AlphaZero No Castling Chess

Discussion of anything and everything relating to chess playing software and machines.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
User avatar
Laskos
Posts: 9744
Joined: Wed Jul 26, 2006 8:21 pm
Full name: Kai Laskos

Re: AlphaZero No Castling Chess

Post by Laskos » Fri Dec 06, 2019 7:51 pm

lkaufman wrote:
Fri Dec 06, 2019 6:31 pm
Laskos wrote:
Fri Dec 06, 2019 5:51 pm
MTaktikos wrote:
Fri Dec 06, 2019 5:17 pm
Laskos wrote:
Fri Dec 06, 2019 3:16 pm

Thanks, seems to work now. Will check for some properties, interesting. Does the engine know something how to play it, or it plays as usual, only discovering that it lost/won at the last move? I have no time looking at PGNs and games.

I mean, the eval has to be changed due to such adjudication rules, otherwise it will play dumb, unrepresentative games. I don't have time to look at the code, games and PGNs, just left it play 40 games at 2:1 time odds, to see the sensitivity of the game to strength.
No the eval is not specially tweaked for this variant.
But I have seen Fairy Stockfish playing other 8x8 variants without tweaked eval, and nevertheless it plays much stronger (> 400 Elo) than known variant engines, which themselves play at superhuman level. Therefore I don't assume that it will play dumb

P.S. After this weekend I will try to change variants.cpp so that Fairy Stockfish can be tested also for Armageddon chess
Thanks, seems to be playing relatively fine, but I saw only 2-3 games. In 40 games 2 min + 2 s versus 1 min + 1 s the score was 27:13 (only wins and losses, as it should be), in line with the Elo gain for doubling time control of the Standard Chess. So, the game is playable and is sensitive to the strength.
Wait a minute, no draws? I thought this version eliminated draws by stalemate, bare king, and rep, but not fifty move rule. Surely there should be some fifty move rule draws in forty games? Unless the engine was just forfeiting players for repeating, rather than forcing them to play moves that did not repeat. It would be just incredible if my three rule changes reduced the draw percentage to less than 2%, rather I should say unbelievable. Or is this version calling fifty move rule draws wins for Black (my alternate proposal)?
Ah, forgot about that, I deselected the 50 move rule too in the GUI.
Last edited by Laskos on Fri Dec 06, 2019 8:00 pm, edited 2 times in total.

User avatar
Laskos
Posts: 9744
Joined: Wed Jul 26, 2006 8:21 pm
Full name: Kai Laskos

Re: AlphaZero No Castling Chess

Post by Laskos » Fri Dec 06, 2019 7:57 pm

MTaktikos wrote:
Fri Dec 06, 2019 7:43 pm
Wait a minute, no draws? I thought this version eliminated draws by stalemate, bare king, and rep, but not fifty move rule. Surely there should be some fifty move rule draws in forty games? Unless the engine was just forfeiting players for repeating, rather than forcing them to play moves that did not repeat. It would be just incredible if my three rule changes reduced the draw percentage to less than 2%, rather I should say unbelievable. Or is this version calling fifty move rule draws wins for Black (my alternate proposal)?
No, the 50 move rule draws remain possible.
But: a single repetition is a loss, this rule may become decisive. I think about a scenario where the stronger side builds a fortress and presses the weaker side in a smaller area, where it is forced to make a repetition before it's 50 moves are over. Would be a very interesting Zugzwang scenario, wouldn't it?
(My hardware, an older AMD 8350, is not fast enough to make serious tests with the engine)

P.S. Cann't wait to see Stefan Pohl's new work. After the weekend I will have the time to implement something for his Armageddon variant. Hope some day there will exist also a "Fairy Leela" project, through Stefan's advices I learned how to use the Lc0 engine with my GPU
I think I deselected the 50-move rule in the GUI too. When I will be at my PC home, will re-do the match.

lkaufman
Posts: 3837
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: AlphaZero No Castling Chess

