Search found 2026 matches

by Michael Sherwin
Wed Jan 30, 2019 2:54 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Lazy move generation and move ordering
Replies: 7
Views: 681

Re: Lazy move generation and move ordering

In RomiChess which is a bitboard engine I do a hybrid staged move generation. First a pseudo move generator creates all the move and attack bitboards. At anytime if the opposing king is captured it exits immediately. Then in the staged part it starts with the hashtable move checks to see if that mov...
by Michael Sherwin
Mon Jan 21, 2019 8:50 am
Forum: Computer Chess Club: General Topics
Topic: Training with RomiChess in console mode
Replies: 23
Views: 3169

Re: Training with RomiChess in console mode

Romichess K (Per) came with a personality capability. I was able to get Romichess K to work as an analysis engine but not Romichess K (Per). Is it not feasible to make an analysis engine that can make use of this personality capability. BTW I'm using SCID vs PC. Obviously the engine is extremely st...
by Michael Sherwin
Thu Jan 17, 2019 11:43 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Reinforcement Learning (RL) in real time paradigm
Replies: 12
Views: 1057

Re: Reinforcement Learning (RL) in real time paradigm

Michael, I will be glad if you prove me wrong. But even then I would search for ways to reduce game length. You want to feed transposition table with additional information, trusting that it will help to shape the final search. The idea looks good, as long as this information has a chance to be acc...
by Michael Sherwin
Tue Jan 15, 2019 7:50 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Reinforcement Learning (RL) in real time paradigm
Replies: 12
Views: 1057

Re: Reinforcement Learning (RL) in real time paradigm

A similar idea has been proposed before I think: https://papers.nips.cc/paper/3722-bootstrapping-from-game-tree-search.pdf It must be over 10 years I have seen Joels name mentioned somewhere, thanks for the paper. https://www.chessprogramming.org/Bodo I didn't even know he created 'Meep', because t...
by Michael Sherwin
Tue Jan 15, 2019 7:49 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Reinforcement Learning (RL) in real time paradigm
Replies: 12
Views: 1057

Re: Reinforcement Learning (RL) in real time paradigm

HOW IT CAN BE DONE USING ONLY ONE THREAD From any given position using more shallow alpha-beta searches play a game of chess (or fragment of a game) until a decisive (or pseudo decisive) result. Add the moves into a tree structure in ram updating the the RL values. Play as many games as possible in...
by Michael Sherwin
Tue Jan 15, 2019 3:14 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Reinforcement Learning (RL) in real time paradigm
Replies: 12
Views: 1057

Re: Reinforcement Learning (RL) in real time paradigm

I'm afraid this algorithm would search too many moves irrelevent for the second search to be competitive. But when I'm analysing positions under Arena, I often do the following: short search, taking back, longer search. Some tactics are found faster that way. I'd suggest the following: try that app...
by Michael Sherwin
Tue Jan 15, 2019 9:36 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Reinforcement Learning (RL) in real time paradigm
Replies: 12
Views: 1057

Re: Reinforcement Learning (RL) in real time paradigm

I am not sure to understand your proposal. To me Reinforcement learning (RL) is what is often called here Texel Tuning, ie a way to tune the eval parameters from played game. Your proposal looks like building an opening book (from the current position) and loading it to the hash table to improve th...
by Michael Sherwin
Tue Jan 15, 2019 9:04 am
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Reinforcement Learning (RL) in real time paradigm
Replies: 12
Views: 1057

Re: Reinforcement Learning (RL) in real time paradigm

Is there nobody that would like to talk about this idea? I think it would be possible to play thousands of training games before each main search. The main search's eval function would not need to be used (but could be used). This hinges on another idea that I have posted about. That would be a sear...
by Michael Sherwin
Mon Jan 14, 2019 7:04 pm
Forum: Computer Chess Club: Programming and Technical Discussions
Topic: Reinforcement Learning (RL) in real time paradigm
Replies: 12
Views: 1057

Reinforcement Learning (RL) in real time paradigm

HOW IT CAN BE DONE USING ONLY ONE THREAD From any given position using more shallow alpha-beta searches play a game of chess (or fragment of a game) until a decisive (or pseudo decisive) result. Add the moves into a tree structure in ram updating the the RL values. Play as many games as possible in ...