def main():
args = parse_arg()
tf.reset_default_graph()
config = tf.ConfigProto()
config.gpu_options.allow_growth = True
with tf.Session(config=config) as sess:
env = gym.make('FeedingCooperation-v0')
set_global_seeds(int(args['random_seed']))
robot_actor_critic_entropy = Robot_Actor_Critic(
sess, float(args['actor_lr']), float(args['critic_lr']),
float(args['value_lr']), float(args['reg_factor']),
float(args['gamma']), float(args['tau']),
float(args['value_weight']), float(args['critic_weight']),
float(args['actor_weight']), float(args['all_lr']),
float(args['max_steps']), float(args['minibatch_size']))
human_actor_critic_entropy = Human_Actor_Critic(
sess, float(args['actor_lr']), float(args['critic_lr']),
float(args['value_lr']), float(args['reg_factor']),
float(args['gamma']), float(args['tau']),
float(args['value_weight']), float(args['critic_weight']),
float(args['actor_weight']), float(args['all_lr']),
float(args['max_steps']), float(args['minibatch_size']))
train(sess, env, args, robot_actor_critic_entropy,
human_actor_critic_entropy)
savepath = osp.join("my_model_sac_cop/", 'final')
os.makedirs(savepath, exist_ok=True)
savepath = osp.join(savepath, 'sacmodel')
save_state(savepath)
def make_vec_env(num_env, seed,copeoperation=False):
current_dir=os.getcwd()
logger_dir=osp.join(current_dir,
datetime.datetime.now().strftime("recoder-%Y-%m-%d-%H-%M-%S-%f"))
os.makedirs(logger_dir, exist_ok=True)
assert isinstance(logger_dir, str)
def make_thunk(rank,cops=False):
return lambda: create_env(
subrank=rank,
logger_dir=logger_dir,cop=cops)
set_global_seeds(seed)
return SubprocVecEnv([make_thunk(i,cops=copeoperation) for i in range(num_env)])
def main(args):
# arguments
arg_parser = common_arg_parser()
args, unknown_args = arg_parser.parse_known_args(args)
args.model = "_".join([args.agent, args.FA, str(args.T)])
# initialization
set_global_seeds(args.seed)
# os.environ["CUDA_VISIBLE_DEVICES"] = str(args.gpu_no)
# logger
logger.configure("./log" + args.data_path.split("/")[-2] + "/" + "_".join([
args.model,
datetime.now().strftime("%Y%m%d_%H%M%S"),
args.data_path.split("/")[-2],
str(args.learning_rate),
str(args.T),
str(args.ST),
str(args.gamma)
]))
logger.log("Training Model: " + args.model)
# environments
envs = get_objects(all_envs)
env = envs[args.environment](args)
# ipdb.set_trace()
# policy network
args.user_num = env.user_num
args.item_num = env.item_num
args.utype_num = env.utype_num
# ipdb.set_trace()
args.saved_path = os.path.join(
os.path.abspath("./"), "saved_path_" + args.data_path.split("/")[-2] +
"_" + str(args.FA) + "_" + str(args.learning_rate) + "_" +
str(args.agent) + "_" + str(args.seed))
nets = get_objects(all_FA)
fa = nets[args.FA].create_model_without_distributed(args)
logger.log("Hype-Parameters: " + str(args))
# # agents
agents = get_objects(all_agents)
agents[args.agent](env, fa, args).train()
def make_vec_env(env_id,
env_type,
num_env,
seed,
wrapper_kwargs=None,
start_index=0,
reward_scale=1.0,
flatten_dict_observations=True,
gamestate=None):
"""
Create a wrapped, monitored SubprocVecEnv for Atari and MuJoCo.
