Skip to content

Conversation

@omiDBaghi
Copy link

Fix: Prevent duplicate score counting by tracking consumed episodes

Adds consumed-episodes directory to track episodes that have already been used for training. Filters out consumed episodes in discover_episode_metadata to prevent the same rewards from being counted multiple times.

Fixes #26

Context

This fix addresses issue #26 where scores were being counted multiple times because episodes were reused across training runs.

Summary of Change

  • Added import shutil for file operations
  • Added get_consumed_episodes_dir() to determine consumed episodes directory path
  • Added mark_episodes_as_consumed() to move used episodes to consumed-episodes directory
  • Modified discover_episode_metadata() to filter out already-consumed episodes
  • Episodes are now tracked in persistent-data/state-service/consumed-episodes after training

Tests

Tested manually by:

  1. Running CodeAssist and earning score
  2. Exiting with Ctrl+C
  3. Restarting and exiting again with Ctrl+C
  4. Verifying that consumed episodes were moved to consumed-episodes directory
  5. Confirming that score is not re-counted on subsequent runs

Discord Username: omid7923

Fix: Prevent duplicate score counting by tracking consumed episodes

Adds consumed-episodes directory to track episodes that have already
been used for training. Filters out consumed episodes in discover_episode_metadata
to prevent the same rewards from being counted multiple times.

Fixes gensyn-ai#26
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Score / reward can be counted multiple times by restarting and pressing Ctrl+C

1 participant