Fix TensorBoard file iterator (#178)

* Move torch import to local scope and catch tensorboard summary iterator exceptions

* Update CHANGELOG

CHANGELOG.md

@@ -11,6 +11,9 @@ The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.0.0/).
 - Move the KL reduction from the PyTorch `KLAdaptiveLR` class to each agent using it in distributed runs
 - Move the PyTorch distributed initialization from the agent base class to the ML framework configuration
 
+### Fixed
+- Catch TensorBoard summary iterator exceptions in `TensorboardFileIterator` postprocessing utils
+
 ## [1.2.0] - 2024-06-23
 ### Added
 - Define the `environment_info` trainer config to log environment info (PyTorch implementation)
@@ -99,7 +102,7 @@ and Omniverse Isaac Gym environments when they are loaded
 
 ## [0.10.1] - 2023-01-26
 ### Fixed
-- Tensorboard writer instantiation when `write_interval` is zero
+- TensorBoard writer instantiation when `write_interval` is zero
 
 ## [0.10.0] - 2023-01-22
 ### Added
@@ -218,7 +221,7 @@ to allow storing samples in memories during evaluation
 ## [0.4.1] - 2022-03-22
 ### Added
 - Examples of all Isaac Gym environments (preview 3)
-- Tensorboard file iterator for data post-processing
+- TensorBoard file iterator for data post-processing
 
 ### Fixed
 - Init and evaluate agents in ParallelTrainer

skrl/utils/postprocessing.py

@@ -6,7 +6,8 @@
 import os
 
 import numpy as np
-import torch
+
+from skrl import logger
 
 
 class MemoryFileIterator():
@@ -78,6 +79,7 @@ def _format_torch(self) -> Tuple[str, dict]:
         :return: Tuple of file name and data
         :rtype: tuple
         """
+        import torch
         filename = os.path.basename(self.file_paths[self.n])
         data = torch.load(self.file_paths[self.n])
 
@@ -158,18 +160,21 @@ def __next__(self) -> Tuple[str, dict]:
         self.n += 1
 
         data = {}
-        for event in summary_iterator(file_path):
-            try:
-                # get Tensorboard data
-                step = event.step
-                tag = event.summary.value[0].tag
-                value = event.summary.value[0].simple_value
-                # record data
-                if tag in self.tags:
-                    if not tag in data:
-                        data[tag] = []
-                    data[tag].append([step, value])
-            except Exception as e:
-                pass
+        try:
+            for event in summary_iterator(file_path):
+                try:
+                    # get Tensorboard data
+                    step = event.step
+                    tag = event.summary.value[0].tag
+                    value = event.summary.value[0].simple_value
+                    # record data
+                    if tag in self.tags:
+                        if not tag in data:
+                            data[tag] = []
+                        data[tag].append([step, value])
+                except Exception as e:
+                    pass
+        except Exception as e:
+            logger.warning(str(e))
 
         return os.path.dirname(file_path).split(os.sep)[-1], data