from baselines.trpo_mpi import defaults
from baselines.common import set_global_seeds, tf_util as U
from baselines.common.input import observation_placeholder
from baselines.common.models import mlp
from baselines.common.policies import build_policy
from baselines.common.vec_env.dummy_vec_env import DummyVecEnv
U.get_session( config=tf.ConfigProto(
allow_soft_placement = True,
inter_op_parallelism_threads = 1,
intra_op_parallelism_threads = 1) )
U.initialize()
# Get dictionary from baselines/trpo_mpi/defaults
defaults = defaults.mara_mlp()
# Create needed folders
try:
logdir = defaults['trained_path'].split('checkpoints')[0] + 'results' + defaults['trained_path'].split('checkpoints')[1]
except:
logdir = '/tmp/ros2learn/' + defaults['env_name'] + '/trpo_mpi_results/'
finally:
logger.configure( os.path.abspath(logdir) )
csvdir = logdir + "csv/"
csv_files = [csvdir + "det_obs.csv", csvdir + "det_acs.csv", csvdir + "det_rew.csv"]
if not os.path.exists(csvdir):
os.makedirs(csvdir)
else:
for f in csv_files:
from baselines.common.models import mlp
from baselines.common.vec_env.dummy_vec_env import DummyVecEnv
def make_env():
env = gym.make(alg_kwargs['env_name'])
env.set_episode_size(alg_kwargs['timesteps_per_batch'])
env = bench.Monitor(env,
logger.get_dir() and os.path.join(logger.get_dir()),
allow_early_resets=True)
return env
# Get dictionary from baselines/trpo_mpi/defaults
alg_kwargs = defaults.mara_mlp()
# Create needed folders
timedate = datetime.now().strftime('%Y-%m-%d_%Hh%Mmin')
logdir = '/tmp/ros2learn/' + alg_kwargs['env_name'] + '/trpo_mpi/' + timedate
# Generate tensorboard file
format_strs = os.getenv('MARA_LOG_FORMAT',
'stdout,log,csv,tensorboard').split(',')
logger.configure(os.path.abspath(logdir), format_strs)
with open(logger.get_dir() + "/parameters.txt", 'w') as out:
out.write('num_layers = ' + str(alg_kwargs['num_layers']) + '\n' +
'num_hidden = ' + str(alg_kwargs['num_hidden']) + '\n' +
'layer_norm = ' + str(alg_kwargs['layer_norm']) + '\n' +
'timesteps_per_batch = ' +