rlgym-learn-algos 0.2.4__cp39-cp39-win_amd64.whl → 0.2.5__cp39-cp39-win_amd64.whl

rlgym_learn_algos/ppo/gae_trajectory_processor_pure_python.py

@@ -117,14 +117,17 @@ class GAETrajectoryProcessorPurePython(
         else:
             avg_return = np.nan
             return_std = np.nan
-        avg_reward = reward_sum[0] / exp_len
+        avg_reward = reward_sum.item() / exp_len
+        average_episode_return = reward_sum.item() / len(trajectories)
         trajectory_processor_data = GAETrajectoryProcessorData(
-            average_undiscounted_episodic_return=avg_reward,
+            average_undiscounted_episodic_return=average_episode_return,
             average_return=avg_return,
             return_standard_deviation=return_std,
+            average_reward=avg_reward
         )
         return (
             (
+                agent_ids,
                 observations,
                 actions,
                 torch.stack(log_probs_list).to(device=self.device),
@@ -146,7 +149,7 @@ class GAETrajectoryProcessorPurePython(
         self.max_returns_per_stats_increment = (
             config.trajectory_processor_config.max_returns_per_stats_increment
         )
-        self.dtype = np.dtype(config.dtype)
+        self.dtype = np.dtype(str(config.dtype).replace("torch.", ""))
         self.device = config.device
         self.checkpoint_load_folder = config.checkpoint_load_folder
         if self.checkpoint_load_folder is not None:

rlgym_learn_algos/ppo/ppo_learner.py

@@ -284,6 +284,15 @@ class PPOLearner(
                     batch_advantages,
                 ) = batch
                 batch_target_values = batch_values + batch_advantages
+                if self.config.learner_config.advantage_normalization:
+                    old_device = batch_advantages.device
+                    batch_advantages = batch_advantages.to(
+                        self.config.learner_config.device
+                    )
+                    std, mean = torch.std_mean(batch_advantages)
+                    batch_advantages = (batch_advantages - mean) / (std + 1e-8)
+                    batch_advantages = batch_advantages.to(old_device)
+
                 self.actor_optimizer.zero_grad()
                 self.critic_optimizer.zero_grad()
 
@@ -306,10 +315,6 @@ class PPOLearner(
                     advantages = batch_advantages[start:stop].to(
                         self.config.learner_config.device
                     )
-                    if self.config.learner_config.advantage_normalization:
-                        advantages = (advantages - torch.mean(advantages)) / (
-                            torch.std(advantages) + 1e-8
-                        )
                     old_probs = batch_old_probs[start:stop].to(
                         self.config.learner_config.device
                     )

{rlgym_learn_algos-0.2.4.dist-info → rlgym_learn_algos-0.2.5.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: rlgym-learn-algos
-Version: 0.2.4
+Version: 0.2.5
 Classifier: Programming Language :: Rust
 Classifier: Programming Language :: Python :: Implementation :: CPython
 Requires-Dist: pydantic>=2.8.2