"""
wrapper_kwargs = wrapper_kwargs or {}
mpi_rank = MPI.COMM_WORLD.Get_rank() if MPI else 0
seed = seed + 10000 * mpi_rank if seed is not None else None
logger_dir = logger.get_dir()
def make_thunk(rank):
return lambda: make_env(env_id=env_id,
env_type=env_type,
mpi_rank=mpi_rank,
subrank=rank,
seed=seed,
reward_scale=reward_scale,
gamestate=gamestate,
flatten_dict_observations=
flatten_dict_observations,
wrapper_kwargs=wrapper_kwargs,
logger_dir=logger_dir)
set_global_seeds(seed)
if num_env > 1:
return SubprocVecEnv(
[make_thunk(i + start_index) for i in range(num_env)])
else:
return DummyVecEnv([make_thunk(start_index)])
import tensorflow as tf
import numpy as np
import gym
import my_envs
from util import set_global_seeds
import os
import time
set_global_seeds(1001)
with tf.Session() as sess:
directory = os.path.join(os.path.dirname(os.path.realpath(__file__)),
'my_model')
file_pos = os.path.join(directory, '00350')
meta_pos = os.path.join(file_pos, 'ppomodel.meta')
saver = tf.train.import_meta_graph(meta_pos)
saver.restore(sess, tf.train.latest_checkpoint(file_pos))
action_op = sess.graph.get_tensor_by_name('ppo_model/add_1:0')
input = sess.graph.get_tensor_by_name('ppo_model/ob:0')
num_env = action_op.shape[0]
env = gym.make('Feeding-v0')
env.play_show()
state = env.reset()
ep_reward = 0
total_step = 0
while True:
state = np.expand_dims(state, axis=0)
state = np.repeat(state, num_env, axis=0)
action = sess.run(action_op, feed_dict={input: state})
action = action[0, :]
def learn(env,
network,
seed=None,
lr=5e-4,
total_timesteps=100000,
buffer_size=50000,
exploration_fraction=0.1,
exploration_final_eps=0.02,
train_freq=1,
batch_size=32,
print_freq=100,
checkpoint_freq=10000,
checkpoint_path=None,
learning_starts=1000,
gamma=1.0,
target_network_update_freq=500,
prioritized_replay=False,
prioritized_replay_alpha=0.6,
prioritized_replay_beta0=0.4,
prioritized_replay_beta_iters=None,
prioritized_replay_eps=1e-6,
param_noise=False,
callback=None,
load_path=None,
save_path=None,
**network_kwargs):
"""Train a deepq model.
Parameters
-------
env: gym.Env
environment to train on
network: string or a function
neural network to use as a q function approximator. If string, has to be one of the names of registered models in baselines.common.models
(mlp, cnn, conv_only). If a function, should take an observation tensor and return a latent variable tensor, which
will be mapped to the Q function heads (see build_q_func in baselines.deepq.models for details on that)
seed: int or None
prng seed. The runs with the same seed "should" give the same results. If None, no seeding is used.
lr: float
learning rate for adam optimizer
total_timesteps: int
number of env steps to optimizer for
buffer_size: int
size of the replay buffer
exploration_fraction: float
fraction of entire training period over which the exploration rate is annealed
exploration_final_eps: float
final value of random action probability
train_freq: int
update the model every `train_freq` steps.
set to None to disable printing
batch_size: int
size of a batched sampled from replay buffer for training
print_freq: int
how often to print out training progress
set to None to disable printing
checkpoint_freq: int
how often to save the model. This is so that the best version is restored
at the end of the training. If you do not wish to restore the best version at
the end of the training set this variable to None.
learning_starts: int
how many steps of the model to collect transitions for before learning starts
gamma: float
discount factor
target_network_update_freq: int
update the target network every `target_network_update_freq` steps.
prioritized_replay: True
if True prioritized replay buffer will be used.
prioritized_replay_alpha: float
alpha parameter for prioritized replay buffer
prioritized_replay_beta0: float
initial value of beta for prioritized replay buffer
prioritized_replay_beta_iters: int
number of iterations over which beta will be annealed from initial value
to 1.0. If set to None equals to total_timesteps.
prioritized_replay_eps: float
epsilon to add to the TD errors when updating priorities.
param_noise: bool
whether or not to use parameter space noise (https://arxiv.org/abs/1706.01905)
callback: (locals, globals) -> None
function called at every steps with state of the algorithm.
