Draughts has not been solved, can a chess programmer solve it ?

[Chessqueen]

From what I read here is that draughts could be solved - but that the game itself has a 100x smaller community than chess - so it would be not so interesting in that sense to do so.

IMO - the 8x8 solvable relatively quickly. The 10x10 maybe a lot more effort.
If I google this:
Is't checkers solved already??
http://sciencenetlinks.com/science-news ... rs-solved/

I believe that since the engine Chinook solved checkers on a 8x8 board Google has not changed the information and most of the time people confuse International Draughts played on 10x10 with 8x8 checkers which has not been solved, that is if Chinook play infinite number of games all the games will ended in draws, whereas, in 10x10 computer Draughts championship as you saw in the last Draughts one most of the games ended in draws, but not all the games, and in order to be solved all the games have to end in a draw even if the engine Damage NNUE by Bert Tuyt play by itself 1000 games all the games will not end in draw

[Chessqueen]

You make 2 mistakes here, first of all Draughts is actually played on a 50 square board, secondly the level of Draughts programs in 2012 wasn't very high, maybe comparable in complexity to chess programs with 2500 CCRL rating. Even today are top level Draughts programs relatively simple and way less advanced than top level Chess engines.

The main reason is that there is hardly any development going on in computer Draughts. If you look at the computer Draughts forum there are just a few posts per year, sometimes several months are passing by without a single message.

I guess there is no comparable massive collaborative effort like Stockfish in draughts, but if Fabien Letouzey's Senpai can have a near 3000 CCRL rating, then I think it's safe to assume that his program Scan is at least having a similarly advanced search function. So I wouldn't disqualify the level of quality of draughts programs. In my opinion, they main difference is that massive amount of testing that goes into Stockfish.

I was talking about the complexity and level of draughts programs back in 2012. The last 6 years or so their level increased by several hundred Elo points, basically due to machine learning, and even today it does not seem very difficult to reach the level of top draughts engines.

When I look at my own draughts engine it is all very basic, just a PVS search without any fancy stuff, the same I already did back in the nineties with my chess engine. The neural net is the main ingredient for it's strength, and not the complexity of the search.

Joost, I believe that some or most Draughts engines that use network have problem understanding and using positions with a few stones left on the board, and there should be an algorithm that needs to be implemented or programmed with a neural network that automatically update itself while playing either by itself or other draughts engines without having to play thousands of games to have the network learn from those games, and in addition to this algorithm it should be combined with an EGDB of at least 7 stones or 7P FDB. Just my two cents.

[Joost Buijs]

Joost, I believe that some or most Draughts engines that use network have problem understanding and using positions with a few stones left on the board, and there should be an algorithm that needs to be implemented or programmed with a neural network that automatically update itself while playing either by itself or other draughts engines without having to play thousands of games to have the network learn from those games, and in addition to this algorithm it should be combined with an EGDB of at least 7 stones or 7P FDB. Just my two cents.

You are right, this is something that I noticed too when I was training the draughts network. To solve the problem you could have the engine always play with an EGDB. Another possibility could be to train a second network for positions with <= 8 pieces using the WDL information from an EGDB.

I trained the draughts network without EGDB and used positions till the very end, it would be easier for the network to learn when it is trained with positions containing more than 8 pieces only. Still it did reasonably well during the 2021 unofficial WK, maybe it would have done better with an EGDB, I dont know.

[Madeleine Birchfield]

You make 2 mistakes here, first of all Draughts is actually played on a 50 square board, secondly the level of Draughts programs in 2012 wasn't very high, maybe comparable in complexity to chess programs with 2500 CCRL rating. Even today are top level Draughts programs relatively simple and way less advanced than top level Chess engines.

The main reason is that there is hardly any development going on in computer Draughts. If you look at the computer Draughts forum there are just a few posts per year, sometimes several months are passing by without a single message.

I guess there is no comparable massive collaborative effort like Stockfish in draughts, but if Fabien Letouzey's Senpai can have a near 3000 CCRL rating, then I think it's safe to assume that his program Scan is at least having a similarly advanced search function. So I wouldn't disqualify the level of quality of draughts programs. In my opinion, they main difference is that massive amount of testing that goes into Stockfish.

