def policy_fn(name, ob_space, ac_space):
if env.env.env.state_self_standardize:
return mlp_mirror_norms_policy.MlpMirrorNormsPolicy(
name=name,
ob_space=ob_space,
ac_space=ac_space,
hid_size=args.hsize,
num_hid_layers=args.layers,
gmm_comp=1,
mirror_loss=True,
observation_permutation=obs_per,
action_permutation=np.array([
5, 6, 7, 8, 9, 0.0001, 1, 2, 3, 4, -10, 11, -12, 17, 18,
19, 20, 13, 14, 15, 16
]))
else:
return mlp_mirror_policy.MlpMirrorPolicy(
name=name,
ob_space=ob_space,
ac_space=ac_space,
hid_size=args.hsize,
num_hid_layers=args.layers,
gmm_comp=1,
mirror_loss=True,
observation_permutation=obs_per,
action_permutation=np.array([
5, 6, 7, 8, 9, 0.0001, 1, 2, 3, 4, -10, 11, -12, 17, 18,
19, 20, 13, 14, 15, 16
]))
def policy_fn(name, ob_space, ac_space):
old_act_permute = [-86, 87, -88, 93, 94, 95, 96, 89, 90, 91, 92]
mus_act_l = np.arange(43)
mus_act_r = mus_act_l + 43
mus_act_l[0] = 0.001
act_permute = np.concatenate([mus_act_r, mus_act_l, old_act_permute])
if env.env.env.state_self_standardize:
return mlp_mirror_norms_policy.MlpMirrorNormsPolicy(
name=name,
ob_space=ob_space,
ac_space=ac_space,
hid_size=args.hsize,
num_hid_layers=args.layers,
gmm_comp=1,
mirror_loss=True,
observation_permutation=obs_per,
action_permutation=act_permute)
else:
return mlp_mirror_policy.MlpMirrorPolicy(
name=name,
ob_space=ob_space,
ac_space=ac_space,
hid_size=args.hsize,
num_hid_layers=args.layers,
gmm_comp=1,
mirror_loss=True,
observation_permutation=obs_per,
action_permutation=act_permute)