If callback returns true training stops.
load_path: str
path to load the model from. (default: None)
**network_kwargs
additional keyword arguments to pass to the network builder.
Returns
-------
act: ActWrapper
Wrapper over act function. Adds ability to save it and load it.
See header of baselines/deepq/categorical.py for details on the act function.
"""
# Create all the functions necessary to train the model
set_global_seeds(seed)
q_func = build_q_func(network, **network_kwargs)
# capture the shape outside the closure so that the env object is not serialized
# by cloudpickle when serializing make_obs_ph
observation_space = env.observation_space
model = deepq.DEEPQ(q_func=q_func,
observation_shape=env.observation_space.shape,
num_actions=env.action_space.n,
lr=lr,
grad_norm_clipping=10,
gamma=gamma,
param_noise=param_noise)
if load_path is not None:
load_path = osp.expanduser(load_path)
ckpt = tf.train.Checkpoint(model=model)
manager = tf.train.CheckpointManager(ckpt, load_path, max_to_keep=None)
ckpt.restore(manager.latest_checkpoint)
print("Restoring from {}".format(manager.latest_checkpoint))
return model
# Create the replay buffer
if prioritized_replay:
replay_buffer = PrioritizedReplayBuffer(buffer_size,
alpha=prioritized_replay_alpha)
if prioritized_replay_beta_iters is None:
prioritized_replay_beta_iters = total_timesteps
beta_schedule = LinearSchedule(prioritized_replay_beta_iters,
initial_p=prioritized_replay_beta0,
final_p=1.0)
else:
replay_buffer = ReplayBuffer(buffer_size)
beta_schedule = None
# Create the schedule for exploration starting from 1.
exploration = LinearSchedule(schedule_timesteps=int(exploration_fraction *
total_timesteps),
initial_p=1.0,
final_p=exploration_final_eps)
model.update_target()
episode_rewards = [0.0]
saved_mean_reward = None
obs = env.reset()
# always mimic the vectorized env
if not isinstance(env, VecEnv):
obs = np.expand_dims(np.array(obs), axis=0)
reset = True
for t in range(total_timesteps):
if callback is not None:
if callback(locals(), globals()):
break
kwargs = {}
if not param_noise:
update_eps = tf.constant(exploration.value(t))
update_param_noise_threshold = 0.
else:
update_eps = tf.constant(0.)
# Compute the threshold such that the KL divergence between perturbed and non-perturbed
# policy is comparable to eps-greedy exploration with eps = exploration.value(t).
# See Appendix C.1 in Parameter Space Noise for Exploration, Plappert et al., 2017
# for detailed explanation.
update_param_noise_threshold = -np.log(1. - exploration.value(t) +
exploration.value(t) /
float(env.action_space.n))
kwargs['reset'] = reset
kwargs[
'update_param_noise_threshold'] = update_param_noise_threshold
kwargs['update_param_noise_scale'] = True
action, _, _, _ = model.step(tf.constant(obs),
update_eps=update_eps,
**kwargs)
action = action[0].numpy()
reset = False
new_obs, rew, done, _ = env.step(action)
# Store transition in the replay buffer.
if not isinstance(env, VecEnv):
new_obs = np.expand_dims(np.array(new_obs), axis=0)
replay_buffer.add(obs[0], action, rew, new_obs[0], float(done))
else:
replay_buffer.add(obs[0], action, rew[0], new_obs[0],
float(done[0]))