Post by lkaufman » Fri Dec 06, 2019 10:02 pm

MTaktikos wrote:
Fri Dec 06, 2019 7:43 pm
Wait a minute, no draws? I thought this version eliminated draws by stalemate, bare king, and rep, but not fifty move rule. Surely there should be some fifty move rule draws in forty games? Unless the engine was just forfeiting players for repeating, rather than forcing them to play moves that did not repeat. It would be just incredible if my three rule changes reduced the draw percentage to less than 2%, rather I should say unbelievable. Or is this version calling fifty move rule draws wins for Black (my alternate proposal)?
No, the 50 move rule draws remain possible.
But: a single repetition is a loss, this rule may become decisive. I think about a scenario where the stronger side builds a fortress and presses the weaker side in a smaller area, where it is forced to make a repetition before it's 50 moves are over. Would be a very interesting Zugzwang scenario, wouldn't it?
(My hardware, an older AMD 8350, is not fast enough to make serious tests with the engine)

P.S. Cann't wait to see Stefan Pohl's new work. After the weekend I will have the time to implement something for his Armageddon variant. Hope some day there will exist also a "Fairy Leela" project, through Stefan's advices I learned how to use the Lc0 engine with my GPU
I got it working, but I think something is wrong. I see it repeating multiple times and only eventually varying, isn't it supposed to refuse to repeat even once? If it sees that playing the normally best move would force a repetition, shouldn't it choose an otherwise inferior but not clearly losing move? Is there some step I could be overlooking in the setup? I am selecting the "larry" variant. How about the adjudication rules, should they remain as default for fifty move and rep rule claims?
Komodo rules!

pohl4711
Posts: 1141
Joined: Sat Sep 03, 2011 5:25 am
Location: Berlin, Germany
Contact:

Re: AlphaZero No Castling Chess

Post by pohl4711 » Sat Dec 07, 2019 12:25 pm

lkaufman wrote:
Fri Dec 06, 2019 7:19 pm

I think that Kai meant that openings from normal chess are not appropriate for NBC chess, they need to have their own openings analyzed by engines that are aware of the rule.
In my new openings (coming until X-mas), which are a new variant of Armageddon, Komodo analyzed all endposition of the opening-lines. And only a small evel-intervall was taken for filtering good lines out of the pool.

pohl4711
Posts: 1141
Joined: Sat Sep 03, 2011 5:25 am
Location: Berlin, Germany
Contact:

Re: AlphaZero No Castling Chess

Post by pohl4711 » Sat Dec 07, 2019 12:27 pm

MTaktikos wrote:
Fri Dec 06, 2019 7:43 pm

P.S. Cann't wait to see Stefan Pohl's new work. After the weekend I will have the time to implement something for his Armageddon variant.
My new openigns are a new variant of Armageddon and well be released soon (before X-mas). The work is almost done, I just have to write the documentation and design the website-section. So, perhaps you should wait for the release...

MTaktikos
Posts: 11
Joined: Fri Oct 25, 2019 5:58 pm
Full name: Michael Taktikos

Re: AlphaZero No Castling Chess

Post by MTaktikos » Sat Dec 07, 2019 1:44 pm

lkaufman wrote:
Fri Dec 06, 2019 10:02 pm
I got it working, but I think something is wrong. I see it repeating multiple times and only eventually varying, isn't it supposed to refuse to repeat even once?
Yes, it is supposed to decide the player with the first repetition as defeated.
With the Winboard GUI I could reproduce the behavior you mentioned, and also some games where a bare king was on the board. After long debugging, it became clear that in the matches the Winboard GUI had switched to usual chess!
A provisional solution was to go to the menu Engine/Engine #1 settings/UCI_variant and chose here larry instead of chess; the same was to be done with Engine #2 settings.
The first game in the match the rules of your chess variant weren't violated, but after a while Winboard switched silently to chess again (argh!)
The only sure way that I found to play the Larrychess variant was to turn the Analysis Mode on and to make manually the moves suggested by the engine.

BTW I have the question which Winboard version used Kai Laskos for his test.
@ Kai: Was it the Winboard Alien version?