I think the lack of activity in computer draughts is largely because, despite its name, international draughts is largely an European phenomenon. Rarely do people from America or China or India, the three most populous countries in the world, play international draughts, while chess is played in all of them.

[Chessqueen]

Joost, I believe that some or most Draughts engines that use network have problem understanding and using positions with a few stones left on the board, and there should be an algorithm that needs to be implemented or programmed with a neural network that automatically update itself while playing either by itself or other draughts engines without having to play thousands of games to have the network learn from those games, and in addition to this algorithm it should be combined with an EGDB of at least 7 stones or 7P FDB. Just my two cents.

You are right, this is something that I noticed too when I was training the draughts network. To solve the problem you could have the engine always play with an EGDB. Another possibility could be to train a second network for positions with <= 8 pieces using the WDL information from an EGDB.

I trained the draughts network without EGDB and used positions till the very end, it would be easier for the network to learn when it is trained with positions containing more than 8 pieces only. Still it did reasonably well during the 2021 unofficial WK, maybe it would have done better with an EGDB, I dont know.

I just noticed that your Draughts engine is one of the few along with Cerberus and Dream that does NOT use base of endings or EGDB and managed to get one of the top 6th best places. Probably in the next Draughts World Championship it will end up in the Top 3 best places after you add EGDB

Top 6 best Draughts engines and their Authors ==> https://www.uwccpi2021.draughtsprograms ... tings.html
Final result ==> https://www.uwccpi2021.draughtsprograms ... usion.html

Note: Anyway, any progress on your latest version of Nightmare chess engines. Have you included NNUE and EDGB to your Nightmare as Well, or planning to in the near future
https://www.chessprogramming.org/Joost_Buijs

[pgn][Site "Internet Chess Club"]
[Date "2013.07.20"]
[Round "6"]
[White "Nightmare"]
[Black "Arasan"]
[Result "1/2-1/2"]

1.e4 c6 2.d4 d5 3.e5 Bf5 4.Nf3 e6 5.Be2 Nd7 6.O-O Bg6 7.Nbd2 Ne7 8.Nh4 c5
9.c3 Qb6 10.Ndf3 Be4 11.dxc5 Qxc5 12.Nd4 Nc6 13.Be3 Qe7 14.f3 Qxh4 15.Bf2
Qg5 16.fxe4 dxe4 17.Qa4 Ncxe5 18.Nb5 Qd8 19.Qxe4 a6 20.Bg3 Qb6+ 21.Kh1 axb5
22.Bxe5 Qc6 23.Qe3 Nxe5 24.Qxe5 f6 25.Rxf6 gxf6 26.Qxf6 Rg8 27.Bh5+ Rg6
28.Bxg6+ hxg6 29.Rf1 Kd7 30.Qf7+ Kd6 31.Rd1+ Ke5 32.h4 Rc8 33.Rd4 Bh6
34.Qxg6 Rh8 35.Rb4 Kd6 36.Qf6 Qe8 37.a4 Be3 38.Re4 Ba7 39.Qe5+ Kd7 40.Qxb5+
Kc8 41.Qe5 Qd8 42.Rc4+ Kd7 43.g3 Qb8 44.Qg7+ Kd6 45.Rb4 Qf8 46.Qxf8+ Rxf8
47.Rxb7 e5 48.Rxa7 e4 49.b3 e3 50.Ra6+ Kd5 51.c4+ Ke5 52.Ra5+ Ke4 53.Ra6 Rf1+
54.Kg2 e2 55.Re6+ Kd3 56.g4 e1=Q 57.Rxe1 Rxe1 58.h5 Ke3 59.b4 Kf4 60.b5 Re2+
61.Kf1 Kf3 62.a5 Ra2 63.Ke1 Ke3 64.Kf1 Rf2+ 65.Kg1 Kf3 66.b6 Rg2+ 67.Kf1 Ra2
68.Ke1 Ke3 69.Kd1 Kd3 70.Kc1 Kc3 71.Kd1 Kd3 72.Kc1 Kc3 73.Kd1 Kd3 1/2-1/2[/pgn]

[Joost Buijs]

I just noticed that your Draughts engine is one of the few along with Cerberus and Dream that does NOT use base of endings or EGDB and managed to get one of the top 6th best places. Probably in the next Draughts World Championship it will end up in the Top 3 best places after you add EGDB

