The Unofficial World Championships Of Computer Programs In International Draughts Blitz 2020

Discussion about development of draughts in the time of computer and Internet.
BertTuyt
Posts: 1515
Joined: Wed Sep 01, 2004 19:42

Re: The Unofficial World Championships Of Computer Programs In International Draughts Blitz 2020

Post by BertTuyt » Tue Sep 15, 2020 09:52

Krzystof,

you are right my program uses fixed time to do the move. This is not optimal, and others use better time management.

The way Damage calculates the time needed for the search is simple, the GUI communicates the number of moves left and the time left, and based upon this info Damage calculates the average.You can see that in the left bottom window, where this information is presented.The search is exactly aborted after this time (others do this in another way by finalizing the current search depth)

Normally with 16 threads and a 7P DB, 3 seconds for a search is a lot. For humans it is quick, but my engine calculates in this case on your machine, most likely 100M positions/seconds.

And with this speed, and the related search depth, one should not expect blunders (certainly not in the late middle game).
What i (and many others i guess) perceive is that around move 40 in most cases programs (at least if they have sufficient endgame DB, like the 6P DB) know the final outcome of the game, and programs tend to play perfect from there.

Also Damage does not clean the hash-table between moves. This has some advantages.
First of all the hash table cleaning takes time, secondly we can use the results from previous searches, which are stored in the hash-table.
For this we have a (so called) age counter, you can see the value in the right bottom window of the engine.

With this additional age information (which is also stored in the hash-table) you can do several things, for example you can just ignore the hashtable info, if the age counter and the age info in the hashtable dont match.

In my case i use the previous search results.
What most likely goes wrong is that the hashtable (for whatever reason) is corrupted, so information is not valid.

With a single thread this is easy to debug, with multi-threading this is not straightforward.

Bert

BertTuyt
Posts: 1515
Joined: Wed Sep 01, 2004 19:42

Re: The Unofficial World Championships Of Computer Programs In International Draughts Blitz 2020

Post by BertTuyt » Tue Sep 15, 2020 10:10

And forgot to mention, match setup is right.

Bert

Krzysztof Grzelak
Posts: 1054
Joined: Thu Jun 20, 2013 17:16
Real name: Krzysztof Grzelak

Re: The Unofficial World Championships Of Computer Programs In International Draughts Blitz 2020

Post by Krzysztof Grzelak » Tue Sep 15, 2020 10:14

I understand and thank you for your answer Bert. I thought to buy an AMD Ryzen Threadripper AMD Ryzen Threadripper 3970X or AMD Ryzen Threadripper 3990X, processor for draughts. But they are still expensive in Poland.

Krzysztof Grzelak
Posts: 1054
Joined: Thu Jun 20, 2013 17:16
Real name: Krzysztof Grzelak

Re: The Unofficial World Championships Of Computer Programs In International Draughts Blitz 2020

Post by Krzysztof Grzelak » Tue Sep 15, 2020 15:32

The match is over - there is no revival - there is no need to publish it.

Sidiki
Posts: 209
Joined: Thu Jan 15, 2015 16:28
Real name: Coulibaly Sidiki

Re: The Unofficial World Championships Of Computer Programs In International Draughts Blitz 2020

Post by Sidiki » Fri Sep 18, 2020 05:27

GWD it's available, but i don,t know if the link exist again.

This's a screenshot

http://www.mediafire.com/view/qhu5rglbh ... D.JPG/file

Krzysztof Grzelak
Posts: 1054
Joined: Thu Jun 20, 2013 17:16
Real name: Krzysztof Grzelak

Re: The Unofficial World Championships Of Computer Programs In International Draughts Blitz 2020

Post by Krzysztof Grzelak » Fri Sep 18, 2020 11:17

Sidiki wrote:
Fri Sep 18, 2020 05:27
GWD it's available, but i don,t know if the link exist again.

This's a screenshot

http://www.mediafire.com/view/qhu5rglbh ... D.JPG/file
Then please share the program GWD. Thank you very much Sidiki.

Sidiki
Posts: 209
Joined: Thu Jan 15, 2015 16:28
Real name: Coulibaly Sidiki

Re: The Unofficial World Championships Of Computer Programs In International Draughts Blitz 2020

Post by Sidiki » Fri Sep 18, 2020 14:13

Hi Krzysztof,
I sent you a private message.
Before today, i don't realized how GWD was settable with pattern ie, you can set it to play a particular kind of game.
Rated 2200 Elo point.

Krzysztof Grzelak
Posts: 1054
Joined: Thu Jun 20, 2013 17:16
Real name: Krzysztof Grzelak

Re: The Unofficial World Championships Of Computer Programs In International Draughts Blitz 2020

Post by Krzysztof Grzelak » Sun May 30, 2021 17:03

Jelle Wiersma wrote:
Mon Sep 14, 2020 22:32
Hi Krzysztof, may be you can also ask Nicolas Guibert for the program Buggy, and Gijsbert Wiesendekker / Klaas Bor for the program GWD. Both strong programs. And thanks for organizing and running the tournament 13 days in a row!

Jelle
As for the Buggy program, I wrote with the author Nicolas Guibert. Unfortunately, the copy of the program has not survived anywhere. Great pity. The author of the GWD program wrote that he would make the program available for the tournament. As for the program Damy - I wrote to the author of the program. So far, the author of the program has not responded to my request.

