Search found 161 matches

by nionita
Wed Sep 11, 2019 7:30 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: New Tool for Tuning with Skopt
Replies: 60
Views: 4873

Re: New Tool for Tuning with Skopt

After trying this script a little on some search parameter I am not in success. Any advice will be welcome about choosing the good script configuration and a reasonable number of games. For now, 1000 games to tune 4 or 5 parameters at a time seems not good. People have had some problems with skopt ...
by nionita
Wed Jul 03, 2019 8:37 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Formula needed for optimizing over-subscription of threads for the sake of batching
Replies: 2
Views: 991

Re: Formula needed for optimizing over-subscription of threads for the sake of batching

What if you make a small NN to estimate performace with the data points you have and 3 entries (delay, threads, cpu cores)? Even if it overfits, it memorizes your results, which is actually what you want.
by nionita
Sat Jun 29, 2019 9:49 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: On-line engine blitz tourney June
Replies: 17
Views: 2811

Re: On-line engine blitz tourney June

Roland Chastain wrote:
Sat Jun 29, 2019 9:25 pm
I tried to participate but apparently it didn't work.
Do you have an account from HGM?
by nionita
Wed Jun 19, 2019 6:14 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Turnover Chess
Replies: 2
Views: 676

Re: Turnover Chess

This kind of games would be ideal for a "zero" approach, as there is no human experience or intuition how the game strategy should look like. Then, especially if the game is interesting (i.e. the complexity not too low and not too high), it would be also interesting to see how quick a "dedicated" ap...
by nionita
Fri Mar 01, 2019 8:37 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Training the trainer: how is it done for Stockfish?
Replies: 22
Views: 4495

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

Stockfish is not trained. When Stockfish developers have a new idea (e.g. a search or evaluation improvement), they implement and test it by playing many games against current best version. If the new Stockfish is better, then it becomes the current best. This is the way classic AB engines are impro...
by nionita
Sat Feb 23, 2019 7:05 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: MCTS implementation question
Replies: 2
Views: 948

Re: MCTS implementation question

The search aim is to approximate the node values. But once you have terminal nodes, there is no need to approximate anymore, their value is exact. So those nodes should not be visited again. Instead their parents should be aware of such descendants, such that: - if one descendant is a win, the paren...
by nionita
Sat Feb 16, 2019 12:02 am
Forum: Computer Chess Club: General Topics
Topic: Barbarossa 0.5.0 released
Replies: 2
Views: 678

Barbarossa 0.5.0 released

Hi all, I just released version 0.5.0 of Barbarossa, which has again many improvements in eval & search, as well as some technical optimizations. In self play it's about 140 elo points better than version 0.4.0. Barbarossa is the best open source Haskell chess engine. I created binaries for Win7 (32...
by nionita
Thu Jan 31, 2019 12:22 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: JavaScript question
Replies: 4
Views: 932

Re: JavaScript question

Yes, for arrays this is true, you just copy a reference. For primitive type you would copy the value itself.
by nionita
Tue Jun 12, 2018 7:25 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: whether or not a piece has moved and how many times
Replies: 11
Views: 2517

Re: whether or not a piece has moved and how many times

Curious about how others have implemented calculation of how many times each piece has moved since the beginning of the game. The concern is when a piece has been captured in a make move and then uncaptured and updating the array. How did you solve this? I'd do it the same way I handle the 50-move ...
by nionita
Wed Dec 13, 2017 11:16 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Google's AlphaGo team has been working on chess
Replies: 203
Views: 62511

Re: Google's AlphaGo team has been working on chess

Note that the AlphaZero's NN is trained to predict the visiting frequencies the nodes will get in an MCTS. That doesn't necessarily correspond to the evaluation score of the moves. Moves with complex tactics will need a lot of visits before they are resolved one way or the other, because the MCTS w...