User avatar
Laskos
Posts: 9744
Joined: Wed Jul 26, 2006 8:21 pm
Full name: Kai Laskos

Re: AlphaZero No Castling Chess

Post by Laskos » Sat Dec 07, 2019 6:24 pm

MTaktikos wrote:
Sat Dec 07, 2019 1:44 pm


BTW I have the question which Winboard version used Kai Laskos for his test.
@ Kai: Was it the Winboard Alien version?
No, your attached Winboard. I deselected the GUI 50-move and 3-fold rules, was not watching the games much, and assumed it was fine, so I messed up that experiment. When this morning I was more careful, the things didn't work out. So, I quit for now, as it seems Winboard overwrites the adjudication rules.

User avatar
Laskos
Posts: 9744
Joined: Wed Jul 26, 2006 8:21 pm
Full name: Kai Laskos

Re: AlphaZero No Castling Chess

Post by Laskos » Sat Dec 07, 2019 6:36 pm

lkaufman wrote:
Fri Dec 06, 2019 4:25 pm
Laskos wrote:
Fri Dec 06, 2019 1:46 pm
pohl4711 wrote:
Fri Dec 06, 2019 12:13 pm
Laskos wrote:
Fri Dec 06, 2019 9:10 am
I have an even simpler modification, but with a scoring modification in addition. Only Black is not allowed to castle and the Draws are adjudicated as Black wins.
You have an even simpler modification? That is my NBC-Armageddon openings idea. Not yours. I released it in August 2019, with 4 openings-sets. And announced it here on talkchess.

https://www.sp-cc.de/armageddon-openings.htm

From my website:
Level 2: NBC (= No Black Castling): White can castle to both sides, black is not allowed to castle. Line: 1. Na3 Nh6 2. Nb1 Rg8 3. Na3 Rh8 4. Nb1 Ng8 5. Nc3 Na6 6. Nb1 Rb8 7. Na3 Ra8 8. Nb1 Nb8

Level 2 (NBC) testing:

NBC_Armageddon_IM_4moves:
White Wins: 296 (59.2 %), Black Wins: 204 (40.8 %), Draws: 0 (0.0 %) White Score: 59.2 %, Black Score: 40.8 %
1 SF 190728 (half time) : 3477 500 (+304,= 0,-196), 60.8 %
2 Komodo 13.01 : 3400 500 (+196,= 0,-304), 39.2 % (Elo-spreading: 77 Elo)

NBC_Armageddon_SuperGM_4moves:
White Wins: 284 (56.8 %), Black Wins: 216 (43.2 %), Draws: 0 (0.0 %), White Score: 56.8 %, Black Score: 43.2 %
1 SF 190728 (half time) : 3451 500 (+286,= 0,-214), 57.2 %
2 Komodo 13.01 : 3400 500 (+214,= 0,-286), 42.8 % (Elo-spreading: 51 Elo)

NBC_Armageddon_FEOBOS:
White Wins: 287 (57.4 %), Black Wins: 213 (42.6 %), Draws: 0 (0.0 %)m, White Score: 57.4 %, Black Score: 42.6 %
1 SF 190728 (half time) : 3490 500 (+313,= 0,-187), 62.6 %
2 Komodo 13.01 : 3400 500 (+187,= 0,-313), 37.4 % (Elo-spreading: 90 Elo)

NBC_Armageddon_6pawnplies:
White Wins: 277 (55.4 %), Black Wins: 223 (44.6 %), Draws: 0 (0.0 %), White Score: 55.4 %, Black Score: 44.6 %
1 SF 190728 (half time) : 3473 500 (+301,= 0,-199), 60.2 %
2 Komodo 13.01 : 3400 500 (+199,= 0,-301), 39.8 % (Elo-spreading: 73 Elo)
Ah, sorry, I didn't know. So, I re-discovered this morning meddling about handicaps your proposal. Seeing your results, it seems a viable proposal, if Kramnik thinks no castling at all is a viable proposal. I am not sure how many top GMs share Kramnik's view.