Krzysztof Grzelak
Posts: 1054
Joined: Thu Jun 20, 2013 17:16
Real name: Krzysztof Grzelak

Re: The Unofficial World Championships Of Computer Programs In International Draughts Blitz 2020

Post by Krzysztof Grzelak » Thu Jun 03, 2021 15:12

He would like to get one more program. A program called Dios - author of the program Chris Jurriens. Once I asked the author of the program to share - so far none of this has worked out - a great pity.

Joost Buijs
Posts: 355
Joined: Wed May 04, 2016 11:45
Real name: Joost Buijs

Re: The Unofficial World Championships Of Computer Programs In International Draughts Blitz 2020

Post by Joost Buijs » Fri Jun 04, 2021 08:11

Maybe I'm wrong, but wasn't the main author of DIOS Erik van Riet Paap?

B.T.W. my engine can only run in DXP server mode, will this suffice to have it run in your tournament? Or do you need something else like a simple text based UI?

Ed Gilbert
Posts: 829
Joined: Sat Apr 28, 2007 14:53
Real name: Ed Gilbert
Location: Morristown, NJ USA
Contact:

Re: The Unofficial World Championships Of Computer Programs In International Draughts Blitz 2020

Post by Ed Gilbert » Fri Jun 04, 2021 14:10

B.T.W. my engine can only run in DXP server mode, will this suffice to have it run in your tournament? Or do you need something else like a simple text based UI?
It might work with Dam2.2 as a UI. I think Moby Dam works that way, using DXP. https://hjetten.home.xs4all.nl/mobydam/mobydam.html

-- Ed

Joost Buijs
Posts: 355
Joined: Wed May 04, 2016 11:45
Real name: Joost Buijs

Re: The Unofficial World Championships Of Computer Programs In International Draughts Blitz 2020

Post by Joost Buijs » Fri Jun 04, 2021 16:53

Ed Gilbert wrote:
Fri Jun 04, 2021 14:10
B.T.W. my engine can only run in DXP server mode, will this suffice to have it run in your tournament? Or do you need something else like a simple text based UI?
It might work with Dam2.2 as a UI. I think Moby Dam works that way, using DXP. https://hjetten.home.xs4all.nl/mobydam/mobydam.html

-- Ed
Ed, Thanks for the info!

Dam 2.2 seems to be freeware, I will check it out. I never tried the engine with anything but Kingsrow. There are still some things I want to add, like better time-control and pondering, maybe something like a book. I'd rather spend my time on these instead of adding another protocol.

Ed Gilbert
Posts: 829
Joined: Sat Apr 28, 2007 14:53
Real name: Ed Gilbert
Location: Morristown, NJ USA
Contact:

Re: The Unofficial World Championships Of Computer Programs In International Draughts Blitz 2020

Post by Ed Gilbert » Sat Jun 05, 2021 00:11

My tests with a rather large opening book indicate that it adds almost nothing to the engine strength in 10x10 draughts. See http://laatste.info/bb3/viewtopic.php?f ... on#p128916. Pondering is probably more useful, but I always found it tricky to implement correctly.

-- Ed

Joost Buijs
Posts: 355
Joined: Wed May 04, 2016 11:45
Real name: Joost Buijs

Re: The Unofficial World Championships Of Computer Programs In International Draughts Blitz 2020

Post by Joost Buijs » Sat Jun 05, 2021 07:36

Since the NN remains slow the idea is to gain some time by adding a book, and not to have it play stronger moves from book. At the level of Kingsrow gaining time does not seem to give an advantage anymore, maybe it could help with bullet games or when running on very slow hardware.

For my chess engine I use a book limited at 70 ply, it happens often that the engine plays 35 full moves from book which (in theory) gives it a 2 to 1 time advance (unfortunately the opponent usually does the same). Many deep lines that occur in chess are pretty well known and are examined to death, for 10x10 draughts this doesn't seem to be the case.

Sidiki
Posts: 209
Joined: Thu Jan 15, 2015 16:28
Real name: Coulibaly Sidiki

Re: The Unofficial World Championships Of Computer Programs In International Draughts Blitz 2020

Post by Sidiki » Sat Jun 05, 2021 13:00

Joost Buijs wrote:
Sat Jun 05, 2021 07:36
Since the NN remains slow the idea is to gain some time by adding a book, not to have it play stronger moves from book. At the level of Kingsrow gaining time does not seem to give an advantage anymore, maybe it could help with bullet games or when running on very slow hardware.

For my chess engine I use a book limited at 70 ply, it happens often that the engine plays 35 full moves from book which (in theory) gives it a 2 to 1 time advance (unfortunately the opponent usually does the same). Many deep lines that occur in chess are pretty well known and are examined to death, for 10x10 draughts this doesn't seem to be the case.
Great,

this put us with the opening solution.

Kingsrow and Scan, without forget the others are, without book, very very strong, due to they evaluation. With the book they have a time advantage.

This let me say that, the opening book must be an entire part of any respectable engine.
If you look carefully even the strongest chess engine that use NNUE or training file haven't the same strenght with/without these training file.

In the same time, opening book help to gain time. I think that opening book must be trained by the engine to just anticipe the best move that it will have to do facing the best possible move of his opponent based on his own evaluation and a certain preferred move.
For Kingsrow the best moves are 1.34-30 or 1.31-26 and for Scan 1.34-30, Dragon 1.32-28 followed by 34-29.

Friendly, Sidiki.

Post Reply