Merge branch 'master' of https://github.com/ShangtongZhang/DeepRL

async_worker/actor_critic.py

@@ -33,6 +33,7 @@ class AdvantageActorCritic:
 
             steps += 1
             total_reward += reward
+            reward = config.reward_shift_fn(reward)
 
             if deterministic:
                 if terminal:

async_worker/n_step_q.py

@@ -34,6 +34,7 @@ class NStepQLearning:
 
             steps += 1
             total_reward += reward
+            reward = config.reward_shift_fn(reward)
 
             if deterministic:
                 if terminal:

async_worker/one_step_q.py

@@ -34,6 +34,7 @@ class OneStepQLearning:
 
             steps += 1
             total_reward += reward
+            reward = config.reward_shift_fn(reward)
 
             if deterministic:
                 if terminal:

async_worker/one_step_sarsa.py

@@ -37,6 +37,7 @@ class OneStepSarsa:
 
             steps += 1
             total_reward += reward
+            reward = config.reward_shift_fn(reward)
 
             if deterministic:
                 if terminal:

async_worker/ppo.py

@@ -104,6 +104,7 @@ class ProximalPolicyOptimization:
                 advs.append(cum_adv)
             advantages = advs[::-1]
             returns = list(returns)
+
             replay.feed([states, actions, returns, advantages])
 
         batched_rewards /= batched_episode