# # Store transition in the replay buffer.
# replay_buffer.add(obs, action, rew, new_obs, float(done))
obs = new_obs
episode_rewards[-1] += rew
if done:
obs = env.reset()
if not isinstance(env, VecEnv):
obs = np.expand_dims(np.array(obs), axis=0)
episode_rewards.append(0.0)
reset = True
if t > learning_starts and t % train_freq == 0:
# Minimize the error in Bellman's equation on a batch sampled from replay buffer.
if prioritized_replay:
experience = replay_buffer.sample(batch_size,
beta=beta_schedule.value(t))
(obses_t, actions, rewards, obses_tp1, dones, weights,
batch_idxes) = experience
else:
obses_t, actions, rewards, obses_tp1, dones = replay_buffer.sample(
batch_size)
weights, batch_idxes = np.ones_like(rewards), None
obses_t, obses_tp1 = tf.constant(obses_t), tf.constant(obses_tp1)
actions, rewards, dones = tf.constant(actions), tf.constant(
rewards), tf.constant(dones)
weights = tf.constant(weights)
td_errors = model.train(obses_t, actions, rewards, obses_tp1,
dones, weights)
if prioritized_replay:
new_priorities = np.abs(td_errors) + prioritized_replay_eps
replay_buffer.update_priorities(batch_idxes, new_priorities)
if t > learning_starts and t % target_network_update_freq == 0:
# Update target network periodically.
model.update_target()
mean_100ep_reward = round(np.mean(episode_rewards[-101:-1]), 1)
num_episodes = len(episode_rewards)
if done and print_freq is not None and len(
episode_rewards) % print_freq == 0:
logger.record_tabular("steps", t)
logger.record_tabular("episodes", num_episodes)
logger.record_tabular("mean 100 episode reward", mean_100ep_reward)
logger.record_tabular("% time spent exploring",
int(100 * exploration.value(t)))
logger.dump_tabular()
if save_path is not None:
save_path = osp.expanduser(save_path)
ckpt = tf.train.Checkpoint(model=model)
manager = tf.train.CheckpointManager(ckpt, save_path, max_to_keep=None)
manager.save()
return model
def learn(env,
total_timesteps,
seed=None,
nsteps=1024,
ent_coef=0.01,
lr=0.01,
vf_coef=0.5,
p_coef=1.0,
max_grad_norm=None,
gamma=0.99,
lam=0.95,
nminibatches=15,
noptepochs=4,
cliprange=0.2,
save_interval=100,
copeoperation=False,
human_ent_coef=0.01,
human_vf_coef=0.5,
human_p_coef=1.0):
set_global_seeds(seed)
sess = get_session()
global_summary = tf.summary.FileWriter(
'summaries/' + 'feeding' +
datetime.datetime.now().strftime('%d-%m-%y%H%M'), sess.graph)
if isinstance(lr, float): lr = constfn(lr)
else: assert callable(lr)
if isinstance(cliprange, float): cliprange = constfn(cliprange)
else: assert callable(cliprange)
# Get the nb of env
nenvs = env.num_envs
# Calculate the batch_size
nbatch = nenvs * nsteps
nbatch_train = nbatch // nminibatches
if copeoperation == True:
human_model = Model(env=env,
nbatch_act=nenvs,
nbatch_train=nbatch_train,
ent_coef=human_ent_coef,
vf_coef=human_vf_coef,
p_coef=human_p_coef,
max_grad_norm=max_grad_norm,
human=True,
robot=False)
robot_model = Model(env=env,
nbatch_act=nenvs,
nbatch_train=nbatch_train,
ent_coef=ent_coef,
vf_coef=vf_coef,
p_coef=p_coef,
max_grad_norm=max_grad_norm,
human=False,
robot=True)
if copeoperation == False:
model = Model(env=env,
nbatch_act=nenvs,
nbatch_train=nbatch_train,
ent_coef=ent_coef,
vf_coef=vf_coef,
p_coef=p_coef,
max_grad_norm=max_grad_norm)
initialize()
# Instantiate the runner object
if copeoperation == True:
runner = Runner(env=env,
model=None,
nsteps=nsteps,
gamma=gamma,
lam=lam,
human_model=human_model,
robot_model=robot_model)
if copeoperation == False:
