Apr 6, 2007

[New Version] v0.31 - Futility pruning and additions to evaluation

  • Made a few arrays static in the SEE, speeding it up quite a bit
  • Futility pruning added
  • X-ray attacks in the evaluation (e.g. queen attacking empty squares behind own bishop)
  • Eval is adjusted towards 0 if a draw is likely
  • Pawns are valued slightly less but becomes a bit more valuable in endings
  • Knight outposts added
  • King tropism added
  • Rook/queen on 7th rank only awarded if enemy king on 8th or enemy pawns still on 7th
  • Fixed bug with bishop positioning and also one with pawn evaluation
Note: One particular implementation of futility pruning (and no additions to evaluation) performed slightly better against v0.3, however I believe this setup works better in general. Only testing will show so I am releasing this now since it is quite an improvement and make any needed adjustments in future releases.

Also I commented out the candidate pawns code. It follows the definition used in Fruit, but it got a bit messy and I did not notice any improvement from it. I will have to keep trying in that area.



Anonymous said...

Hi Jonathan

Thanks for the new Mediocre chess version,Can you tell me where I can find the executable file of an engine,I see only sources and a batch file from the download.


Anonymous said...

I ran several gauntlets with v0.3 against bibichess 0.5, gibbon 2.31g, matheus 2.3, clueless v1.4 and firefly v2.2.0

These are all engines of similar strength as v0.3.

I have now run some gauntlets with v0.31 against the same engines, and my initial results suggests that the new version is a little bit weaker than v0.3.

Jonatan Pettersson said...

The executables become available as soon as Jim Ablett gets time to compile them. I have now added links to the executables for v0.31

As for v0.31 being weaker than v0.3 that is not what my tests show, however I'm aware there is some tuning to do.

I have come to a point where new features doesn't automatically mean stronger play, it all need to fit with the rest of the program. Hopefully I can tune it better for the next version.

Anonymous said...

I have now run more tests, and I'll eat my words :) The new version is stronger.