rlgym-learn-algos 0.2.0__cp312-cp312-win32.whl → 0.2.2__cp312-cp312-win32.whl

rlgym_learn_algos/conversion/convert_rlgym_ppo_checkpoint.py

@@ -0,0 +1,27 @@
+import json
+import os
+import time
+from typing import Optional
+
+
+def convert_rlgym_ppo_checkpoint(
+    rlgym_ppo_checkpoint_folder: str, out_folder: Optional[str]
+):
+
+    if out_folder is None:
+        out_folder = f"rlgym_ppo_converted_checkpoint_{time.time_ns()}"
+    print(f"Saving converted checkpoint to folder {out_folder}")
+
+    os.makedirs(out_folder, exist_ok=True)
+
+    PPO_FILES = [
+        ("PPO_POLICY_OPTIMIZER.pt", "actor_optimizer.pt"),
+        ("PPO_POLICY.pt", "actor.pt"),
+        ("PPO_VALUE_NET_OPTIMIZER.pt", "critic_optimizer.pt"),
+        ("PPO_VALUE_NET.pt", "critic.pt"),
+    ]
+    os.makedirs(f"{out_folder}/ppo_learner", exist_ok=True)
+    for file in PPO_FILES:
+        with open(f"{rlgym_ppo_checkpoint_folder}/{file[0]}", "rb") as fin:
+            with open(f"{out_folder}/ppo_learner/{file[1]}", "wb") as fout:
+                fout.write(fin.read())

rlgym_learn_algos/logging/wandb_metrics_logger.py

@@ -164,7 +164,7 @@ class WandbMetricsLogger(
                 self.run_id = None
         except FileNotFoundError:
             print(
-                f"{self.config.agent_controller_name}: Tried to load from checkpoint, but checkpoint didn't contain a wandb run! A new run will be created based on the config values."
+                f"{self.config.agent_controller_name}: Tried to load wandb run from checkpoint using the file at location {str(os.path.join(self.config.checkpoint_load_folder, self.checkpoint_file_name))}, but there is no such file! A new run will be created based on the config values instead."
             )
             self.run_id = None
 

rlgym_learn_algos/ppo/experience_buffer.py

@@ -25,6 +25,7 @@ EXPERIENCE_BUFFER_FILE = "experience_buffer.pkl"
 class ExperienceBufferConfigModel(BaseModel, extra="forbid"):
     max_size: int = 100000
     device: PydanticTorchDevice = "auto"
+    save_experience_buffer_in_checkpoint: bool = True
     trajectory_processor_config: Dict[str, Any] = Field(default_factory=dict)
 
     @model_validator(mode="before")
@@ -41,21 +42,11 @@ class ExperienceBufferConfigModel(BaseModel, extra="forbid"):
                     data["trajectory_processor_config"] = data[
                         "trajectory_processor_config"
                     ].model_dump()
-            if "device" not in data or data["device"] == "auto":
-                data["device"] = get_device("auto")
+            if "device" not in data:
+                data["device"] = "auto"
+            data["device"] = get_device(data["device"])
         return data
 
-    # device: PydanticTorchDevice = "auto"
-
-    # @model_validator(mode="before")
-    # @classmethod
-    # def set_device(cls, data):
-    #     if isinstance(data, dict) and (
-    #         "device" not in data or data["device"] == "auto"
-    #     ):
-    #         data["device"] = get_device("auto")
-    #     return data
-
 
 @dataclass
 class DerivedExperienceBufferConfig:
@@ -139,6 +130,7 @@ class ExperienceBuffer(
         self.trajectory_processor.load(
             DerivedTrajectoryProcessorConfig(
                 trajectory_processor_config=trajectory_processor_config,
+                agent_controller_name=config.agent_controller_name,
                 dtype=config.dtype,
                 device=config.learner_device,
             )
@@ -171,26 +163,27 @@ class ExperienceBuffer(
             self.advantages = state_dict["advantages"]
         except FileNotFoundError:
             print(
-                f"{self.config.agent_controller_name}: Tried to load from checkpoint, but checkpoint didn't contain a saved experience buffer! A blank experience buffer will be used instead."
+                f"{self.config.agent_controller_name}: Tried to load experience buffer from checkpoint using the file at location {str(os.path.join(self.config.checkpoint_load_folder, EXPERIENCE_BUFFER_FILE))}, but there is no such file! A blank experience buffer will be used instead."
             )
 
