def __init__(self, hyperparams):
config = copy.deepcopy(ALG_MF)
config.update(hyperparams)
Algorithm.__init__(self, config)
self.policy_opt = self._hyperparams['policy_opt']['type'](
self._hyperparams['policy_opt'], self.dO, self.dU)
self.baseline = LinearFeatureBaseline()
def _advance_iteration_variables(self):
"""
Move all 'cur' variables to 'prev', reinitialize 'cur'
variables, and advance iteration counter.
"""
Algorithm._advance_iteration_variables(self)
for m in range(self.M):
self.cur[m].traj_info.last_kl_step = \
self.prev[m].traj_info.last_kl_step
self.cur[m].pol_info = copy.deepcopy(self.prev[m].pol_info)
def __init__(self, hyperparams):
config = copy.deepcopy(ALG_MDGPS)
config.update(hyperparams)
Algorithm.__init__(self, config)
policy_prior = self._hyperparams['policy_prior']
for m in range(self.M):
self.cur[m].pol_info = PolicyInfo(self._hyperparams)
self.cur[m].pol_info.policy_prior = \
policy_prior['type'](policy_prior)
self.policy_opt = self._hyperparams['policy_opt']['type'](
self._hyperparams['policy_opt'], self.dO, self.dU)