New engine: Clover

[lucametehau]

Hello again!

Found a major tt bug that should affect tournament play, thanks to Connor, Seer author.

Link: https://github.com/lucametehau/CloverEn ... tag/v2.3.1

[connor_mcmonigle]

Thanks, though I'm unsure this resolves the issues I was experiencing... Were you able to reproduce the buggy moves by filling the hash with entries from earlier searches? And did this change fix it? Making a new release when your testing indicates maybe -2 elo doesn't make a lot of sense imo.

I just pointed out a position where Clover played a buggy move in a match at LTC on 1th. There were several other positions where clear blunders were played that were irreproducible from the given position in my testing suggesting some issue with the transposition table entries being corrupted or not entered with correct bounds or possibly zobrist collisions. Perhaps the issue was specific to my Linux compile. I'm unsure.

[lucametehau]

What was wrong, was that I was not correctly reseting age when starting a new game, so this affected the tt entries by a lot.

I played the exact same game as you showed me and this time Clover didn't blunder so I hope this bug is gone

[mar]

Thanks, though I'm unsure this resolves the issues I was experiencing...

funny thing is that aging is not even called, because it's commented out so this fix solves absolutely nothing

[lucametehau]

Thanks, though I'm unsure this resolves the issues I was experiencing...

funny thing is that aging is not even called, because it's commented out so this fix solves absolutely nothing

Ah I age the tt when the go command is given (check uci.h). I think you looked at search.h and saw on line 720 TT->age() commented

[mar]

Ah I age the tt when the go command is given (check uci.h). I think you looked at search.h and saw on line 720 TT->age() commented

ok right, I missed that. either way I very much doubt aging would be the problem here (blundering), so I'd probably look elsewhere.
I simply clear hashtable on ucinewgame btw

[lucametehau]

Ah I age the tt when the go command is given (check uci.h). I think you looked at search.h and saw on line 720 TT->age() commented

ok right, I missed that. either way I very much doubt aging would be the problem here (blundering), so I'd probably look elsewhere.
I simply clear hashtable on ucinewgame btw

Well about the blundering, I think you might be right, there might be another bug. Still, I don't see why not reseting the age before a new game wouldn't be considered a bug. Reseting it makes the engine more stable (maybe I can get some results from ctt now ).

[Sylwy]

After only 16 games in my test....

[Sylwy]

Here is one of my tests:

http://talkchess.com/forum3/viewtopic.php?f=6&t=77364

[lucametehau]

New version of Clover is here: https://github.com/lucametehau/CloverEn ... s/tag/v2.4. It is 100+ elo stronger than the previous version and I hope that it's at least 50 elo stronger against other engines.
Tell me if there are any problems!