     def save_checkpoint(self, folder_path):
         os.makedirs(folder_path, exist_ok=True)
-        with open(
-            os.path.join(folder_path, EXPERIENCE_BUFFER_FILE),
-            "wb",
-        ) as f:
-            pickle.dump(
-                {
-                    "agent_ids": self.agent_ids,
-                    "observations": self.observations,
-                    "actions": self.actions,
-                    "log_probs": self.log_probs,
-                    "values": self.values,
-                    "advantages": self.advantages,
-                },
-                f,
-            )
+        if self.config.experience_buffer_config.save_experience_buffer_in_checkpoint:
+            with open(
+                os.path.join(folder_path, EXPERIENCE_BUFFER_FILE),
+                "wb",
+            ) as f:
+                pickle.dump(
+                    {
+                        "agent_ids": self.agent_ids,
+                        "observations": self.observations,
+                        "actions": self.actions,
+                        "log_probs": self.log_probs,
+                        "values": self.values,
+                        "advantages": self.advantages,
+                    },
+                    f,
+                )
         self.trajectory_processor.save_checkpoint(folder_path)
 
     # TODO: update docs

rlgym_learn_algos/ppo/gae_trajectory_processor.py

@@ -115,6 +115,7 @@ class GAETrajectoryProcessor(
         self.max_returns_per_stats_increment = (
             config.trajectory_processor_config.max_returns_per_stats_increment
         )
+        self.agent_controller_name = config.agent_controller_name
         self.dtype = config.dtype
         self.device = config.device
         self.checkpoint_load_folder = config.checkpoint_load_folder
@@ -127,12 +128,17 @@ class GAETrajectoryProcessor(
         )
 
     def _load_from_checkpoint(self):
-        with open(
-            os.path.join(self.checkpoint_load_folder, TRAJECTORY_PROCESSOR_FILE),
-            "rt",
-        ) as f:
-            state = json.load(f)
-        self.return_stats.load_state_dict(state["return_running_stats"])
+        try:
+            with open(
+                os.path.join(self.checkpoint_load_folder, TRAJECTORY_PROCESSOR_FILE),
+                "rt",
+            ) as f:
+                state = json.load(f)
+            self.return_stats.load_state_dict(state["return_running_stats"])
+        except FileNotFoundError:
+            print(
+                f"{self.agent_controller_name}: Tried to load trajectory processor from checkpoint using the trajectory processor file at location {str(os.path.join(self.checkpoint_load_folder, TRAJECTORY_PROCESSOR_FILE))}, but there is no such file! Running stats will be initialized as if this were a new run instead."
+            )
 
     def save_checkpoint(self, folder_path):
         state = {

rlgym_learn_algos/ppo/ppo_agent_controller.py

@@ -253,6 +253,7 @@ class PPOAgentController(
         self.learner.load(
             DerivedPPOLearnerConfig(
                 learner_config=learner_config,
+                agent_controller_name=config.agent_controller_name,
                 obs_space=self.obs_space,
                 action_space=self.action_space,
                 checkpoint_load_folder=learner_checkpoint_load_folder,
@@ -304,33 +305,57 @@ class PPOAgentController(
         random.seed(self.config.base_config.random_seed)
 
