PK :
Thanks. Yeah I think figuring out how to increase search depth is the most obvious path to increasing strength. The Slow Chess search depth for most tactics is probably about 1-ply deeper than other programs (when a threat move is played at last ply, it stays in main search for one more ply instead of going to q-search), but still overall way less.
Endgame play always seems hit or miss despite the improvements I've made since WV2.1.
Has a bigger random assortment of knowledge like knows this looks like draw : 8/5k2/8/5K1p/1P1b3P/3B3P/8/8 w - -
but play in some common endgames can be lacking, like there is no eval or bitbase for KRP-KR win/draw, which would help with rook endgame search. And occasionally manages to do something like this (as white) : 7k/6R1/6PP/5r2/8/4K3/8/8 b - -
Max :
Unfortunately I think a Mac version is not likely at this point, but maybe eventually. The code base not very portable and contains many windows function calls (mainly for the GUI code which is still from 2004 using win32 api.)
As a vague future project I was considering more cleaning up code and making engine-side fully modular for expanding to other games (like Othello could be fun quick test for that.) If I do that I will look into other platforms. I do have a Macbook but haven't tried testing on it.
Slow Chess Blitz Classic
Moderators: hgm, Rebel, chrisw
-
- Posts: 11
- Joined: Sun Aug 04, 2019 6:44 am
- Full name: Jonathan Kreuzer
-
- Posts: 640
- Joined: Sat Jun 08, 2013 10:07 am
- Location: France
- Full name: Roland Chastain
Re: Slow Chess Blitz Classic
Thank you Jonathan for the update. Both problems solved.
Here are the results of some engines tournaments I made yesterday with CuteChess Cli. The time control is 40/60. I use Bayeselo for the ratings, with a zero value for Iota 0.3.
By the way, I saw one or two games where Slow Chess Classic 1.4 lost on time. And the same for Rodent IV beta (I say that for Pawel).
You already answered to the question of other platforms, so I won't repeat it. But if ever you need, I would be glad to help for compiling and testing under Linux.
There is a Slow Chess 2.960, by Alexander Schmidt, derived from 2.96 and playing Fischer random chess. I don't know if you are interested in that variant. It would be my suggestion for a future version.
Here are the results of some engines tournaments I made yesterday with CuteChess Cli. The time control is 40/60. I use Bayeselo for the ratings, with a zero value for Iota 0.3.
Code: Select all
Rank Name Elo + - games score oppo. draws
1 Senpai 2.0 1087 111 92 80 94% 530 8%
2 Rodent IV beta 897 110 106 40 73% 608 20%
3 SlowChess Classic 1.4 876 83 79 80 74% 572 10%
4 SlowChess Blitz WV 641 145 146 20 53% 596 15%
5 SlowChess Blitz WV2 632 155 173 20 43% 677 5%
6 SlowChess 2.96 604 159 161 20 53% 568 5%
7 SlowChess 2.94 594 156 159 20 50% 570 10%
8 SlowChess Blitz 0 558 144 154 20 43% 612 15%
9 SlowChess Blitz WV21 524 117 133 40 31% 683 8%
10 Viper 0.1 476 171 201 20 30% 677 0%
11 Raven 0.50 457 156 174 20 33% 632 5%
12 Tony's chess 0.03 284 172 203 20 20% 632 0%
13 Iota 0.3 0 138 220 80 0% 747 0%
You already answered to the question of other platforms, so I won't repeat it. But if ever you need, I would be glad to help for compiling and testing under Linux.
There is a Slow Chess 2.960, by Alexander Schmidt, derived from 2.96 and playing Fischer random chess. I don't know if you are interested in that variant. It would be my suggestion for a future version.
Qui trop embrasse mal étreint.
-
- Posts: 11
- Joined: Sun Aug 04, 2019 6:44 am
- Full name: Jonathan Kreuzer
Re: Slow Chess Blitz Classic
Roland :
Thanks for the testing. Looks about 300 elo stronger than old versions if they are lumped together and averaged.
Do you know if the time control losses were just slightly too late on last move, or engine was hanging? There is some chance it's already fixed in latest (a day or 2 ago I put up 1.4c, which fixes for me a bug with UCI ponder-on in cutechess where super fast moves like the final checkmate could hang, but also was refactoring UCI handling in general including time control. ) In testing I didn't notice it go over time limit, actually not sure if it auto-flags though.
1.4c also has larger hash table sizes selectable from the menu in the GUI for own GUI/winboard. (and larger limit for uci hash option.)
Fischer Random looks like a possibility for next version, from reading wikipedia looks like mainly just needs a separate handling of castling.
I will probably take a break from chess for a bit (I think much less than 14 years this time! And hopefully come back with a clearly better version 2.)
Thanks for the testing. Looks about 300 elo stronger than old versions if they are lumped together and averaged.
Do you know if the time control losses were just slightly too late on last move, or engine was hanging? There is some chance it's already fixed in latest (a day or 2 ago I put up 1.4c, which fixes for me a bug with UCI ponder-on in cutechess where super fast moves like the final checkmate could hang, but also was refactoring UCI handling in general including time control. ) In testing I didn't notice it go over time limit, actually not sure if it auto-flags though.
