454f620fce3e6fe2f87a91bffd74667d1f8a94f9,slm_lab/agent/algorithm/actor_critic.py,ActorCritic,init_algorithm_params,#ActorCritic#,58

Before Change

    def init_algorithm_params(self):
        """Initialize other algorithm parameters"""
        if self.algorithm_spec["action_policy"] == "default":
            if self.body.is_discrete:
                self.algorithm_spec["action_policy"] = "softmax"
            else:
                self.algorithm_spec["action_policy"] = "gaussian"
        util.set_attr(self, self.algorithm_spec, [
            "action_policy",
            "gamma",  // the discount factor
            "add_entropy",

After Change

    def init_algorithm_params(self):
        """Initialize other algorithm parameters"""
        // set default
        util.set_attr(self, dict(
            action_pdtype="default",
            action_policy="default",
            action_policy_update="no_update",
            explore_var_start=np.nan,
            explore_var_end=np.nan,
            explore_anneal_epi=np.nan,
        ))
        util.set_attr(self, self.algorithm_spec, [
            "action_policy",
            // theoretically, AC does not have policy update; but in this implementation we have such option
            "action_policy_update",

Instances