Skip to content

v1.0.0

Choose a tag to compare

View all tags
@TomaszJaworski777 TomaszJaworski777 released this 04 Dec 13:45
· 75 commits to main since this release
1.0.0
b6f6b5a
This commit was created on GitHub.com and signed with GitHub’s verified signature.
GPG key ID: B5690EEEBB952194
Verified
Learn about vigilant mode.

This engine was developed to harvest some aggressiveness from the MCTS search method. Current estimated Elo is around 3000.

Implemented Features:

  • Tree reuse
  • LRU
  • FPU
  • Gini scaling
  • Scaling C with search duration
  • Variance scaling
  • Exploration scaling
  • Threats in value net
  • Defenses in value net
  • Horizontal mirroring in value net
  • Softmax policy temperature at root
  • Extended time manager
  • Multithreading
  • Hashtable

Jackal currently supports multiple threads, but due to some unidentified bug it doesn't play better with more than 1 thread. ($100 for fixing PR)

Features to Increase Its EAS:

  • Policy bonus to sacrifices
  • Filtering positions to only those where sides with lower material won the game
  • Asymmetrical contempt in PUCT
  • Bonus to positions that have less material than root position

I measured Jackal's EAS to be around 230k, while I also noticed it is very slow to end the games and draws a lot of winning positions. My current guess is that MCTS heavily relies on its neural nets and my current data is just not strong enough to be efficient in end games.

EAS Score Sacs Shorts Bad draws Avg. win moves Engine/player
229966 40.00% 26.71% 17.36% 103 Jackal
Assets 5
Loading