def build_env(args):
ncpu = multiprocessing.cpu_count()
if sys.platform == 'darwin': ncpu //= 2
nenv = args.num_env or ncpu
alg = args.alg
seed = args.seed
env_type, env_id = get_env_type(args)
if env_type in {'atari', 'retro'}:
if alg == 'deepq':
env = make_env(env_id, env_type, seed=seed, wrapper_kwargs={'frame_stack': True})
elif alg == 'trpo_mpi':
env = make_env(env_id, env_type, seed=seed)
else:
print("are we here?")
frame_stack_size = 4
env = make_vec_env(env_id, env_type, nenv, seed, gamestate=args.gamestate, reward_scale=args.reward_scale)
env = VecFrameStack(env, frame_stack_size)
else:
config = tf.ConfigProto(allow_soft_placement=True,
intra_op_parallelism_threads=1,
inter_op_parallelism_threads=1)
config.gpu_options.allow_growth = True
get_session(config=config)
flatten_dict_observations = alg not in {'her'}
env = make_vec_env(env_id, env_type, args.num_env or 1, seed, reward_scale=args.reward_scale, flatten_dict_observations=flatten_dict_observations)
if env_type == 'mujoco':
env = VecNormalize(env, use_tf=True)
return env
def build_env(args):
"""
Build gym environment
Args:
args: (argparse.ArgumentParser) parsed command line arguments
"""
# number of cpu cores
ncpu = multiprocessing.cpu_count()
if sys.platform == 'darwin':
# for 'darwin' system halve the processor cores
ncpu //= 2
# set nenv = cpu if args.num_env == 0 or None
nenv = args.num_env or ncpu
# agent
alg = args.alg
# seed
seed = args.seed
# environment type & id
env_type, env_id = get_env_type(args)
if env_type in ['atari', 'retro']:
if alg == 'deepq':
env = make_env(env_id,
env_type,
seed=seed,
wrapper_kwargs={'frame_stack': True})
elif alg == 'trpo_mpi':
env = make_env(env_id, env_type, seed=seed)
else:
frame_stack_size = 4
env = make_env(env_id,
env_type,
nenv,
seed,
gamestate=args.gamestate,
reward_scale=args.reward_scale)
# vectorize environments (TBD)
# make environments for MuJoCo games (TBD)
return env
def build_env(cloth_cfg_path=None,
render_path=None,
start_state_path=None,
num_env=1,
seed=1,
alg='ddpg'):
"""Daniel: actually construct the env, using 'vector envs' for parallelism.
For now our cloth env can follow the non-atari and non-retro stuff, because
I don't think we need a similar kind of 'wrapping' that they do. Note that
`VecFrameStack` is needed to stack frames, e.g., in Atari we do 4 frame
stacking. Without that, the states would be size (84,84,1).
The non-`args` parameters here are for the cloth env.
"""
#Adi: Need to modify the next section because no 'args' parameter
ncpu = multiprocessing.cpu_count()
if sys.platform == 'darwin': ncpu //= 2
#nenv = args.num_env or ncpu
#alg = args.alg
#seed = args.seed
#env_type, env_id = get_env_type(args)
env_type = 'cloth'
env_id = 'cloth'
if env_type in {'atari', 'retro'}:
if alg == 'deepq':
env = make_env(env_id,
env_type,
seed=seed,
wrapper_kwargs={'frame_stack': True})
elif alg == 'trpo_mpi':
env = make_env(env_id, env_type, seed=seed)
else:
frame_stack_size = 4
env = make_vec_env(env_id,
env_type,
nenv,
seed,
gamestate=args.gamestate,
reward_scale=args.reward_scale)
env = VecFrameStack(env, frame_stack_size)
else:
config = tf.ConfigProto(allow_soft_placement=True,
intra_op_parallelism_threads=1,
inter_op_parallelism_threads=1)
config.gpu_options.allow_growth = True
get_session(config=config)
flatten_dict_observations = alg not in {'her'}
#Adi: I don't think we want to make a vector environment for now because it's causing a lot of trouble temporarily.. let's just start with a single non-vec env
#env = make_vec_env(env_id, env_type, num_env or 1, seed,
# reward_scale=1,
# flatten_dict_observations=flatten_dict_observations,
# cloth_cfg_path=cloth_cfg_path,
# render_path=render_path,
# start_state_path=start_state_path)
#Adi: I have to directly define a few more variables because we are now making a single environment instead of a vector environment
#Adi: These values are subject to change
mpi_rank = 0
subrank = 0
reward_scale = 1.0
gamestate = None
wrapper_kwargs = None
logger_dir = logger.get_dir()
env = make_env(env_id=env_id,
env_type=env_type,
mpi_rank=mpi_rank,
subrank=subrank,
seed=seed,
reward_scale=reward_scale,
gamestate=gamestate,
flatten_dict_observations=flatten_dict_observations,
wrapper_kwargs=wrapper_kwargs,
logger_dir=logger_dir,
cloth_cfg_path=cloth_cfg_path,
render_path=render_path,
start_state_path=start_state_path)
if env_type == 'mujoco':
env = VecNormalize(env)
return env