da21952253a0ea6797faa7c59a3b6e9f117b58a6,slm_lab/agent/algorithm/dqn.py,VanillaDQN,init_algo_params,#VanillaDQN#,58

Before Change

        algorithm_spec = self.agent.spec["algorithm"]
        self.action_policy = act_fns[algorithm_spec["action_policy"]]
        self.action_policy_update = act_update_fns[algorithm_spec["action_policy_update"]]
        util.set_attr(self, _.pick(algorithm_spec, [
            // explore_var is epsilon, tau or etc. depending on the action policy
            // these control the trade off between exploration and exploitaton
            "explore_var_start", "explore_var_end", "explore_anneal_epi",
            "gamma",  // the discount factor
            "training_min_timestep",  // how long before starting training
            "training_frequency",  // how often to train (once a few timesteps)
            "training_epoch",  // how many batches to train each time
            "training_iters_per_batch",  // how many times to train each batch
        ]))
        self.nanflat_explore_var_a = [
            self.explore_var_start] * self.agent.body_num

After Change

            "training_epoch",  // how many batches to train each time
            "training_iters_per_batch",  // how many times to train each batch
        ]))
        util.set_attr(self, _.pick(net_spec, [
            "decay_lr", "decay_lr_timestep", "start_decay_lr_timestep",
        ]))
        self.nanflat_explore_var_a = [
            self.explore_var_start] * self.agent.body_num

Instances