Engine results: a surprise! (again)

[jdart]

But even with this, is randomness not needed?

You shouldn't use a small-sized PGN file to run a match with a large number of games, because you will get repeated openings. You need a big file, either PGN or polyglot (which cutechess can read).

--Jon

[xmas79]

But even with this, is randomness not needed?

You shouldn't use a small-sized PGN file to run a match with a large number of games, because you will get repeated openings. You need a big file, either PGN or polyglot (which cutechess can read).

--Jon

as i'm out all day, i'm running now 50k games with 2M games pgn file and 12 plies Max as openings... should be that random enough?

[jdart]

That sounds fine. You can use order=random to get random selections from the book. Note though that cutechess is actually deterministic. It uses a pseudo-random move generator and by default will pick the same openings in the same order each time. If using multiple cores in parallel you can use the -srand option to make each core use a different sequence through the book.

--Jon

[xmas79]

Ok test completed. Something around 27k games with around one hundred duplicated games, I had a cutechess-cli crash (QTCore4.dll moule to be precise...):

ELOStat says (base 2400):

White Wins : 12544 (46.7 %)
Black Wins : 11474 (42.7 %)
Draws : 2833 (10.6 %)
Unfinished : 2

White Perf. : 52.0 %
Black Perf. : 48.0 %

ECO A = 8618 Games (32.1 %)
ECO B = 8246 Games (30.7 %)
ECO C = 4528 Games (16.9 %)
ECO D = 2877 Games (10.7 %)
ECO E = 2582 Games ( 9.6 %)

Program Elo + - Games Score Av.Op. Draws

1 MyEngine : 2499 4 4 26851 75.7 % 2301 10.6 %
2 FairyMax : 2301 4 4 26851 24.3 % 2499 10.6 %


So difference is 198 ELO points. Definitely a good start!

Thanks,
Natale.

[jd1]

Actually my name is Natale

Sorry , my apologies! Your latest results look good. Keep up the good work!