v0.23.0 - Centerfold Release

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog and this project adheres to Semantic Versioning.

---

[0.23.0] - 2026-03-28 @ 01:55 - Centerfold Release

Added

• Additional Configurable Simulation Settings
  • Epsilon Nice Settings
    • Added TUI controls for setting the NICE_P_VALUE and NICE_STEPS values.
  • ATH Replay Memory Settings
    • Added TUI controls for setting the DOWNSHIFT_COUNT_THRESHOLD, MAX_FRAMES, MAX_GEAR, MAX_HARD_RESET_EPISODES, MAX_STAGNANT_EPISODES, MAX_TRAINING_FRAMES, NUM_COOLDOWN_EPISODES, and UPSHIFT_COUNT_THRESHOLD values.
  • Reward Settings
    • Added TUI constrols for setting the FOOD_REWARD, COLLISION_PENALTY, MAX_MOVES_PENALTY, EMPTY_MOVE_REWARD, CLOSER_TO_FOOD, FURTHER_FROM_FOOD, and MAX_MOVES_MULTIPLIER values.
  • Wired these settings through the entire stack i.e. enabled support in the SimCfg, HydraMgr, ATHMemory, ATHCommon, ATHDataMgr, ATHDataStore, ATHGearbox, and EpsilonNicePolicy classes.
• Updated PyPI and RTD documentation.
  • Added detailed configuration information with screenshots.

Changed

• Removed THRESHOLD_BUCKETS. Instead, the system uses the last three buckets and limits the MAX_BUCKETS to have a minimum of 3.
• Reworked how the Linear Model runs; made it consistent with the GRU and RNN:
  • Training now happens at the end of the episode.
  • The Linear model now uses the ATH Replay Memory
• Enabled the layers TUI option for the Linear model; it denotes the number of hidden layers that the NN contains.
• Changed the Snapshot Report to use the TUI labels instead of the backend variable names.

Fixed

• Set MAX_MOVES_PENALTY to -10 (not to 0)