Search found 1992 matches

by Michel
Sat May 18, 2019 5:35 am
Forum: Computer Chess Club: General Topics
Topic: SF Dev need 150s (2.5 min) to find Qh6
Replies: 13
Views: 642

Re: SF Dev need 150s (2.5 min) to find Qh6

It appears SF misses some zugzwang. Sf-dev with Null Move Reduction disabled: info depth 41 seldepth 58 multipv 1 score cp 9 lowerbound nodes 1800916083 nps 35147369 hashfull 252 tbhits 0 time 51239 pv c1h6 info depth 40 currmove c1h6 currmovenumber 1 info depth 41 seldepth 58 multipv 1 score cp 19...
by Michel
Wed May 15, 2019 3:30 pm
Forum: Computer Chess Club: General Topics
Topic: SF Dev need 150s (2.5 min) to find Qh6
Replies: 13
Views: 642

Re: SF Dev need 150s (2.5 min) to find Qh6

It appears SF misses some zugzwang.
by Michel
Tue Mar 05, 2019 2:49 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Training the trainer: how is it done for Stockfish?
Replies: 22
Views: 2434

Re: Training the trainer: how is it done for Stockfish?

Regarding: "Very simple. MCTS is the only viable approach given the 80000 nps they got using 4 TPUs." Well, A/B fully replaced any other scheme when computers where much slower than 80Knps, and MCTS never won above A/B in any period of chess engine history...so maybe not so simple :-) Regrading: "i...
by Michel
Tue Feb 12, 2019 11:21 am
Forum: Computer Chess Club: General Topics
Topic: Bad eval with 7 men syzygy
Replies: 9
Views: 905

Re: Bad eval with 7 men syzygy

I vaguely recall Tord Romastad's post on fishcooking list where he explained that overvaluing drawn positions is actually good for modern selective search. Getting "good drawn position" might indicate that there is some conversion to superior ending along the way, but it has been pruned one way or ...
by Michel
Sun Jan 27, 2019 1:32 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Linux friendly engines at all levels
Replies: 24
Views: 2457

Re: Linux friendly engines at all levels

GnuCheese

http://hardy.uhasselt.be/GnuCheese/

A cleaned up version of GnuChess 5 (not 6 which is just Fruit 2.1!).

+-2800 on CCRL 40/40.
by Michel
Thu Jan 24, 2019 12:41 am
Forum: Computer Chess Club: General Topics
Topic: Schizophrenic rating model for Leela
Replies: 27
Views: 3916

Re: Schizophrenic rating model for Leela

Stockfish has a score of more than +8 for 24.Bf3 at small depth simply because it prunes for some reason safe moves by the queen. This is because SF allows LMP in PV nodes. This is well-known to cause very visible blunders at depth <=5, but since it was not an elo loss at normal time controls it we...
by Michel
Tue Jan 22, 2019 12:22 pm
Forum: Computer Chess Club: General Topics
Topic: Schizophrenic rating model for Leela
Replies: 27
Views: 3916

Re: Schizophrenic rating model for Leela

Nice! Lots of ideas! So a fully schizophrenic Leela is just two players. The score of a regular engine against a fully schizophrenic Leela with given elo1,elo2 would be the same as its score against two regular players with elo1 and elo2 respectively. A partially schizophrenic Leela is more complica...
by Michel
Sun Jan 13, 2019 10:09 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Writing bugs
Replies: 66
Views: 6830

Re: Writing bugs

Of course it is obvious how to emulate the correct object oriented design in C in this simple case. That was left as an exercise to the reader. You were claiming "state is evil" (as usual a statement devoid of any nuance). I am claiming instead: "there is no problem with state as long as it is prope...
by Michel
Sun Jan 13, 2019 8:57 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Writing bugs
Replies: 66
Views: 6830

Re: Writing bugs

State is fine but it should be encapsulated in an object. Like so: rg=random_generator(seed) r=rg.rand() The state maintained by the rg object could be much more complicated than a single number (this would likely be the case for a cryptographically secure random generator). The rg object completely...
by Michel
Sat Jan 12, 2019 6:25 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Fixed nodes games and the pentanomial model.
Replies: 25
Views: 2565

Re: Fixed nodes games and the pentanomial model.

Yes everything seems to be consistent now. It is interesting to compare the real game measurements with SF's eval (the Gauss curve you posted). It seems that 1cp of SFeval is worth about 1.6 elo. This is quite a bit higher than what I had expected. A small issue is that the ratio between SFeval and ...