[Feature] keep_checkpoints_num and checkpoint_at_end (facebookres…

…earch#102)

* amend

* amend

* amend

benchmarl/conf/experiment/base_experiment.yaml

@@ -98,3 +98,8 @@ restore_file: null
 # Interval for experiment saving in terms of collected frames (this should be a multiple of on/off_policy_collected_frames_per_batch).
 # Set it to 0 to disable checkpointing
 checkpoint_interval: 0
+# Wether to checkpoint when the experiment is done
+checkpoint_at_end: False
+# How many checkpoints to keep. As new checkpoints are taken, temporally older checkpoints are deleted to keep this number of
+# checkpoints. The checkpoint at the end is included in this number. Set to `null` to keep all checkpoints.
+keep_checkpoints_num: 3

benchmarl/experiment/experiment.py

@@ -11,7 +11,7 @@
 
 import os
 import time
-from collections import OrderedDict
+from collections import deque, OrderedDict
 from dataclasses import dataclass, MISSING
 from pathlib import Path
 from typing import Dict, List, Optional
@@ -96,7 +96,9 @@ class ExperimentConfig:
 
     save_folder: Optional[str] = MISSING
     restore_file: Optional[str] = MISSING
-    checkpoint_interval: float = MISSING
+    checkpoint_interval: int = MISSING
+    checkpoint_at_end: bool = MISSING
+    keep_checkpoints_num: Optional[int] = MISSING
 
     def train_batch_size(self, on_policy: bool) -> int:
         """
@@ -280,6 +282,8 @@ def validate(self, on_policy: bool):
                 f"checkpoint_interval ({self.checkpoint_interval}) "
                 f"is not a multiple of the collected_frames_per_batch ({self.collected_frames_per_batch(on_policy)})"
             )
+        if self.keep_checkpoints_num is not None and self.keep_checkpoints_num <= 0:
+            raise ValueError("keep_checkpoints_num must be greater than zero or null")
         if self.max_n_frames is None and self.max_n_iters is None:
             raise ValueError("n_iters and total_frames are both not set")
 
@@ -483,6 +487,7 @@ def _setup_name(self):
         self.model_name = self.model_config.associated_class().__name__.lower()
         self.environment_name = self.task.env_name().lower()
         self.task_name = self.task.name.lower()
+        self._checkpointed_files = deque([])
 
         if self.config.restore_file is not None and self.config.save_folder is not None:
             raise ValueError(
@@ -668,6 +673,8 @@ def _collection_loop(self):
             pbar.update()
             sampling_start = time.time()
 
+        if self.config.checkpoint_at_end:
+            self._save_experiment()
         self.close()
 
     def close(self):
@@ -835,10 +842,16 @@ def load_state_dict(self, state_dict: Dict) -> None:
 
     def _save_experiment(self) -> None:
         """Checkpoint trainer"""
+        if self.config.keep_checkpoints_num is not None:
+            while len(self._checkpointed_files) >= self.config.keep_checkpoints_num:
+                file_to_delete = self._checkpointed_files.popleft()
+                file_to_delete.unlink(missing_ok=False)
+
         checkpoint_folder = self.folder_name / "checkpoints"
         checkpoint_folder.mkdir(parents=False, exist_ok=True)
         checkpoint_file = checkpoint_folder / f"checkpoint_{self.total_frames}.pt"
         torch.save(self.state_dict(), checkpoint_file)
+        self._checkpointed_files.append(checkpoint_file)
 
     def _load_experiment(self) -> Experiment:
         """Load trainer from checkpoint"""