env, actor, critic, actor_noise, batch_size, saver)
with tf.Session(config=tf_config) as sess:
#state_dim : 1d, action_spec : scalar
if actor_type == "basic":
actor = Actor.Actor(sess, state_dim, action_dim, actor_lr, tau,
batch_size)
elif actor_type == "rnn":
actor = RNNActor.Actor(sess, state_dim, action_dim, actor_lr, tau,
batch_size, num_of_action)
critic = Critic.Critic(sess, state_dim, action_dim, critic_lr, tau,
gamma, actor.get_num_trainable_vars(),
critic_reg_weight)
if noise_type == "gaussian":
actor_noise = Noise.GaussianNoise(action_dim=action_dim,
sigma=sigma)
elif noise_type == "ou":
actor_noise = Noise.OrnsteinUhlenbeckActionNoise(mu=np.zeros(
[int(action_dim / num_of_action)]),
sigma=sigma)
exp_detail = utils.experiment_detail_saver(domain_name, task_name,
step_size, actor_lr,
critic_lr, tau, gamma,
sigma, batch_size,
critic_reg_weight)
print(exp_detail)
utils.append_file_writer(video_dir, "experiment_detail.txt", "num of action : " \
+ str(num_of_action) + "\n")
print("num of action : " + str(num_of_action))
critic_lr = 1e-3
tau = 5e-3
gamma = 0.99
sigma = 0.2
critic_reg_weight = 0.0
noise_type = "ou"
assert noise_type in ["ou","gaussian"]
with tf.Session(config=tf_config) as sess:
#state_dim : 1d, action_spec : scalar
actor = Actor.Actor(sess, state_dim, env.action_spec().shape[0], actor_lr, tau, batch_size)
critic = Critic.Critic(sess, state_dim, env.action_spec().shape[0], critic_lr, tau, gamma, actor.get_num_trainable_vars(),critic_reg_weight)
if noise_type == "gaussian":
actor_noise = Noise.GaussianNoise(action_dim=env.action_spec().shape[0],sigma=sigma)
elif noise_type == "ou":
actor_noise = Noise.OrnsteinUhlenbeckActionNoise(mu=np.zeros([env.action_spec().shape[0]]), sigma=sigma)
exp_detail = utils.experiment_detail_saver(
domain_name, task_name, step_size,
actor_lr, critic_lr, tau,
gamma, sigma, batch_size,
critic_reg_weight)
print(exp_detail)
utils.append_file_writer(video_dir, "experiment_detail.txt", "Critic origin type : "\
+critic.critic_origin_type+"\n")
utils.append_file_writer(video_dir, "experiment_detail.txt", "Noise type : " \
+ noise_type + "\n")