saymrwulf/stable-baselines3
1
0
Fork 0
mirror of https://github.com/saymrwulf/stable-baselines3.git synced 2026-09-17 22:30:59 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Separate feature extractor networks for DQN networks (#132)

Browse source 
* Separate feature extractor networks for DQN networks

* [ci skip] Bump version

Co-authored-by: Antonin RAFFIN <antonin.raffin@ensta.org>
This commit is contained in:
Anssi 2020-07-30 21:48:30 +03:00 committed by GitHub
parent 8f9aaaebe9
commit 77cb3dd0ab
No known key found for this signature in database
GPG key ID: 4AEE18F83AFDEB23
3 changed files with 7 additions and 7 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

3
docs/misc/changelog.rst
View file

@ -3,7 +3,7 @@
Changelog Changelog
========== ==========
Pre-Release 0.8.0a5 (WIP) Pre-Release 0.8.0a6 (WIP)
------------------------------ ------------------------------
Breaking Changes: Breaking Changes:
@ -33,6 +33,7 @@ Bug Fixes:
- Use ``cloudpickle.load`` instead of ``pickle.load`` in ``CloudpickleWrapper``. (@shwang) - Use ``cloudpickle.load`` instead of ``pickle.load`` in ``CloudpickleWrapper``. (@shwang)
- Fixed a bug with orthogonal initialization when `bias=False` in custom policy (@rk37) - Fixed a bug with orthogonal initialization when `bias=False` in custom policy (@rk37)
- Fixed approximate entropy calculation in PPO and A2C. (@andyshih12) - Fixed approximate entropy calculation in PPO and A2C. (@andyshih12)
- Fixed DQN target network sharing feature extractor with the main network.
Deprecations: Deprecations:
^^^^^^^^^^^^^ ^^^^^^^^^^^^^

9
stable_baselines3/dqn/policies.py
View file

@ -133,8 +133,6 @@ class DQNPolicy(BasePolicy):
else: else:
net_arch = [] net_arch = []
self.features_extractor = features_extractor_class(self.observation_space, **self.features_extractor_kwargs)
self.features_dim = self.features_extractor.features_dim
self.net_arch = net_arch self.net_arch = net_arch
self.activation_fn = activation_fn self.activation_fn = activation_fn
self.normalize_images = normalize_images self.normalize_images = normalize_images
@ -142,8 +140,6 @@ class DQNPolicy(BasePolicy):
self.net_args = { self.net_args = {
"observation_space": self.observation_space, "observation_space": self.observation_space,
"action_space": self.action_space, "action_space": self.action_space,
"features_extractor": self.features_extractor,
"features_dim": self.features_dim,
"net_arch": self.net_arch, "net_arch": self.net_arch,
"activation_fn": self.activation_fn, "activation_fn": self.activation_fn,
"normalize_images": normalize_images, "normalize_images": normalize_images,
@ -169,7 +165,10 @@ class DQNPolicy(BasePolicy):
self.optimizer = self.optimizer_class(self.parameters(), lr=lr_schedule(1), **self.optimizer_kwargs) self.optimizer = self.optimizer_class(self.parameters(), lr=lr_schedule(1), **self.optimizer_kwargs)
def make_q_net(self) -> QNetwork: def make_q_net(self) -> QNetwork:
return QNetwork(**self.net_args).to(self.device) # Make sure we always have separate networks for feature extractors etc
features_extractor = self.features_extractor_class(self.observation_space, **self.features_extractor_kwargs)
features_dim = features_extractor.features_dim
return QNetwork(features_extractor=features_extractor, features_dim=features_dim, **self.net_args).to(self.device)
def forward(self, obs: th.Tensor, deterministic: bool = True) -> th.Tensor: def forward(self, obs: th.Tensor, deterministic: bool = True) -> th.Tensor:
return self._predict(obs, deterministic=deterministic) return self._predict(obs, deterministic=deterministic)

2
stable_baselines3/version.txt
View file

@ -1 +1 @@
0.8.0a5 0.8.0a6