     def _load_from_checkpoint(self):
-        with open(
-            os.path.join(
-                self.config.agent_controller_config.checkpoint_load_folder,
-                CURRENT_TRAJECTORIES_FILE,
-            ),
-            "rb",
-        ) as f:
-            current_trajectories: Dict[
-                int,
-                EnvTrajectories[AgentID, ActionType, ObsType, RewardType],
-            ] = pickle.load(f)
-        with open(
-            os.path.join(
-                self.config.agent_controller_config.checkpoint_load_folder,
-                ITERATION_SHARED_INFOS_FILE,
-            ),
-            "rb",
-        ) as f:
-            iteration_shared_infos: List[Dict[str, Any]] = pickle.load(f)
-        with open(
-            os.path.join(
-                self.config.agent_controller_config.checkpoint_load_folder,
-                PPO_AGENT_FILE,
-            ),
-            "rt",
-        ) as f:
-            state = json.load(f)
+        try:
+            with open(
+                os.path.join(
+                    self.config.agent_controller_config.checkpoint_load_folder,
+                    CURRENT_TRAJECTORIES_FILE,
+                ),
+                "rb",
+            ) as f:
+                current_trajectories: Dict[
+                    int,
+                    EnvTrajectories[AgentID, ActionType, ObsType, RewardType],
+                ] = pickle.load(f)
+        except FileNotFoundError:
+            print(
+                f"{self.config.agent_controller_name}: Tried to load current trajectories from checkpoint using the file at location {str(os.path.join(self.config.agent_controller_config.checkpoint_load_folder, CURRENT_TRAJECTORIES_FILE))}, but there is no such file! Current trajectories will be initialized as an empty dict instead."
+            )
+            current_trajectories = {}
+        try:
+            with open(
+                os.path.join(
+                    self.config.agent_controller_config.checkpoint_load_folder,
+                    ITERATION_SHARED_INFOS_FILE,
+                ),
+                "rb",
+            ) as f:
+                iteration_shared_infos: List[Dict[str, Any]] = pickle.load(f)
+        except FileNotFoundError:
+            print(
+                f"{self.config.agent_controller_name}: Tried to load iteration shared info data from checkpoint using the file at location {str(os.path.join(self.config.agent_controller_config.checkpoint_load_folder, ITERATION_SHARED_INFOS_FILE))}, but there is no such file! Iteration shared info data will be initialized as an empty list instead."
+            )
+            current_trajectories = {}
+        try:
+            with open(
+                os.path.join(
+                    self.config.agent_controller_config.checkpoint_load_folder,
+                    PPO_AGENT_FILE,
+                ),
+                "rt",
+            ) as f:
+                state = json.load(f)
+        except FileNotFoundError:
+            print(
+                f"{self.config.agent_controller_name}: Tried to load PPO agent miscellaneous state data from checkpoint using the file at location {str(os.path.join(self.config.agent_controller_config.checkpoint_load_folder, PPO_AGENT_FILE))}, but there is no such file! This state data will be initialized as if this were a new run instead."
+            )
+            state = {
+                "cur_iteration": 0,
+                "iteration_timesteps": 0,
+                "cumulative_timesteps": 0,
+                "iteration_start_time": time.perf_counter(),
+                "timestep_collection_start_time": time.perf_counter(),
+            }
 
         self.current_trajectories = current_trajectories
         self.iteration_shared_infos = iteration_shared_infos

rlgym_learn_algos/ppo/ppo_learner.py

@@ -16,12 +16,13 @@ from rlgym.api import (
     ObsType,
     RewardType,
 )
+from torch import nn as nn
+
 from rlgym_learn_algos.util.torch_functions import get_device
 from rlgym_learn_algos.util.torch_pydantic import (
     PydanticTorchDevice,
     PydanticTorchDtype,
 )
-from torch import nn as nn
 
 from .actor import Actor
 from .critic import Critic
@@ -38,43 +39,30 @@ class PPOLearnerConfigModel(BaseModel, extra="forbid"):
     clip_range: float = 0.2
     actor_lr: float = 3e-4
     critic_lr: float = 3e-4
+    advantage_normalization: bool = True
     device: PydanticTorchDevice = "auto"
+    cudnn_benchmark_mode: bool = True
 
     @model_validator(mode="before")
     @classmethod
     def set_device(cls, data):
-        if isinstance(data, dict) and (
-            "device" not in data or data["device"] == "auto"
-        ):
-            data["device"] = get_device("auto")
+        if isinstance(data, dict):
+            if "device" not in data:
+                data["device"] = "auto"
+            data["device"] = get_device(data["device"])
         return data
 
