This update involved a lot of code refactoring but did include some improvements to net about 100 Elo in self-play. Some of the changes include:
- Pawn push threats in evaluation
- Passed pawn evaluation tweaks
- 2-ply continuation history
- improving heuristic
I've also included a Linux binary for the first time. I have done minimal testing with the Linux binary, basically just to see if it actually runs. If you encounter any issues with that binary, please let me know.
Self-play LTC (60+0.6):
Score of Puffin 5.0 vs Puffin 4.0: 117 - 29 - 152 [0.648] 298
... Puffin 5.0 playing White: 53 - 13 - 83 [0.634] 149
... Puffin 5.0 playing Black: 64 - 16 - 69 [0.661] 149
... White vs Black: 69 - 77 - 152 [0.487] 298
Elo difference: 105.7 +/- 27.5, LOS: 100.0 %, DrawRatio: 51.0 %
SPRT: llr 2.95 (100.1%), lbound -2.94, ubound 2.94 - H1 was accepted