Not logged inRybka Chess Community Forum
Up Topic The Rybka Lounge / Computer Chess / Bluefish BMI2 (Beta)
- - By Eelco de Groot (***) Date 2019-09-27 04:45 Edited 2019-09-27 04:51 Upvotes 4
This is just an experiment for Uly mainly, what happens if you throw a bit of Deep Junior ingredients in Stockfish mix, latest development version. Getting the Deep Junior - Kasparov Bishop sac from New York is not so hard, but it only plays it for about 5 minutes. Trouble is White can with little effort reach a repetition draw I think, what also happened in the game. So score sinks again for Black to zero and then lower.

But Uly mainly wants an engine with "style", probably needs more changes than that. In this version I only changed the material value of a pawn from 128 to a round 100. Then all the other piece values roughly the same, and midgame limit and endgame limit to match. Also the material imbalance table but could be argued you want actually more imbalance, not more positional versus material. In that case the material imbalances should be turned not as material down but up. Not tried. Then the same 100/128 increase for King Safety, by dividing kingdanger * kingdanger by 3200 instead of 4096. Another nice round number there, coincidence but 4096 also strange as power of two. That's it, no other changes. I did no extensive checks if there are no overflows anywhere, its a bit of wild code on a rainy night only.
Attachment: BluefishBetaStockfish-clone.7z - BMI2 compile (plus sources), TDM-GCC compiled (450k)
Parent - By Eelco de Groot (***) Date 2019-09-27 05:50 Upvotes 1
Forgot to say that contempt should be set to zero in the UCI settings if you want the exact same engine as in the three games, Analysis contempt can be set to Off for both sides but with zero contempt that should not matter. Dynamic contempt is not so easy to switch off, it is unchanged from Stockfish dynamic contempt. In the games Stockfish had the same UCI settings for contempt as Bluefish,
Parent - By Eelco de Groot (***) Date 2019-09-28 08:28 Edited 2019-09-28 08:48 Upvotes 2
Here's a generic compile for processors that can't run the BMI2 compile, benchnumber is of course the same, 4176743. This should run on all the x86 architecture if it is 64 bit.

Total time (ms) : 3123
Nodes searched  : 4176743
Nodes/second    : 1337413

Also a -modern compile added now.
Attachment: Bluefish.exe - Just the executable, called Bluefish this time, for sources and licence etc. see the other compile. (1151k)
Attachment: Bluefish-modern.exe - -modern compile (in speed, this should fall in between the other two but does not run on older proc) (1143k)
Up Topic The Rybka Lounge / Computer Chess / Bluefish BMI2 (Beta)

Powered by mwForum 2.27.4 © 1999-2012 Markus Wichitill