Top 6 best Draughts engines and their Authors ==> https://www.uwccpi2021.draughtsprograms ... tings.html
Final result ==> https://www.uwccpi2021.draughtsprograms ... usion.html

Note: Anyway, any progress on your latest version of Nightmare chess engines. Have you included NNUE and EDGB to your Nightmare as Well, or planning to in the near future
https://www.chessprogramming.org/Joost_Buijs

[pgn][Site "Internet Chess Club"]
[Date "2013.07.20"]
[Round "6"]
[White "Nightmare"]
[Black "Arasan"]
[Result "1/2-1/2"]

1.e4 c6 2.d4 d5 3.e5 Bf5 4.Nf3 e6 5.Be2 Nd7 6.O-O Bg6 7.Nbd2 Ne7 8.Nh4 c5
9.c3 Qb6 10.Ndf3 Be4 11.dxc5 Qxc5 12.Nd4 Nc6 13.Be3 Qe7 14.f3 Qxh4 15.Bf2
Qg5 16.fxe4 dxe4 17.Qa4 Ncxe5 18.Nb5 Qd8 19.Qxe4 a6 20.Bg3 Qb6+ 21.Kh1 axb5
22.Bxe5 Qc6 23.Qe3 Nxe5 24.Qxe5 f6 25.Rxf6 gxf6 26.Qxf6 Rg8 27.Bh5+ Rg6
28.Bxg6+ hxg6 29.Rf1 Kd7 30.Qf7+ Kd6 31.Rd1+ Ke5 32.h4 Rc8 33.Rd4 Bh6
34.Qxg6 Rh8 35.Rb4 Kd6 36.Qf6 Qe8 37.a4 Be3 38.Re4 Ba7 39.Qe5+ Kd7 40.Qxb5+
Kc8 41.Qe5 Qd8 42.Rc4+ Kd7 43.g3 Qb8 44.Qg7+ Kd6 45.Rb4 Qf8 46.Qxf8+ Rxf8
47.Rxb7 e5 48.Rxa7 e4 49.b3 e3 50.Ra6+ Kd5 51.c4+ Ke5 52.Ra5+ Ke4 53.Ra6 Rf1+
54.Kg2 e2 55.Re6+ Kd3 56.g4 e1=Q 57.Rxe1 Rxe1 58.h5 Ke3 59.b4 Kf4 60.b5 Re2+
61.Kf1 Kf3 62.a5 Ra2 63.Ke1 Ke3 64.Kf1 Rf2+ 65.Kg1 Kf3 66.b6 Rg2+ 67.Kf1 Ra2
68.Ke1 Ke3 69.Kd1 Kd3 70.Kc1 Kc3 71.Kd1 Kd3 72.Kc1 Kc3 73.Kd1 Kd3 1/2-1/2[/pgn]

Nightmare progressed a lot since 2013, I removed a lot of bugs, tuned the evaluation function with Texel tuning back in 2016, about 1 year ago I replaced my HCE with NNUE evaluation. I only use it to play the monthly blitz tourney and I have no idea what it's level is. My guess is that it currently sits at the same level as Arasan and Nemorino.

In the ACCA tournament of 2019 it got 4th place.

http://aigames.net/ACCA/ACCAWCRCC/2019A ... sults.html

In the last ACCA tournament of 2020 it got 7th place, this was still without NNUE evaluation.

http://aigames.net/ACCA/ACCAWCRCC/2020A ... sults.html

I prefer to keep the engine private, if I release it I have to give support and that is something I don't want. There are already enough engines for everybody to download, and I don't see the added value of yet another engine.

Edit: Nightmare uses 6P Nalimov EGTB, with permission from Eugene Nalimov and Andrew Kadatch, I never cared to add syzygy, maybe I will in the future.

