Re: AlphaZero No Castling Chess
Posted: Fri Dec 06, 2019 9:16 am
Good to know the tastes of super-GMs like Kramnik for what they consider as acceptable Chess variants. This variant is cute and if adopted, will soon develop its own opening theory. The problem is, the draw rate decreases just a bit, if played competently. I can mimic competent play by pitting the best T40 and T30 nets to have some diversity, on a pretty powerful OC-ed RTX 2070 GPU. I used a temperature of 0.5 for first 5 moves, so opening diversity is guaranteed. The results are here:Javier Ros wrote: ↑Tue Dec 03, 2019 5:30 pm It seems that AlphaZero has been trained to play "no-castling chess", see the interesting article
Kramnik And AlphaZero: How To Rethink Chess
at
https://www.chess.com/article/view/no-c ... -alphazero
Regarding small changes, I think the first side that can capture all opposing pieces could get a free pawn, but the opponent is allowed to choose where on the second row the pawn is placed.lkaufman wrote: ↑Thu Dec 05, 2019 6:58 pmIf you want a much-superior game related to but very unlike chess, just take up shogi! That's what I did. But a lot of people want to play a game that retains all the elements of chess, the same tactics, same positional judgment, etc., but just without the need for memorizing theory and without so many draws. It's easy to fix either one of these problems, by FRC in the first case or by scoring changes for various types of draws in the second. To fix both, I suppose you need to combine FRC with the scoring changes.Ovyron wrote: ↑Thu Dec 05, 2019 9:38 am Hold on, back when the queen was made powerful in chess, wasn't that a radical change that made the game completely different? Yet it was adopted, so why do people want to keep the nature of chess unchanged? Why are people hesitant to add rules that destroy the nature of chess, if that was done in the past and the end-product was a superior game?
Just, stop being afraid of ending with a game very unlike chess, if it's a much superior game people will adopt it and welcome the change.
I think no castling increases the tactical sharpness (if that's the right word) of the game.Laskos wrote: ↑Fri Dec 06, 2019 10:10 amGood to know the tastes of super-GMs like Kramnik for what they consider as acceptable Chess variants. This variant is cute and if adopted, will soon develop its own opening theory. The problem is, the draw rate decreases just a bit, if played competently. I can mimic competent play by pitting the best T40 and T30 nets to have some diversity, on a pretty powerful OC-ed RTX 2070 GPU. I used a temperature of 0.5 for first 5 moves, so opening diversity is guaranteed. The results are here:Javier Ros wrote: ↑Tue Dec 03, 2019 5:30 pm It seems that AlphaZero has been trained to play "no-castling chess", see the interesting article
Kramnik And AlphaZero: How To Rethink Chess
at
https://www.chess.com/article/view/no-c ... -alphazero
TC 60'' + 0.6''
Standard Chess:
Score of lc0_T40B4_200 vs lc0_32930: 39 - 4 - 157 [0.588] 200
Elo difference: 61.43 +/- 21.33
Finished match
No Castling Chess:
Score of lc0_T40B4_200 vs lc0_32930: 43 - 4 - 153 [0.598] 200
Elo difference: 68.63 +/- 22.22
Finished match
The draw rate is only decreased by a little, hardly significantly. The White performance in both cases is about 55%. I am not sure whether it is worth to adopt this variant, which quite possibly, after competent play is achieved, will again go to 80%+ draws as the Standard Chess in super-GM matches.
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. Two things are achieved: the draw rate will hover at some 40%-50% in super-GM games, even when played very competently, the starting position is very borderline White win / old Draw. The second thing is that it inflates the Elo difference, therefore the resolution power of the variant compared to the Standard Chess (if discarding all reciprocal wins in pairs as draws, when calculating the error margins). But this variant has to be played in pairs of two games, reversed colors. Here is the result and White wins share:
Black No Castling Chess (side and reversed):
Score of lc0_T40B4_200 vs lc0_32930: 83 - 22 - 95 [0.652] 200
Elo difference: 109.45 +/- 34.99
Finished match
You see, the Elo difference between T40 and T30 here increased significantly. The Breakdown on colors is the following:
White wins: 98/200
Draws: 95/200
Black Wins: 7/200
So, almost a complete balance White wins versus the rest, and the draw rate is below 50%.
I will recheck with a randomized Komodo on 4 threads this result.
The source code is included, I had made only the following addition in the file variants.cpp;
As you said, the side that first makes a repetition should lose, also the first side with a bare king.Variant* larry_variant() {
Variant* v = chess_variant();
v->nFoldValue = -VALUE_MATE;
v->nFoldRule = 1;
v->stalemateValue = -VALUE_MATE;
v->bareKingValue = -VALUE_MATE;
return v;
}
@Laskos
To make the pure engine was easy, but now we need a GUI that support this variant...
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.
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.MTaktikos wrote: ↑Fri Dec 06, 2019 1:10 pm@Laskos
Looks like the incorrect adjucation comes not from the engine SFLarry, but is an Xboard adjucation by the GUI
Please try the following:
1) In Winboard's menu Options/Adjucations uncheck the points, where Winboard adjucates itself insufficient material, trivial draws,
and verifies engine claims
2) In Winboard Options/General uncheck the "test legality" by Winboard
3) Chose New variant/larry
4) In Edit/Paste position from clipboard copy-paste the position
4b1NK/3pPp1N/3P1PpP/6Pk/6p1/6p1/6P1/8 w - - 0 1
(in this position, Black ist the side that has by Zugzwang to make the first repetition
5.) Chose Mode/Analysis mode. SFLarry should (correctly) anounce a mate
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.pohl4711 wrote: ↑Fri Dec 06, 2019 1:13 pmYou 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)