def learn_hoof_no_lambgam(env,
env_type,
timesteps_per_batch,
total_timesteps,
kl_range,
gamma_range,
lam_range,
num_kl=25,
num_gamma_lam=20,
**network_kwargs):
params = defaults.mujoco()
if gamma_range is 'fixed' and lam_range is 'fixed':
num_gamma_lam = 1
if kl_range is 'fixed':
num_kl = 1
run_hoof_no_lamgam(
network=params['network'],
env=env,
total_timesteps=total_timesteps,
timesteps_per_batch=int(timesteps_per_batch / env.num_envs),
kl_range=kl_range if kl_range is not 'fixed' else params['max_kl'],
gamma_range=gamma_range
if gamma_range is not 'fixed' else params['gamma'],
lam_range=lam_range if lam_range is not 'fixed' else params['lam'],
num_kl=num_kl,
num_gamma_lam=num_gamma_lam,
cg_iters=params['cg_iters'],
seed=None,
cg_damping=params['cg_damping'],
vf_stepsize=params['vf_stepsize'],
vf_iters=params['vf_iters'],
normalize_observations=params['normalize_observations'])
def learn_npg_variant(algo, env, env_type, timesteps_per_batch, total_timesteps, **network_kwargs):
params = defaults.mujoco()
params['network'] = mlp(num_hidden=64, num_layers=2)
run_pg(algo,
network=params['network'],
env=env,
total_timesteps=total_timesteps,
timesteps_per_batch=int(timesteps_per_batch/env.num_envs),
max_kl=params['max_kl'],
cg_iters=params['cg_iters'],
gamma=params['gamma'],
lam=params['lam'],
seed=None,
cg_damping=params['cg_damping'],
vf_stepsize=params['vf_stepsize'],
vf_iters=params['vf_iters'],
normalize_observations=params['normalize_observations']
)