def run_experiment(mode="local", keys=None, params=dict()):
flags = FLAGS.__flags
flags = deepcopy(flags)
for k, v in params.items():
print('Modifying flags.%s from %r to %r' % (k, flags[k], v))
flags[k] = v
n_episodes = flags["max_episode"] # max episodes before termination
env = get_env(record_video=False, record_log=False, **flags)
policy = get_policy(env=env, **flags)
baseline = get_baseline(env=env, **flags)
qf = get_qf(env=env, **flags)
es = get_es(env=env, **flags)
info, _ = get_env_info(**flags)
max_path_length = info['horizon']
n_itr = int(
np.ceil(float(n_episodes * max_path_length) / flags['batch_size']))
algo = get_algo(n_itr=n_itr,
env=env,
policy=policy,
baseline=baseline,
qf=qf,
es=es,
max_path_length=max_path_length,
**flags)
#exp_prefix='%s-%s-%d'%(flags["exp"], flags["env_name"], flags["batch_size"])
exp_prefix = '%s' % (flags["exp"])
exp_name = get_annotations_string(keys=keys, **flags)
if flags["normalize_obs"]: flags["env_name"] += 'norm'
exp_name = '%s-%d--' % (flags["env_name"], flags["batch_size"]) + exp_name
log_dir = config.LOG_DIR + "/local/" + exp_prefix.replace(
"_", "-") + "/" + exp_name
if flags["seed"] is not None:
log_dir += '--s-%d' % flags["seed"]
if not flags["overwrite"] and osp.exists(log_dir):
ans = input("Overwrite %s?: (yes/no)" % log_dir)
if ans != 'yes': sys.exit(0)
run_experiment_lite(
algo.train(),
exp_prefix=exp_prefix,
exp_name=exp_name,
# Number of parallel workers for sampling
n_parallel=1,
snapshot_mode="last_best",
# Specifies the seed for the experiment. If this is not provided, a random seed
# will be used
seed=flags["seed"],
# plot=True,
terminate_machine=True,
sync_s3_pkl=True,
periodic_sync_interval=1200,
# terminate_machine=False,
# fast_code_sync=False,
mode=mode,
)
def set_experiment(mode="local", keys=None, params=dict()):
flags = FLAGS.__flags
# VGG: fix error handling flags after Tensorflow 1.4
for name in flags.keys():
flags[name] = flags[name].value
print('{}: {}'.format(name, flags[name]))
flags = deepcopy(flags)
for k, v in params.items():
print('Modifying flags.%s from %r to %r' % (k, flags[k], v))
flags[k] = v
n_episodes = flags["max_episode"] # max episodes before termination
info, _ = get_env_info(**flags)
max_path_length = 200 #info['horizon'] #200 **VGG: TODO: fix Gym env to include this
print('n_episodes: ', n_episodes)
print('max_path_length: ', max_path_length)
print('flags[batch_size]: ', flags['batch_size'])
print('flags[obs_space]: ', info['obs_space'])
n_itr = int(
np.ceil(float(n_episodes * max_path_length) / flags['batch_size']))
exp_prefix = '%s' % (flags["exp"])
exp_name = get_annotations_string(keys=keys, **flags)
if flags["normalize_obs"]: flags["env_name"] += 'norm'
exp_name = '%s-%d--' % (flags["env_name"], flags["batch_size"]) + exp_name
log_dir = config.LOG_DIR + "/local/" + exp_prefix.replace(
"_", "-") + "/" + exp_name
if flags["seed"] is not None:
log_dir += '--s-%d' % flags["seed"]
if not flags["overwrite"] and osp.exists(log_dir):
ans = input("Overwrite %s?: (yes/no)" % log_dir)
if ans != 'yes': sys.exit(0)
env = get_env(record_video=False, record_log=False, **flags)
policy = get_policy(env=env, info=info, **flags)
baseline = get_baseline(env=env, **flags)
qf = get_qf(env=env, info=info, **flags)
es = get_es(env=env, info=info, **flags)
algo = get_algo(n_itr=n_itr,
env=env,
policy=policy,
baseline=baseline,
qf=qf,
es=es,
max_path_length=max_path_length,
plot=True,
**flags)
return algo, dict(
exp_prefix=exp_prefix,
exp_name=exp_name,
mode=mode,
seed=flags["seed"],
)
def set_experiment(mode="local", keys=None, params=dict()):
flags = FLAGS.__flags
flags = deepcopy(flags)
for k, v in params.items():
print('Modifying flags.%s from %r to %r' % (k, flags[k], v))
flags[k] = v
n_episodes = flags["max_episode"] # max episodes before termination
info, _ = get_env_info(**flags)
max_path_length = info['horizon']
n_itr = int(
np.ceil(float(n_episodes * max_path_length) / flags['batch_size']))
exp_prefix = '%s' % (flags["exp"])
exp_name = get_annotations_string(keys=keys, **flags)
if flags["normalize_obs"]: flags["env_name"] += 'norm'
exp_name = '%s-%d--' % (flags["env_name"], flags["batch_size"]) + exp_name
log_dir = config.LOG_DIR + "/local/" + exp_prefix.replace(
"_", "-") + "/" + exp_name
if flags["seed"] is not None:
log_dir += '--s-%d' % flags["seed"]
# if not flags["overwrite"] and osp.exists(log_dir):
# ans = input("Overwrite %s?: (yes/no)"%log_dir)
# if ans != 'yes': sys.exit(0)
env = get_env(record_video=False, record_log=False, **flags)
policy = get_policy(env=env, info=info, **flags)
baseline = get_baseline(env=env, **flags)
qf = get_qf(env=env, info=info, **flags)
es = get_es(env=env, info=info, **flags)
pf = get_pf(env=env, info=info, **flags)
algo = get_algo(n_itr=n_itr,
env=env,
policy=policy,
baseline=baseline,
qf=qf,
es=es,
pf=pf,
max_path_length=max_path_length,
**flags)
return algo, dict(
exp_prefix=exp_prefix,
exp_name=exp_name,
mode=mode,
seed=flags["seed"],
)
def main(argv=None):
info, _ = get_env_info(**FLAGS.__flags)
pprint(info)