Merge branch 'dev' into feature/1386-enable-a-repeatable-or-deterministic-baseline-test

src/primaite/agents/agent.py

@@ -257,14 +257,19 @@ class AgentSessionABC(ABC):
             raise FileNotFoundError(msg)
         pass
 
+    @property
+    def _saved_agent_path(self) -> Path:
+        file_name = (
+            f"{self._training_config.agent_framework}_"
+            f"{self._training_config.agent_identifier}_"
+            f"{self.timestamp_str}.zip"
+        )
+        return self.learning_path / file_name
+
+    @abstractmethod
     def save(self):
         """Save the agent."""
-        agent_path = (
-            self.session_path
-            / f"{self._training_config.agent_framework}_{self._training_config.agent_identifier}_{self.timestamp_str}"
-        )
-        _LOGGER.debug(f"Saving agent: {agent_path}")
-        self._agent.save(agent_path)
+        pass
 
     @abstractmethod
     def export(self):

src/primaite/agents/rllib.py

@@ -1,9 +1,11 @@
 from __future__ import annotations
 
 import json
+import shutil
 from datetime import datetime
 from pathlib import Path
 from typing import Union
+from uuid import uuid4
 
 from ray.rllib.algorithms import Algorithm
 from ray.rllib.algorithms.a2c import A2CConfig
@@ -121,9 +123,11 @@ class RLlibAgent(AgentSessionABC):
     def _save_checkpoint(self):
         checkpoint_n = self._training_config.checkpoint_every_n_episodes
         episode_count = self._current_result["episodes_total"]
-        if checkpoint_n > 0 and episode_count > 0:
-            if (episode_count % checkpoint_n == 0) or (episode_count == self._training_config.num_episodes):
-                self._agent.save(str(self.checkpoints_path))
+        save_checkpoint = False
+        if checkpoint_n:
+            save_checkpoint = episode_count % checkpoint_n == 0
+        if episode_count and save_checkpoint:
+            self._agent.save(str(self.checkpoints_path))
 
     def learn(
         self,
@@ -141,6 +145,7 @@ class RLlibAgent(AgentSessionABC):
         for i in range(episodes):
             self._current_result = self._agent.train()
             self._save_checkpoint()
+        self.save()
         self._agent.stop()
 
         super().learn()
@@ -167,6 +172,26 @@ class RLlibAgent(AgentSessionABC):
         """Load an agent from file."""
         raise NotImplementedError
 
+    def save(self, overwrite_existing: bool = True):
+        """Save the agent."""
+        # Make temp dir to save in isolation
+        temp_dir = self.learning_path / str(uuid4())
+        temp_dir.mkdir()
+
+        # Save the agent to the temp dir
+        self._agent.save(str(temp_dir))
+
+        # Capture the saved Rllib checkpoint inside the temp directory
+        for file in temp_dir.iterdir():
+            checkpoint_dir = file
+            break
+
+        # Zip the folder
+        shutil.make_archive(str(self._saved_agent_path).replace(".zip", ""), "zip", checkpoint_dir)  # noqa
+
+        # Drop the temp directory
+        shutil.rmtree(temp_dir)
+
     def export(self):
         """Export the agent to transportable file format."""
         raise NotImplementedError

src/primaite/agents/sb3.py

@@ -65,11 +65,13 @@ class SB3Agent(AgentSessionABC):
     def _save_checkpoint(self):
         checkpoint_n = self._training_config.checkpoint_every_n_episodes
         episode_count = self._env.episode_count
-        if checkpoint_n > 0 and episode_count > 0:
-            if (episode_count % checkpoint_n == 0) or (episode_count == self._training_config.num_episodes):
-                checkpoint_path = self.checkpoints_path / f"sb3ppo_{episode_count}.zip"
-                self._agent.save(checkpoint_path)
-                _LOGGER.debug(f"Saved agent checkpoint: {checkpoint_path}")
+        save_checkpoint = False
+        if checkpoint_n:
+            save_checkpoint = episode_count % checkpoint_n == 0
+        if episode_count and save_checkpoint:
+            checkpoint_path = self.checkpoints_path / f"sb3ppo_{episode_count}.zip"
+            self._agent.save(checkpoint_path)
+            _LOGGER.debug(f"Saved agent checkpoint: {checkpoint_path}")
 
     def _get_latest_checkpoint(self):
         pass
@@ -91,6 +93,7 @@ class SB3Agent(AgentSessionABC):
             self._agent.learn(total_timesteps=time_steps)
             self._save_checkpoint()
         self._env.reset()
+        self.save()
         self._env.close()
         super().learn()
 
@@ -134,6 +137,10 @@ class SB3Agent(AgentSessionABC):
         """Load an agent from file."""
         raise NotImplementedError
 
+    def save(self):
+        """Save the agent."""
+        self._agent.save(self._saved_agent_path)
+
     def export(self):
         """Export the agent to transportable file format."""
         raise NotImplementedError