{rlgym_learn_algos-0.2.4.dist-info → rlgym_learn_algos-0.2.5.dist-info}/RECORD

@@ -1,6 +1,6 @@
-rlgym_learn_algos-0.2.4.dist-info/METADATA,sha256=JtoluYo3NjNSC68j3U6S7xb79gvrSC74O3TNy_r8v8k,2431
-rlgym_learn_algos-0.2.4.dist-info/WHEEL,sha256=SqgFUQC3rmJBHv6XKhA7wOKmU2ih3aAfz0FSmR1FvyM,94
-rlgym_learn_algos-0.2.4.dist-info/licenses/LICENSE,sha256=HrhfyXIkWY2tGFK11kg7vPCqhgh5DcxleloqdhrpyMY,11558
+rlgym_learn_algos-0.2.5.dist-info/METADATA,sha256=_Hd86nz00PS2t06xX-wHch7-50wlZr8c_FUPjiPmh30,2431
+rlgym_learn_algos-0.2.5.dist-info/WHEEL,sha256=xZ8SzCfbQYp83o2fyM7cU7vfL9QFqpx9fTDOTKg2VCE,94
+rlgym_learn_algos-0.2.5.dist-info/licenses/LICENSE,sha256=HrhfyXIkWY2tGFK11kg7vPCqhgh5DcxleloqdhrpyMY,11558
 rlgym_learn_algos/__init__.py,sha256=C7cRdL4lZrpk3ge_4_lGAbGodqWJXM56FfgO0keRPAY,207
 rlgym_learn_algos/conversion/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 rlgym_learn_algos/conversion/convert_rlgym_ppo_checkpoint.py,sha256=A9nvzjp3DQNRNL5TAt-u3xE80JDIpYEDqAGNReHvFG0,908
@@ -18,15 +18,15 @@ rlgym_learn_algos/ppo/env_trajectories.py,sha256=gzQBRkzwZhlZeSvWL50cc8AOgBfsg5z
 rlgym_learn_algos/ppo/experience_buffer.py,sha256=xDm8NIMdErpv3GyWUBcTvzkLBQa8tW1TXb7OrKRDIu4,11059
 rlgym_learn_algos/ppo/experience_buffer_numpy.py,sha256=Apk4x-pfRnitKJPW6LBZyOPIhgeJs_5EG7BbTCqMwjk,4761
 rlgym_learn_algos/ppo/gae_trajectory_processor.py,sha256=JK958vasIIiuf3ALcFNlvBgGNhFshK8MhQJjwvxhrAM,5453
-rlgym_learn_algos/ppo/gae_trajectory_processor_pure_python.py,sha256=RpyDR6GQ1JXvwtoKkx5V3z3WvU9ElJdzfNtpPiZDaTc,6831
+rlgym_learn_algos/ppo/gae_trajectory_processor_pure_python.py,sha256=whnxjwrETiX_DSxjSy7ZidSqWHTTJAJ5F4pQqBVmDw4,7013
 rlgym_learn_algos/ppo/multi_discrete_actor.py,sha256=zSYeBBirjguSv_wO-peo06hioHiVhZQjnd-NYwJxmag,3127
 rlgym_learn_algos/ppo/ppo_agent_controller.py,sha256=-qAAlkv8qX2FccAT1Uvr7d40v6j4Vi1Rzl9EdmYRI50,25636
-rlgym_learn_algos/ppo/ppo_learner.py,sha256=Cbbuz0AMwPCmkQ1YPDdZLkbgZOdyrOLEx89Camn-nGE,15942
+rlgym_learn_algos/ppo/ppo_learner.py,sha256=NXj7_CDEpMIVpZwZLO91oYUVDiUPx_E4M7Tu_Zs7jzs,16195
 rlgym_learn_algos/ppo/ppo_metrics_logger.py,sha256=niW8xgQLEBCGgTaVyiE_JqsU6RTjV6h-JzM-7c3JT38,2868
 rlgym_learn_algos/ppo/trajectory.py,sha256=IIH_IG8B_HkyxRPf-YsCyF1jQqNjDx752hgzAehG25I,719
 rlgym_learn_algos/ppo/trajectory_processor.py,sha256=5eY_mNGjqIkhqnbKeaqDvqIWPdg6wD6Ai3fXH2WoXbw,2091
 rlgym_learn_algos/py.typed,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-rlgym_learn_algos/rlgym_learn_algos.cp39-win_amd64.pyd,sha256=Cm7rNAYop9cn-cIYQLzmF8G4NQegX92VlsyELPC8ehQ,416256
+rlgym_learn_algos/rlgym_learn_algos.cp39-win_amd64.pyd,sha256=kGXk-rmyRKfJ0j-DbVnbWDu5ba7eD8jJp3yzaDNPdRk,408576
 rlgym_learn_algos/rlgym_learn_algos.pyi,sha256=NwY-sDZWM06TUiKPzxpfH1Td6G6E8TdxtRPgBSh-PPE,1203
 rlgym_learn_algos/stateful_functions/__init__.py,sha256=QS0KYjuzagNkYiYllXQmjoJn14-G7KZawq1Zvwh8alY,236
 rlgym_learn_algos/stateful_functions/batch_reward_type_numpy_converter.py,sha256=1yte5qYyl9LWdClHZ_YsF7R9dJqQeYfINMdgNF_59Gs,767
@@ -36,4 +36,4 @@ rlgym_learn_algos/util/__init__.py,sha256=VPM6SN4T_625H9t30s9EiLeXiEEWgcyRVHa-LL
 rlgym_learn_algos/util/running_stats.py,sha256=0tiGFpKtHWzMa1CxM_ueBzd_ryX4bJBriC8MXcSLg8w,4479
 rlgym_learn_algos/util/torch_functions.py,sha256=_uAXhq1YYPneWI3_XXRYsSA3Hn1a8wGjUnI3m9UojdU,3411
 rlgym_learn_algos/util/torch_pydantic.py,sha256=5AbXQcfQtVgLRBSgCj0Hvi_H42WHLu4Oty4l_i22nAo,3531
-rlgym_learn_algos-0.2.4.dist-info/RECORD,,
+rlgym_learn_algos-0.2.5.dist-info/RECORD,,

{rlgym_learn_algos-0.2.4.dist-info → rlgym_learn_algos-0.2.5.dist-info}/WHEEL

@@ -1,4 +1,4 @@
 Wheel-Version: 1.0
-Generator: maturin (1.9.0)
+Generator: maturin (1.9.1)
 Root-Is-Purelib: false
 Tag: cp39-cp39-win_amd64