[dangi12012]

I just noticed that your Draughts engine is one of the few along with Cerberus and Dream that does NOT use base of endings or EGDB and managed to get one of the top 6th best places. Probably in the next Draughts World Championship it will end up in the Top 3 best places after you add EGDB

Top 6 best Draughts engines and their Authors ==> https://www.uwccpi2021.draughtsprograms ... tings.html
Final result ==> https://www.uwccpi2021.draughtsprograms ... usion.html

Note: Anyway, any progress on your latest version of Nightmare chess engines. Have you included NNUE and EDGB to your Nightmare as Well, or planning to in the near future
https://www.chessprogramming.org/Joost_Buijs

[pgn][Site "Internet Chess Club"]
[Date "2013.07.20"]
[Round "6"]
[White "Nightmare"]
[Black "Arasan"]
[Result "1/2-1/2"]

1.e4 c6 2.d4 d5 3.e5 Bf5 4.Nf3 e6 5.Be2 Nd7 6.O-O Bg6 7.Nbd2 Ne7 8.Nh4 c5
9.c3 Qb6 10.Ndf3 Be4 11.dxc5 Qxc5 12.Nd4 Nc6 13.Be3 Qe7 14.f3 Qxh4 15.Bf2
Qg5 16.fxe4 dxe4 17.Qa4 Ncxe5 18.Nb5 Qd8 19.Qxe4 a6 20.Bg3 Qb6+ 21.Kh1 axb5
22.Bxe5 Qc6 23.Qe3 Nxe5 24.Qxe5 f6 25.Rxf6 gxf6 26.Qxf6 Rg8 27.Bh5+ Rg6
28.Bxg6+ hxg6 29.Rf1 Kd7 30.Qf7+ Kd6 31.Rd1+ Ke5 32.h4 Rc8 33.Rd4 Bh6
34.Qxg6 Rh8 35.Rb4 Kd6 36.Qf6 Qe8 37.a4 Be3 38.Re4 Ba7 39.Qe5+ Kd7 40.Qxb5+
Kc8 41.Qe5 Qd8 42.Rc4+ Kd7 43.g3 Qb8 44.Qg7+ Kd6 45.Rb4 Qf8 46.Qxf8+ Rxf8
47.Rxb7 e5 48.Rxa7 e4 49.b3 e3 50.Ra6+ Kd5 51.c4+ Ke5 52.Ra5+ Ke4 53.Ra6 Rf1+
54.Kg2 e2 55.Re6+ Kd3 56.g4 e1=Q 57.Rxe1 Rxe1 58.h5 Ke3 59.b4 Kf4 60.b5 Re2+
61.Kf1 Kf3 62.a5 Ra2 63.Ke1 Ke3 64.Kf1 Rf2+ 65.Kg1 Kf3 66.b6 Rg2+ 67.Kf1 Ra2
68.Ke1 Ke3 69.Kd1 Kd3 70.Kc1 Kc3 71.Kd1 Kd3 72.Kc1 Kc3 73.Kd1 Kd3 1/2-1/2[/pgn]

Nightmare progressed a lot since 2013, I removed a lot of bugs, tuned the evaluation function with Texel tuning back in 2016, about 1 year ago I replaced my HCE with NNUE evaluation. I only use it to play the monthly blitz tourney and I have no idea what it's level is. My guess is that it currently sits at the same level as Arasan and Nemorino.

In the ACCA tournament of 2019 it got 4th place.

http://aigames.net/ACCA/ACCAWCRCC/2019A ... sults.html

In the last ACCA tournament of 2020 it got 7th place, this was still without NNUE evaluation.

http://aigames.net/ACCA/ACCAWCRCC/2020A ... sults.html

I prefer to keep the engine private, if I release it I have to give support and that is something I don't want. There are already enough engines for everybody to download, and I don't see the added value of yet another engine.

Edit: Nightmare uses 6P Nalimov EGTB, with permission from Eugene Nalimov and Andrew Kadatch, I never cared to add syzygy, maybe I will in the future.

You add a lot of value if your engine is written differently or has new approaches. If its an optimized clone there is no value in releasing.

[Joost Buijs]

You add a lot of value if your engine is written differently or has new approaches. If its an optimized clone there is no value in releasing.

Differently than what??? If you mean differently than SF and clones, it is, my engine already existed many years before SF first saw the light.

I started with computer chess back in 1977, one of the reasons that I'm not very interested in it anymore.