1.4c also has larger hash table sizes selectable from the menu in the GUI for own GUI/winboard. (and larger limit for uci hash option.)
Fischer Random looks like a possibility for next version, from reading wikipedia looks like mainly just needs a separate handling of castling.
I will probably take a break from chess for a bit (I think much less than 14 years this time! And hopefully come back with a clearly better version 2.)
-
- Posts: 640
- Joined: Sat Jun 08, 2013 10:07 am
- Location: France
- Full name: Roland Chastain
Re: Slow Chess Blitz Classic
Thank you Jonathan. I downloaded your latest build. If I see again games lost on time I will report it here. Glad that you think of adding FRC support.
Qui trop embrasse mal étreint.
-
- Posts: 11
- Joined: Sun Aug 04, 2019 6:44 am
- Full name: Jonathan Kreuzer
Re: Slow Chess Blitz Classic
Well, I made one more update for now, Blitz Classic 1.5 is now up on the web page http://www.3dkingdoms.com/chess/slow.htm
Changes :
- Support for Chess960 / Fischer Random Chess under UCI and in own GUI.
- Search one step more selective ( multi-ply lmr, probcut)
- some small bug fixes and improvements
Hopefully no new bugs... despite it being "just" changing castling handling first tests were pretty broken, but now everything appears to be fixed, 100 FRC and normal games with no issues noticed (under cutechess gui)
For the playing strength it appears likely to be stronger than 1.4 ( I tried running 400 bullet games over-night every night against version 1.4, and always averaged to about 40 elo stronger after 400 games, although at just 100 games results varied from +3 to + 80.)
Despite initial promising results (+40 elo self-play test after one hour just adjusting search more selective), Slow Chess appears still resistant to modern search techniques. Any other changes I made ( way more selective, different logic, etc.) any overnight test with enough games was almost exactly +40 against 1.4. So I think I prefer my usual strategy of not trying to measure playing strength
Changes :
- Support for Chess960 / Fischer Random Chess under UCI and in own GUI.
- Search one step more selective ( multi-ply lmr, probcut)
- some small bug fixes and improvements
Hopefully no new bugs... despite it being "just" changing castling handling first tests were pretty broken, but now everything appears to be fixed, 100 FRC and normal games with no issues noticed (under cutechess gui)
For the playing strength it appears likely to be stronger than 1.4 ( I tried running 400 bullet games over-night every night against version 1.4, and always averaged to about 40 elo stronger after 400 games, although at just 100 games results varied from +3 to + 80.)
Despite initial promising results (+40 elo self-play test after one hour just adjusting search more selective), Slow Chess appears still resistant to modern search techniques. Any other changes I made ( way more selective, different logic, etc.) any overnight test with enough games was almost exactly +40 against 1.4. So I think I prefer my usual strategy of not trying to measure playing strength
-
- Posts: 3553
- Joined: Thu Jun 07, 2012 11:02 pm
Re: Slow Chess Blitz Classic
I will have to give this a try ! There are over 800 games for 1.4 in today's standard chess blitz update, so the rating there will inform me on opponents to choose for chess960 matches.jonk wrote: ↑Tue Sep 17, 2019 7:34 pm Well, I made one more update for now, Blitz Classic 1.5 is now up on the web page http://www.3dkingdoms.com/chess/slow.htm
Changes :
- Support for Chess960 / Fischer Random Chess under UCI and in own GUI.
In Cutechess GUI, did you run it as winboard or UCI for those 100 chess960 games?
-
- Posts: 3553
- Joined: Thu Jun 07, 2012 11:02 pm
-
- Posts: 640
- Joined: Sat Jun 08, 2013 10:07 am
- Location: France
- Full name: Roland Chastain
Re: Slow Chess Blitz Classic
Great! Thank you Jonathan. I can't wait to try it.
Qui trop embrasse mal étreint.
-
- Posts: 11
- Joined: Sun Aug 04, 2019 6:44 am
- Full name: Jonathan Kreuzer
Re: Slow Chess Blitz Classic
Modern Times :
I used UCI/CuteChess gui for the FRC game testing.
I'm not sure about winboard FRC, under the hood v1.5 actually only plays FRC and considers standard position as an FRC position (and if castle flags in a FEN are set and valid for FRC assumes they are correct.) Oh but format for sending castle move in winboard probably not correct, UCI it goes by the UCI_Chess960 option flag to send King X Rook.
Thanks for the testing 2966 looks like a good result so far to me!
Let me know if any issues with FRC UCI (or own GUI.)
I used UCI/CuteChess gui for the FRC game testing.
I'm not sure about winboard FRC, under the hood v1.5 actually only plays FRC and considers standard position as an FRC position (and if castle flags in a FEN are set and valid for FRC assumes they are correct.) Oh but format for sending castle move in winboard probably not correct, UCI it goes by the UCI_Chess960 option flag to send King X Rook.
Thanks for the testing 2966 looks like a good result so far to me!
Let me know if any issues with FRC UCI (or own GUI.)
-
- Posts: 3553
- Joined: Thu Jun 07, 2012 11:02 pm
Re: Slow Chess Blitz Classic
OK good, I was going to use UCI but though it was worth checking.
The standard chess results are from another tester.
The standard chess results are from another tester.