-
-# @model_validator(mode="before")
-# @classmethod
-# def set_agent_controllers_config(cls, data):
-#     if isinstance(data, LearningCoordinatorConfigModel):
-#         agent_controllers_config = {}
-#         for k, v in data.agent_controllers_config.items():
-#             if isinstance(v, BaseModel):
-#                 agent_controllers_config[k] = v.model_dump()
-#             else:
-#                 agent_controllers_config[k] = v
-#         data.agent_controllers_config = agent_controllers_config
-#     elif isinstance(data, dict) and "agent_controllers_config" in data:
-#         agent_controllers_config = {}
-#         for k, v in data["agent_controllers_config"].items():
-#             if isinstance(v, BaseModel):
-#                 agent_controllers_config[k] = v.model_dump()
-#             else:
-#                 agent_controllers_config[k] = v
-#         data["agent_controllers_config"] = agent_controllers_config
-#     return data
+    @model_validator(mode="after")
+    def validate_cudnn_benchmark(self):
+        if self.device.type != "cuda":
+            self.cudnn_benchmark_mode = False
+        return self
 
 
 @dataclass
 class DerivedPPOLearnerConfig:
     learner_config: PPOLearnerConfigModel
+    agent_controller_name: str
     obs_space: ObsSpaceType
     action_space: ActionSpaceType
     checkpoint_load_folder: Optional[str] = None
@@ -127,6 +115,12 @@ class PPOLearner(
     def load(self, config: DerivedPPOLearnerConfig):
         self.config = config
 
+        if (
+            config.learner_config.cudnn_benchmark_mode
+            and config.learner_config.device.type == "cuda"
+        ):
+            torch.backends.cudnn.benchmark = True
+
         self.actor = self.actor_factory(
             config.obs_space, config.action_space, config.learner_config.device
         )
@@ -155,16 +149,26 @@ class PPOLearner(
         total_parameters = actor_params_count + critic_params_count
 
         # Display in a structured manner
-        print("Trainable Parameters:")
-        print(f"{'Component':<10} {'Count':<10}")
+        print(f"{self.config.agent_controller_name}: Trainable Parameters:")
+        print(f"{self.config.agent_controller_name}: {'Component':<10} {'Count':<10}")
         print("-" * 20)
-        print(f"{'Policy':<10} {actor_params_count:<10}")
-        print(f"{'Critic':<10} {critic_params_count:<10}")
+        print(
+            f"{self.config.agent_controller_name}: {'Policy':<10} {actor_params_count:<10}"
+        )
+        print(
+            f"{self.config.agent_controller_name}: {'Critic':<10} {critic_params_count:<10}"
+        )
         print("-" * 20)
-        print(f"{'Total':<10} {total_parameters:<10}")
+        print(
+            f"{self.config.agent_controller_name}: {'Total':<10} {total_parameters:<10}"
+        )
 
-        print(f"Current Policy Learning Rate: {self.config.learner_config.actor_lr}")
-        print(f"Current Critic Learning Rate: {self.config.learner_config.critic_lr}")
+        print(
+            f"{self.config.agent_controller_name}: Current Policy Learning Rate: {self.config.learner_config.actor_lr}"
+        )
+        print(
+            f"{self.config.agent_controller_name}: Current Critic Learning Rate: {self.config.learner_config.critic_lr}"
+        )
         self.cumulative_model_updates = 0
 
         if self.config.checkpoint_load_folder is not None:
@@ -180,7 +184,7 @@ class PPOLearner(
 
         assert os.path.exists(
             self.config.checkpoint_load_folder
-        ), f"PPO Learner cannot find folder: {self.config.checkpoint_load_folder}"
+        ), f"{self.config.agent_controller_name}: PPO Learner cannot find folder: {self.config.checkpoint_load_folder}"
 
         self.actor.load_state_dict(
             torch.load(
@@ -206,11 +210,17 @@ class PPOLearner(
                 map_location=self.config.learner_config.device,
             )
         )
-        with open(
-            os.path.join(self.config.checkpoint_load_folder, MISC_STATE), "rt"
-        ) as f:
-            misc_state = json.load(f)
-            self.cumulative_model_updates = misc_state["cumulative_model_updates"]
+        try:
+            with open(
+                os.path.join(self.config.checkpoint_load_folder, MISC_STATE), "rt"
+            ) as f:
+                misc_state = json.load(f)
+                self.cumulative_model_updates = misc_state["cumulative_model_updates"]
+        except FileNotFoundError:
+            print(
+                f"{self.config.agent_controller_name}: Tried to load the PPO learner's misc state from the file at location {str(os.path.join(self.config.checkpoint_load_folder, MISC_STATE))}, but there is no such file! Miscellaneous stats will be initialized as if this were a new run instead."
+            )
+            self.cumulative_model_updates = 0
 
     def save_checkpoint(self, folder_path):
         os.makedirs(folder_path, exist_ok=True)
@@ -296,6 +306,10 @@ 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/ppo/trajectory_processor.py

