def _init(args):
env_name = args.env_name
print('Using environment %s' % env_name)
params_dict = {
'env_name': [env_name],
'rundir': [args.rundir],
'ent_wt': [args.trpo_ent],
'trpo_step': [args.trpo_step],
'hid_size': [args.hid_size],
'hid_layers': [args.hid_layers],
'many_runs': [args.repeat > 1]
}
if args.repeat > 1:
# stacked parallel thing doesn't work, bleh
warnings.warn(
"You're trying to use --repeat N for N > 1, but that "
"disables parallel sampling. This is probably going to be "
"heinously slow or something, use at own risk.")
# parallel_sampler.initialize(n_parallel=1)
# parallel_sampler.set_seed(1)
run_sweep_parallel(main, params_dict, repeat=args.repeat)
else:
parallel_sampler.initialize(n_parallel=8)
parallel_sampler.set_seed(1)
run_sweep_serial(main, params_dict, repeat=1)
default=None,
type=int,
help='size of each hidden layer in discriminator (or policy)')
parser.add_argument('--method',
choices=('airl', 'vairl', 'gail', 'vail'),
default=None,
help='IRL method to use')
if __name__ == "__main__":
args = parser.parse_args()
if args.method is None:
args.method = infer_method(args.pkl_path)
print('Assuming method %s' % args.method)
assert (args.trpo_anneal_init_ent is None) \
== (args.trpo_anneal_steps is None), \
"must supply both of --trpo-anneal-{init-ent,steps} or neither"
params_dict = {
'irl_pkl': [args.pkl_path],
'ent_wt': [args.trpo_ent],
'trpo_step': [args.trpo_step],
'trpo_anneal_steps': [args.trpo_anneal_steps],
'trpo_anneal_init_ent': [args.trpo_anneal_init_ent],
'method': [args.method],
'rundir': [args.rundir],
'hid_size': [args.hid_size],
'hid_layers': [args.hid_layers],
'switch_env': [args.switch_env],
}
# run_sweep_parallel(main, params_dict, repeat=3)
run_sweep_serial(main, params_dict, repeat=1)