[Chessqueen]

I just noticed that your Draughts engine is one of the few along with Cerberus and Dream that does NOT use base of endings or EGDB and managed to get one of the top 6th best places. Probably in the next Draughts World Championship it will end up in the Top 3 best places after you add EGDB

Top 6 best Draughts engines and their Authors ==> https://www.uwccpi2021.draughtsprograms ... tings.html
Final result ==> https://www.uwccpi2021.draughtsprograms ... usion.html

Note: Anyway, any progress on your latest version of Nightmare chess engines. Have you included NNUE and EDGB to your Nightmare as Well, or planning to in the near future
https://www.chessprogramming.org/Joost_Buijs

[pgn][Site "Internet Chess Club"]
[Date "2013.07.20"]
[Round "6"]
[White "Nightmare"]
[Black "Arasan"]
[Result "1/2-1/2"]

1.e4 c6 2.d4 d5 3.e5 Bf5 4.Nf3 e6 5.Be2 Nd7 6.O-O Bg6 7.Nbd2 Ne7 8.Nh4 c5
9.c3 Qb6 10.Ndf3 Be4 11.dxc5 Qxc5 12.Nd4 Nc6 13.Be3 Qe7 14.f3 Qxh4 15.Bf2
Qg5 16.fxe4 dxe4 17.Qa4 Ncxe5 18.Nb5 Qd8 19.Qxe4 a6 20.Bg3 Qb6+ 21.Kh1 axb5
22.Bxe5 Qc6 23.Qe3 Nxe5 24.Qxe5 f6 25.Rxf6 gxf6 26.Qxf6 Rg8 27.Bh5+ Rg6
28.Bxg6+ hxg6 29.Rf1 Kd7 30.Qf7+ Kd6 31.Rd1+ Ke5 32.h4 Rc8 33.Rd4 Bh6
34.Qxg6 Rh8 35.Rb4 Kd6 36.Qf6 Qe8 37.a4 Be3 38.Re4 Ba7 39.Qe5+ Kd7 40.Qxb5+
Kc8 41.Qe5 Qd8 42.Rc4+ Kd7 43.g3 Qb8 44.Qg7+ Kd6 45.Rb4 Qf8 46.Qxf8+ Rxf8
47.Rxb7 e5 48.Rxa7 e4 49.b3 e3 50.Ra6+ Kd5 51.c4+ Ke5 52.Ra5+ Ke4 53.Ra6 Rf1+
54.Kg2 e2 55.Re6+ Kd3 56.g4 e1=Q 57.Rxe1 Rxe1 58.h5 Ke3 59.b4 Kf4 60.b5 Re2+
61.Kf1 Kf3 62.a5 Ra2 63.Ke1 Ke3 64.Kf1 Rf2+ 65.Kg1 Kf3 66.b6 Rg2+ 67.Kf1 Ra2
68.Ke1 Ke3 69.Kd1 Kd3 70.Kc1 Kc3 71.Kd1 Kd3 72.Kc1 Kc3 73.Kd1 Kd3 1/2-1/2[/pgn]

Nightmare progressed a lot since 2013, I removed a lot of bugs, tuned the evaluation function with Texel tuning back in 2016, about 1 year ago I replaced my HCE with NNUE evaluation. I only use it to play the monthly blitz tourney and I have no idea what it's level is. My guess is that it currently sits at the same level as Arasan and Nemorino.

In the ACCA tournament of 2019 it got 4th place.

http://aigames.net/ACCA/ACCAWCRCC/2019A ... sults.html

In the last ACCA tournament of 2020 it got 7th place, this was still without NNUE evaluation.

http://aigames.net/ACCA/ACCAWCRCC/2020A ... sults.html

I prefer to keep the engine private, if I release it I have to give support and that is something I don't want. There are already enough engines for everybody to download, and I don't see the added value of yet another engine.

Edit: Nightmare uses 6P Nalimov EGTB, with permission from Eugene Nalimov and Andrew Kadatch, I never cared to add syzygy, maybe I will in the future.

You do not need to give support to it, simply send it to CCRL once a year lets say every Christmas with instruction not to release it, that is is for testing purpose only.

[Joost Buijs]

You do not need to give support to it, simply send it to CCRL once a year lets say every Christmas with instruction not to release it, that is is for testing purpose only.

Thank you, I'm quite capable of testing the engine myself. What I don't like about CCRL is that they test with different hardware, with time controls adjusted to get the performance of an I7-4770K. This has an impact that is not for all engines the same, so you will always get skewed results.

Also using hyper-threading or SMT is something to avoid, in the past I noticed that some testers use hyper-threading, e.g. running 7 threads on 4 cores with hyper-threading or something like that. I agree that with enough games it will probably average out but I simply don't like it.