@@ -16,6 +16,7 @@ TRAJECTORY_PROCESSOR_FILE = "trajectory_processor.json"
 @dataclass
 class DerivedTrajectoryProcessorConfig(Generic[TrajectoryProcessorConfig]):
     trajectory_processor_config: TrajectoryProcessorConfig
+    agent_controller_name: str
     dtype: dtype
     device: device
     checkpoint_load_folder: Optional[str] = None

rlgym_learn_algos/util/torch_functions.py

@@ -13,7 +13,6 @@ import torch.nn as nn
 
 def get_device(device: str):
     if device in ["auto", "gpu"] and torch.cuda.is_available():
-        torch.backends.cudnn.benchmark = True
         return "cuda:0"
     elif device == "auto" and not torch.cuda.is_available():
         return "cpu"

rlgym_learn_algos/util/torch_pydantic.py

@@ -42,7 +42,7 @@ device_str_regex = (
             "privateuseone",
         ]
     )
-    + ")(:\d+)"
+    + ")(:\d+)?"
 )
 
 

{rlgym_learn_algos-0.2.0.dist-info → rlgym_learn_algos-0.2.2.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: rlgym-learn-algos
-Version: 0.2.0
+Version: 0.2.2
 Classifier: Programming Language :: Rust
 Classifier: Programming Language :: Python :: Implementation :: CPython
 Requires-Dist: pydantic>=2.8.2