Thanks for opening suites too! These are from normal human and computer games? I guess the openings will be somewhat different if one side doesn't castle, maybe I will build with Lc0 and Komodo some short opening suites which are played for this variant.
I was involved with this idea too at the time, along with variants of it such as Black can only castle long, or white can only castle short while Black can only castle long. The pure version you tested seems to favor White too much even with Black winning draws. A more aesthetic version that might be more balanced (perhaps a bit in Black's favor between engines) would be that White can castle normally, but neither player can castle on the same side as his opponent has done. But this requires new programming, not so trivial to test. My subjective opinion as a GM is that the pure version you tested would be seen as too favorable for White (even with the Armageddon rule) for human play.
I built an opening suite with Komodo on 4 threads and Variety option, almost 500 different 5-movers (the unique starting position is set to no Black castling):
http://s000.tinyupload.com/?file_id=235 ... 1942827279

The pure No Black Castling seems balanced with that Armageddon scoring in Komodo self-matches at different time controls from this suite, for example:

60+0.6
White wins: 101/200
Draws: 94/200
Black Wins: 5/200


240+2.4
White wins: 20/40
Draws: 18/40
Black Wins: 2/40

But if you feel that it is too much an advantage for White, maybe I am missing something with these engine matches (Lc0 also sees it as pretty balanced).

lkaufman
Posts: 3837
Joined: Sun Jan 10, 2010 5:15 am
Location: Maryland USA
Contact:

Re: AlphaZero No Castling Chess

Post by lkaufman » Sat Dec 07, 2019 6:54 pm

Laskos wrote:
Sat Dec 07, 2019 6:36 pm
lkaufman wrote:
Fri Dec 06, 2019 4:25 pm
Laskos wrote:
Fri Dec 06, 2019 1:46 pm
pohl4711 wrote:
Fri Dec 06, 2019 12:13 pm
Laskos wrote:
Fri Dec 06, 2019 9:10 am
I have an even simpler modification, but with a scoring modification in addition. Only Black is not allowed to castle and the Draws are adjudicated as Black wins.
You have an even simpler modification? That is my NBC-Armageddon openings idea. Not yours. I released it in August 2019, with 4 openings-sets. And announced it here on talkchess.

https://www.sp-cc.de/armageddon-openings.htm

From my website:
Level 2: NBC (= No Black Castling): White can castle to both sides, black is not allowed to castle. Line: 1. Na3 Nh6 2. Nb1 Rg8 3. Na3 Rh8 4. Nb1 Ng8 5. Nc3 Na6 6. Nb1 Rb8 7. Na3 Ra8 8. Nb1 Nb8

Level 2 (NBC) testing:

NBC_Armageddon_IM_4moves:
White Wins: 296 (59.2 %), Black Wins: 204 (40.8 %), Draws: 0 (0.0 %) White Score: 59.2 %, Black Score: 40.8 %
1 SF 190728 (half time) : 3477 500 (+304,= 0,-196), 60.8 %
2 Komodo 13.01 : 3400 500 (+196,= 0,-304), 39.2 % (Elo-spreading: 77 Elo)

NBC_Armageddon_SuperGM_4moves:
White Wins: 284 (56.8 %), Black Wins: 216 (43.2 %), Draws: 0 (0.0 %), White Score: 56.8 %, Black Score: 43.2 %
1 SF 190728 (half time) : 3451 500 (+286,= 0,-214), 57.2 %
2 Komodo 13.01 : 3400 500 (+214,= 0,-286), 42.8 % (Elo-spreading: 51 Elo)

NBC_Armageddon_FEOBOS:
White Wins: 287 (57.4 %), Black Wins: 213 (42.6 %), Draws: 0 (0.0 %)m, White Score: 57.4 %, Black Score: 42.6 %
1 SF 190728 (half time) : 3490 500 (+313,= 0,-187), 62.6 %
2 Komodo 13.01 : 3400 500 (+187,= 0,-313), 37.4 % (Elo-spreading: 90 Elo)

NBC_Armageddon_6pawnplies:
White Wins: 277 (55.4 %), Black Wins: 223 (44.6 %), Draws: 0 (0.0 %), White Score: 55.4 %, Black Score: 44.6 %
1 SF 190728 (half time) : 3473 500 (+301,= 0,-199), 60.2 %
2 Komodo 13.01 : 3400 500 (+199,= 0,-301), 39.8 % (Elo-spreading: 73 Elo)
Ah, sorry, I didn't know. So, I re-discovered this morning meddling about handicaps your proposal. Seeing your results, it seems a viable proposal, if Kramnik thinks no castling at all is a viable proposal. I am not sure how many top GMs share Kramnik's view.