runner = Runner(env=env,
model=model,
nsteps=nsteps,
gamma=gamma,
lam=lam)
epinfobuf = deque(maxlen=10) #recent 10 episode
pbar = tqdm(total=total_timesteps, dynamic_ncols=True)
tfirststart = time.perf_counter()
nupdates = total_timesteps // nbatch
for update in range(1, nupdates + 1):
assert nbatch % nminibatches == 0
# Start timer
frac = 1.0 - (update - 1.0) / nupdates
# Calculate the learning rate
lrnow = lr(frac)
# Calculate the cliprange
cliprangenow = cliprange(frac)
# Get minibatch
if copeoperation == False:
obs, returns, masks, actions, values, neglogpacs, epinfos = runner.run(
)
if copeoperation == True:
obs, human_returns, robot_returns, masks, human_actions, robot_actions, human_values, robot_values, human_neglogpacs, robot_neglogpacs, epinfos = runner.coop_run(
)
epinfobuf.extend(epinfos)
mblossvals = []
human_mblossvals = []
robot_mblossvals = []
inds = np.arange(nbatch)
for _ in range(noptepochs):
# Randomize the indexes
np.random.shuffle(inds)
for start in range(0, nbatch, nbatch_train):
end = start + nbatch_train
mbinds = inds[start:end]
if copeoperation == True:
human_slices = (arr[mbinds]
for arr in (obs[:, 24:], human_returns,
human_actions, human_values,
human_neglogpacs))
robot_slices = (arr[mbinds]
for arr in (obs[:, :24], robot_returns,
robot_actions, robot_values,
robot_neglogpacs))
human_mblossvals.append(
human_model.train(lrnow, cliprangenow, *human_slices))
robot_mblossvals.append(
robot_model.train(lrnow, cliprangenow, *robot_slices))
if copeoperation == False:
slices = (arr[mbinds] for arr in (obs, returns, actions,
values, neglogpacs))
mblossvals.append(model.train(lrnow, cliprangenow,
*slices)) #None
# Feedforward --> get losses --> update
if copeoperation == True:
human_lossvals = np.mean(human_mblossvals, axis=0)
robot_lossvals = np.mean(robot_mblossvals, axis=0)
if copeoperation == False:
lossvals = np.mean(mblossvals, axis=0)
summary = tf.Summary()
if copeoperation == True:
human_ev = explained_variance(human_values, human_returns)
robot_ev = explained_variance(robot_values, robot_returns)
if copeoperation == False:
ev = explained_variance(values, returns)
performance_r = np.mean([epinfo['r'] for epinfo in epinfobuf])
performance_len = np.mean([epinfo['l'] for epinfo in epinfobuf])
success_time = np.mean(
[epinfo['success_time'] for epinfo in epinfobuf])
fall_time = np.mean([epinfo['fall_time'] for epinfo in epinfobuf])
summary.value.add(tag='Perf/Reward', simple_value=performance_r)
summary.value.add(tag='Perf/episode_len', simple_value=performance_len)
summary.value.add(tag='Perf/success_time', simple_value=success_time)
summary.value.add(tag='Perf/fall_time', simple_value=fall_time)
if copeoperation == True:
summary.value.add(tag='Perf/human_explained_variance',
simple_value=float(human_ev))
summary.value.add(tag='Perf/robot_explained_variance',
simple_value=float(robot_ev))
if copeoperation == False:
summary.value.add(tag='Perf/explained_variance',
simple_value=float(ev))
if copeoperation == True:
for (human_lossval, human_lossname) in zip(human_lossvals,
human_model.loss_names):
if human_lossname == 'grad_norm':
summary.value.add(tag='grad/' + human_lossname,
simple_value=human_lossval)
else:
summary.value.add(tag='human_loss/' + human_lossname,
simple_value=human_lossval)
for (robot_lossval, robot_lossname) in zip(robot_lossvals,
robot_model.loss_names):
if robot_lossname == 'grad_norm':
summary.value.add(tag='grad/' + robot_lossname,
simple_value=robot_lossval)