{rlgym_learn_algos-0.2.0.dist-info → rlgym_learn_algos-0.2.2.dist-info}/RECORD

@@ -1,11 +1,13 @@
-rlgym_learn_algos-0.2.0.dist-info/METADATA,sha256=lhDW1yMI9RhHxSWjTN8WFWJ6O_WlgtFsm8NI1e-GLgI,2431
-rlgym_learn_algos-0.2.0.dist-info/WHEEL,sha256=iip_4jp1VbEiG4s9Isod9Q3MUK3z1pesoEwCO-vGFn0,92
-rlgym_learn_algos-0.2.0.dist-info/licenses/LICENSE,sha256=HrhfyXIkWY2tGFK11kg7vPCqhgh5DcxleloqdhrpyMY,11558
+rlgym_learn_algos-0.2.2.dist-info/METADATA,sha256=4wwr9xqqVWvZ7HYM4cumHiRdz79gkixfpe11b4MyvSU,2431
+rlgym_learn_algos-0.2.2.dist-info/WHEEL,sha256=_hRredGV19BUVIQG50bodYTQ5yHFvjyY42Je1tLhKbY,92
+rlgym_learn_algos-0.2.2.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
 rlgym_learn_algos/logging/__init__.py,sha256=ouItskWI4ItuoFdL--rt9YXCt7MasA473lYPhmJnrFA,423
 rlgym_learn_algos/logging/dict_metrics_logger.py,sha256=qmqr0HSiHpm5rjyxfAdmXOeBSbgP_t36-e-enpOccnE,1991
 rlgym_learn_algos/logging/metrics_logger.py,sha256=0l69GSSrxRcPm0xAjvF7yEIis7jGNu70unXu3hnK0XE,4122
-rlgym_learn_algos/logging/wandb_metrics_logger.py,sha256=Kxi8y-nfoh3EI_OqLm4pDS-zhUWEjkS8F4TdD01dr9U,6939
+rlgym_learn_algos/logging/wandb_metrics_logger.py,sha256=OXyOJzGP4zz0mgy3-FAvR6LW7aZet3Ii8CsI5csw4c4,7051
 rlgym_learn_algos/ppo/__init__.py,sha256=o6B8wCRfeyipSNEGJFyB3SHYmxUytaQelX2zsted5cg,1184
 rlgym_learn_algos/ppo/actor.py,sha256=LZevg0kqRrb4PwF05ePK9b1JIBX04YkWjsPs7swZ9JY,1767
 rlgym_learn_algos/ppo/basic_critic.py,sha256=oyyo8x9K6mi2BsbA6_tRy2Av8Pimb35WspJkPpe8XdQ,1022
@@ -13,18 +15,18 @@ rlgym_learn_algos/ppo/continuous_actor.py,sha256=1vdBUw2mQNFNu6A6ZrAztBjd4DmwjGk
 rlgym_learn_algos/ppo/critic.py,sha256=RB89WtiN52BEq5QCpGAPrASUnasac-Bpg7B0lM3UXHw,689
 rlgym_learn_algos/ppo/discrete_actor.py,sha256=Nuc3EndIQud3NGrkBIQgy-Z-okhXVrj6p6okSGD1KNY,2620
 rlgym_learn_algos/ppo/env_trajectories.py,sha256=gzQBRkzwZhlZeSvWL50cc8AOgBfsg5zUys0aTJj6aZU,3775
-rlgym_learn_algos/ppo/experience_buffer.py,sha256=py7kwhRJFsPx5lyvcUVywLAsu5zbU_0wV_52Fb6Kb_4,11012
+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=r-o5ajNSTNr5nZxsUc17KMuZR6c4l4NHHTIs2-WbMgE,4956
+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/multi_discrete_actor.py,sha256=zSYeBBirjguSv_wO-peo06hioHiVhZQjnd-NYwJxmag,3127
-rlgym_learn_algos/ppo/ppo_agent_controller.py,sha256=CH-xpO2mOCwe4iu_n9wPPsqLxEPYcbQhnvcJYs46qgM,23270
-rlgym_learn_algos/ppo/ppo_learner.py,sha256=z14GaL52mx7b20mQsuOLFXlpYVlR0_9Nn5HinImWaLY,15295
+rlgym_learn_algos/ppo/ppo_agent_controller.py,sha256=h0UR-o2k-_LyeFTzvII3HQHHWyeMJewqLlca8ThtyfA,25105
+rlgym_learn_algos/ppo/ppo_learner.py,sha256=Cbbuz0AMwPCmkQ1YPDdZLkbgZOdyrOLEx89Camn-nGE,15942
 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=9-JE8hJkOgVJ-R3_9JYjFTUUTlnVq2-U43VP0HiY-sM,2059
+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.cp312-win32.pyd,sha256=BSLr5luwBqUNmWQe5Er-4hQl6hwUA_bDMBOr2mlgsls,342528
+rlgym_learn_algos/rlgym_learn_algos.cp312-win32.pyd,sha256=auas_7bpNMlCT5PTCIAYgepldvq3LvXcpE4LniUUr64,342016
 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
@@ -32,6 +34,6 @@ rlgym_learn_algos/stateful_functions/numpy_obs_standardizer.py,sha256=OgtwCaxBGT
 rlgym_learn_algos/stateful_functions/obs_standardizer.py,sha256=qPPc3--J_3mpJJ-QHJjta6dbWWBobL7SYdK5MUP-XMw,606
 rlgym_learn_algos/util/__init__.py,sha256=VPM6SN4T_625H9t30s9EiLeXiEEWgcyRVHa-LLVNrn4,47
 rlgym_learn_algos/util/running_stats.py,sha256=0tiGFpKtHWzMa1CxM_ueBzd_ryX4bJBriC8MXcSLg8w,4479
-rlgym_learn_algos/util/torch_functions.py,sha256=CTTHzTIi7u1O9HyX0cVJOrnYVbAtnlVs0g1fO9s3ano,3458
-rlgym_learn_algos/util/torch_pydantic.py,sha256=pgj3I-3q8iW9qtOCv1fgjNkZgA00G_Rdkb4qJPk5gxo,3530
-rlgym_learn_algos-0.2.0.dist-info/RECORD,,
+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-0.2.2.dist-info/RECORD,,

{rlgym_learn_algos-0.2.0.dist-info → rlgym_learn_algos-0.2.2.dist-info}/WHEEL

@@ -1,4 +1,4 @@
 Wheel-Version: 1.0
-Generator: maturin (1.8.6)
+Generator: maturin (1.9.0)
 Root-Is-Purelib: false
 Tag: cp312-cp312-win32