Thanks for opening suites too! These are from normal human and computer games? I guess the openings will be somewhat different if one side doesn't castle, maybe I will build with Lc0 and Komodo some short opening suites which are played for this variant.
I was involved with this idea too at the time, along with variants of it such as Black can only castle long, or white can only castle short while Black can only castle long. The pure version you tested seems to favor White too much even with Black winning draws. A more aesthetic version that might be more balanced (perhaps a bit in Black's favor between engines) would be that White can castle normally, but neither player can castle on the same side as his opponent has done. But this requires new programming, not so trivial to test. My subjective opinion as a GM is that the pure version you tested would be seen as too favorable for White (even with the Armageddon rule) for human play.
I built an opening suite with Komodo on 4 threads and Variety option, almost 500 different 5-movers (the unique starting position is set to no Black castling):
http://s000.tinyupload.com/?file_id=235 ... 1942827279

The pure No Black Castling seems balanced with that Armageddon scoring in Komodo self-matches at different time controls from this suite, for example:

60+0.6
White wins: 101/200
Draws: 94/200
Black Wins: 5/200


240+2.4
White wins: 20/40
Draws: 18/40
Black Wins: 2/40

But if you feel that it is too much an advantage for White, maybe I am missing something with these engine matches (Lc0 also sees it as pretty balanced).
Wow, that's a surprise, rather different result from what S. Pohl got, but he used "normal" openings so I think your result is more likely correct. I don't think you are "missing something", it's just that humans much prefer to attack than to try to hold a bad position, and so nearly every GM would rather play White in this scenario, and most likely White would score well over 50%. Engines can't replicate human behavior perfectly yet, and so such simulations are only a rough guide to what would happen with human play. But your result is quite significant, it may mean that this variant is ideal for correspondence or engine vs engine matches, and perhaps even with human OTB play White's advantage would be within acceptable bounds given that the players always play two game sets. But then tie scores become likely again!
Well, one point you may have missed is that if the engines were aware that draws counted as Black wins and modified to play accordingly, White would surely have won at least a few of the drawn games, so I would say that the result does show some White advantage, but perhaps not too much. Humans would avoid anything looking even somewhat drawish as White, preferring an unclear attack to a small endgame plus.
Komodo rules!

User avatar
Laskos
Posts: 9744
Joined: Wed Jul 26, 2006 8:21 pm
Full name: Kai Laskos

Re: AlphaZero No Castling Chess

Post by Laskos » Sat Dec 07, 2019 7:22 pm

lkaufman wrote:
Sat Dec 07, 2019 6:54 pm
Laskos wrote:
Sat Dec 07, 2019 6:36 pm
lkaufman wrote:
Fri Dec 06, 2019 4:25 pm
Laskos wrote:
Fri Dec 06, 2019 1:46 pm
pohl4711 wrote:
Fri Dec 06, 2019 12:13 pm
Laskos wrote:
Fri Dec 06, 2019 9:10 am
I have an even simpler modification, but with a scoring modification in addition. Only Black is not allowed to castle and the Draws are adjudicated as Black wins.
You have an even simpler modification? That is my NBC-Armageddon openings idea. Not yours. I released it in August 2019, with 4 openings-sets. And announced it here on talkchess.

https://www.sp-cc.de/armageddon-openings.htm

From my website:
Level 2: NBC (= No Black Castling): White can castle to both sides, black is not allowed to castle. Line: 1. Na3 Nh6 2. Nb1 Rg8 3. Na3 Rh8 4. Nb1 Ng8 5. Nc3 Na6 6. Nb1 Rb8 7. Na3 Ra8 8. Nb1 Nb8

Level 2 (NBC) testing:

NBC_Armageddon_IM_4moves:
White Wins: 296 (59.2 %), Black Wins: 204 (40.8 %), Draws: 0 (0.0 %) White Score: 59.2 %, Black Score: 40.8 %
1 SF 190728 (half time) : 3477 500 (+304,= 0,-196), 60.8 %
2 Komodo 13.01 : 3400 500 (+196,= 0,-304), 39.2 % (Elo-spreading: 77 Elo)

