gaspersic/baselines
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Merge pull request #52 from farbeiza/patch-1

Browse Source 
Effectively apply weights from the replay buffer
This commit is contained in:
Szymon Sidor
2017-07-12 23:37:28 +02:00
committed by GitHub
parent 248aad1c3b d76cd1297a
commit 04cd0dcf64
1 changed files with 5 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
baselines/deepq/simple.py
View File

@@ -89,6 +89,7 @@ def learn(env,
gamma=1.0, gamma=1.0,
target_network_update_freq=500, target_network_update_freq=500,
prioritized_replay=False, prioritized_replay=False,
prioritized_importance_sampling=False,
prioritized_replay_alpha=0.6, prioritized_replay_alpha=0.6,
prioritized_replay_beta0=0.4, prioritized_replay_beta0=0.4,
prioritized_replay_beta_iters=None, prioritized_replay_beta_iters=None,
@@ -232,7 +233,10 @@ def learn(env,
else: else:
obses_t, actions, rewards, obses_tp1, dones = replay_buffer.sample(batch_size) obses_t, actions, rewards, obses_tp1, dones = replay_buffer.sample(batch_size)
weights, batch_idxes = np.ones_like(rewards), None weights, batch_idxes = np.ones_like(rewards), None
td_errors = train(obses_t, actions, rewards, obses_tp1, dones, np.ones_like(rewards)) if prioritized_importance_sampling:
td_errors = train(obses_t, actions, rewards, obses_tp1, dones, weights)
else:
td_errors = train(obses_t, actions, rewards, obses_tp1, dones, np.ones_like(rewards))
if prioritized_replay: if prioritized_replay:
new_priorities = np.abs(td_errors) + prioritized_replay_eps new_priorities = np.abs(td_errors) + prioritized_replay_eps
replay_buffer.update_priorities(batch_idxes, new_priorities) replay_buffer.update_priorities(batch_idxes, new_priorities)