else:
summary.value.add(tag='robot_loss/' + robot_lossname,
simple_value=robot_lossval)
if copeoperation == False:
for (lossval, lossname) in zip(lossvals, model.loss_names):
if lossname == 'grad_norm':
summary.value.add(tag='grad/' + lossname,
simple_value=lossval)
else:
summary.value.add(tag='loss/' + lossname,
simple_value=lossval)
global_summary.add_summary(summary, int(update * nbatch))
global_summary.flush()
print('finish one update')
if update % 10 == 0:
msg = 'step: {},episode reward: {},episode len: {},success_time: {},fall_time: {}'
pbar.update(update * nbatch)
pbar.set_description(
msg.format(update * nbatch, performance_r, performance_len,
success_time, fall_time))
if update % save_interval == 0:
tnow = time.perf_counter()
print('consume time', tnow - tfirststart)
if copeoperation == True:
savepath = osp.join("my_model_cop/", '%.5i' % update)
if copeoperation == False:
savepath = osp.join("my_model/", '%.5i' % update)
os.makedirs(savepath, exist_ok=True)
savepath = osp.join(savepath, 'ppomodel')
print('Saving to', savepath)
save_state(savepath)
pbar.close()
return model
def main():
config = Config()
set_global_seeds(config.seed)
env = gym.make(config.env_name)
env = CartPoleWrapper(env)
# with tf.device("/gpu:0"): # gpuを使用する場合
with tf.device("/cpu:0"):
ppo = PPO(num_actions=env.action_space.n,
input_shape=env.observation_space.shape,
config=config)
num_episodes = 0
episode_rewards = deque([0] * 100, maxlen=100)
memory = Memory(env.observation_space.shape, config)
reward_sum = 0
obs = env.reset()
for t in tqdm(range(config.num_update)):
# ===== get samples =====
for _ in range(config.num_step):
policy, value = ppo.step(obs[np.newaxis, :])
policy = policy.numpy()
action = np.random.choice(2, p=policy)
next_obs, reward, done, _ = env.step(action)
memory.add(obs, action, reward, done, value, policy[action])
obs = next_obs
reward_sum += reward
if done:
episode_rewards.append(env.steps)
num_episodes += 1
reward_sum = 0
obs = env.reset()
_, last_value = ppo.step(obs[np.newaxis, :])
memory.add(None, None, None, None, last_value, None)
# ===== make mini-batch and update parameters =====
memory.compute_gae()
for _ in range(config.num_epoch):
idxes = [idx for idx in range(config.num_step)]
random.shuffle(idxes)
for start in range(0, len(memory), config.batch_size):
minibatch_indexes = idxes[start:start + config.batch_size]
batch_obs, batch_act, batch_adv, batch_sum, batch_pi_old = memory.sample(
minibatch_indexes)
loss, policy_loss, value_loss, entropy_loss, policy, kl, frac = ppo.train(
batch_obs, batch_act, batch_pi_old, batch_adv, batch_sum)
memory.reset()
if t % config.log_step == 0:
logger.info("\nnum episodes: {}".format(num_episodes))
logger.info("loss: {}".format(loss.numpy()))
logger.info("policy loss: {}".format(policy_loss.numpy()))
logger.info("value loss: {}".format(value_loss.numpy()))
logger.info("entropy loss: {}".format(entropy_loss.numpy()))
logger.info("kl: {}".format(kl.numpy()))
logger.info("frac: {}".format(frac.numpy()))
logger.info("mean 100 episode reward: {}".format(
np.mean(episode_rewards)))
logger.info("max 100 episode reward: {}".format(
np.max(episode_rewards)))
logger.info("min 100 episode reward: {}".format(
np.min(episode_rewards)))
# ===== finish training =====
if config.play:
obs = env.reset()
while True:
action, _ = ppo.step(obs[np.newaxis, :])
action = int(action.numpy()[0])
obs, _, done, _ = env.step(action)
env.render()
if done:
obs = env.reset()