NBC_Armageddon_SuperGM_4moves:
White Wins: 284 (56.8 %), Black Wins: 216 (43.2 %), Draws: 0 (0.0 %), White Score: 56.8 %, Black Score: 43.2 %
1 SF 190728 (half time) : 3451 500 (+286,= 0,-214), 57.2 %
2 Komodo 13.01 : 3400 500 (+214,= 0,-286), 42.8 % (Elo-spreading: 51 Elo)

NBC_Armageddon_FEOBOS:
White Wins: 287 (57.4 %), Black Wins: 213 (42.6 %), Draws: 0 (0.0 %)m, White Score: 57.4 %, Black Score: 42.6 %
1 SF 190728 (half time) : 3490 500 (+313,= 0,-187), 62.6 %
2 Komodo 13.01 : 3400 500 (+187,= 0,-313), 37.4 % (Elo-spreading: 90 Elo)

NBC_Armageddon_6pawnplies:
White Wins: 277 (55.4 %), Black Wins: 223 (44.6 %), Draws: 0 (0.0 %), White Score: 55.4 %, Black Score: 44.6 %
1 SF 190728 (half time) : 3473 500 (+301,= 0,-199), 60.2 %
2 Komodo 13.01 : 3400 500 (+199,= 0,-301), 39.8 % (Elo-spreading: 73 Elo)
Ah, sorry, I didn't know. So, I re-discovered this morning meddling about handicaps your proposal. Seeing your results, it seems a viable proposal, if Kramnik thinks no castling at all is a viable proposal. I am not sure how many top GMs share Kramnik's view.

Thanks for opening suites too! These are from normal human and computer games? I guess the openings will be somewhat different if one side doesn't castle, maybe I will build with Lc0 and Komodo some short opening suites which are played for this variant.
I was involved with this idea too at the time, along with variants of it such as Black can only castle long, or white can only castle short while Black can only castle long. The pure version you tested seems to favor White too much even with Black winning draws. A more aesthetic version that might be more balanced (perhaps a bit in Black's favor between engines) would be that White can castle normally, but neither player can castle on the same side as his opponent has done. But this requires new programming, not so trivial to test. My subjective opinion as a GM is that the pure version you tested would be seen as too favorable for White (even with the Armageddon rule) for human play.
I built an opening suite with Komodo on 4 threads and Variety option, almost 500 different 5-movers (the unique starting position is set to no Black castling):
http://s000.tinyupload.com/?file_id=235 ... 1942827279

The pure No Black Castling seems balanced with that Armageddon scoring in Komodo self-matches at different time controls from this suite, for example:

60+0.6
White wins: 101/200
Draws: 94/200
Black Wins: 5/200


240+2.4
White wins: 20/40
Draws: 18/40
Black Wins: 2/40

But if you feel that it is too much an advantage for White, maybe I am missing something with these engine matches (Lc0 also sees it as pretty balanced).
Wow, that's a surprise, rather different result from what S. Pohl got, but he used "normal" openings so I think your result is more likely correct. I don't think you are "missing something", it's just that humans much prefer to attack than to try to hold a bad position, and so nearly every GM would rather play White in this scenario, and most likely White would score well over 50%. Engines can't replicate human behavior perfectly yet, and so such simulations are only a rough guide to what would happen with human play. But your result is quite significant, it may mean that this variant is ideal for correspondence or engine vs engine matches, and perhaps even with human OTB play White's advantage would be within acceptable bounds given that the players always play two game sets. But then tie scores become likely again!
Well, one point you may have missed is that if the engines were aware that draws counted as Black wins and modified to play accordingly, White would surely have won at least a few of the drawn games, so I would say that the result does show some White advantage, but perhaps not too much. Humans would avoid anything looking even somewhat drawish as White, preferring an unclear attack to a small endgame plus.
Ah, forgot to specify:
I used
Contempt = 25
White Contempt = True

Does that make sense to you? I am not sure how Black behaves in this case, maybe you can tell me. My goal was that White would chase Wins, Black would chase Draws and Wins.

Post Reply