Chess Engine: Next Steps

[oriyonay]

Hi you beautiful people! I hope you're having a phenomenal day
I've been working on my engine for 3 months now, and have negamax, qsearch, ID, aspiration windows, check extension, MVV/LVA, null-move pruning, late-move reductions, TTs, killer/history heuristics, PVS, tapered evaluation (PeSTO) + (bishop pair, doubled/isolated/passed pawn, semi/fully open rook/queen files, basic piece mobility and king safety), and beta pruning.
My engine currently plays on the level of other 2200-2300 engines.

What are the next steps to improve my engine? My goal is 2900 by 2022!
Any input is highly appreciated as always!!

Have a fantastic rest of your day!!

- Ori

[supersharp77]

Hi you beautiful people! I hope you're having a phenomenal day
I've been working on my engine for 3 months now, and have negamax, qsearch, ID, aspiration windows, check extension, MVV/LVA, null-move pruning, late-move reductions, TTs, killer/history heuristics, PVS, tapered evaluation (PeSTO) + (bishop pair, doubled/isolated/passed pawn, semi/fully open rook/queen files, basic piece mobility and king safety), and beta pruning.
My engine currently plays on the level of other 2200-2300 engines.

What are the next steps to improve my engine? My goal is 2900 by 2022!
Any input is highly appreciated as always!!

Have a fantastic rest of your day!!

- Ori

Github Shows "Ori Yonay Starfish Chess Engine Java"...........


https://github.com/oriyonay/Starfish

https://github.com/oriyonay/Starfish/ra ... arfish.jar

[oriyonay]

Github Shows "Ori Yonay Starfish Chess Engine Java"...........

Haha that was my first engine.. My new one is bitboard-based, can do en passant, and is 500x faster - that’s quite an improvement!

[supersharp77]

Github Shows "Ori Yonay Starfish Chess Engine Java"...........

Haha that was my first engine.. My new one is bitboard-based, can do en passant, and is 500x faster - that’s quite an improvement!

Well does this "New" Engine have a name? Are their engine links?......... Thx AR

[j.t.]

- futility pruning / futility reductions+SEE
- delta pruning(+SEE)
- SEE for move ordering
- tuning all evaluation parameter together with PSQT values (e.g. by using gradient descent)
- test if really all features add performance (for example for me internal iterative deeping and aspiration windows didn't work, also any use of TT in quiesce (neither looking up nor writing) worked).
- You can tweak lots of values:

• - test if 3 killers work better than two
  - test if the maximum value for the history heuristic table would be better if higher/lower
  - test slightly different move ordering schemes (e.g. killers before losing captures of killer after losing captures)
  - test if it helps to allow check-avoiding quiet moves in quiesce (or not)
  - increase/descrease LMR and nullmove reductions

- find all the bugs (IMPORTANT)

Good luck!