A difficult endgame in the game Coulibali - Ndjofang

[TAILLE]

Hi,

The following position was reach during the recent African championship, in the game Coulibali - Ndjofang:

diag.jpg (170.33 KiB) Viewed 9148 times

White to move

Is your program able to find here the white drawing move ?

Note : I am not quite happy with Damy because it needs 28 minutes (!?) to find this drawing move. What are the performance of your program?

[Krzysztof Grzelak]

The program Kingsrow rated position after 14 minutes as a movement 22-18 a draw at depth 30. The test was performed in the following equipment:

processor - I7 2670QM
ram - 16 GB DDR3 1333
hard drive - Samsung ssd 1 TB

hashtable size - 512 mb
cache size - 14500 mb
max pieces - 8

[Ed Gilbert]

I get a much faster result than Krzysztof's. Perhaps you were using a small hashtable setting?

Draw 22-18 in 25 seconds at depth 31.

CPU ES-2630 v3 @ 2.4 GHz
hashtable: 512mb
search threads: 5
egdb cache: 16gb

[Krzysztof Grzelak]

In fact, I had set the option Search threads on 1. I forgot to switch on 8. Ed too good computer you have.

[TAILLE]

Thank you for your answers. I try to build a new search algorithm but I have still a lot of work to do!

[BertTuyt]

Gerard, at least welcome back in this forum.
Would like to learn a lot about your new Damy developments and your new insights...

Bert

[TAILLE]

Gerard, at least welcome back in this forum.
Would like to learn a lot about your new Damy developments and your new insights...

Bert

HI,

A lot of work has still to be done but I am already happy with the first results. My new algorithm is based on several new ideas:
1) Instead of developping a tree I develop a network in order to detect easily the loops
2) With the availability of large memory, instead of keeping results in a hashtable, I keep in memory millions of positions link together in a large network
3) Instead of managing bound values associated to alpha-beta procedure I manage only a real evaluation of each position which is improved each time the position is visited by the exploration algorithm

BTW I do not see on the forum how to build a diagram? can somebody help me?

[Krzysztof Grzelak]

To present the diagram on the forum, you must enter the site http://fmjd.org/dias2/index.php. Set the appropriate position click "send " and "Save On Server" and later paste the link to the diagram on the forum. Give an example.




Only I'm not sure about that matter.

Krzysztof

[TAILLE]

Thank you Krzysztof for your help!


White to play

With the relevant 8 pieces in memory it takes less than 2 minutes to Damy to prove the loss. I am wondering if it is a good start for my new algorithm?
Any reaction to help me?

[Ed Gilbert]

Hi Gerard,

Kingsrow cannot see a db win after 1/2 hour of search, only a nearly 2-man advantage for black. What is your PV?

-- Ed

[TAILLE]

Hi Ed,


White to play

My PV is the following
1.35-30 24x35 2.28-22 23-28 3.32x12 21x41 4.12-7 19-23 5.7-1 41-46 6.1x29 46-28 7.38-32 28x6 8.29-1 6-17 9.32-27 17-3 10.1-7 3-21
11.27-22 26-31 12.7-23 21-8 13.23-7 13-19 14.33-28 31-36 15.7-23 8-3 16.23x5 3-12 17.22-17 12x26 18.34-29 26-48 19.39-33 25-30 20.29-23 48-42 and the db is reached

[Ed Gilbert]

Gerard, that is a very good result. I would say your new program is working very well.

The position looks somewhat like the Woldouby position, W:W25,27,28,30,32,33,34,35,37,38:B12,13,14,16,18,19,21,23,24,26. Does Damy now see a draw there?

-- Ed

[TAILLE]

Gerard, that is a very good result. I would say your new program is working very well.

The position looks somewhat like the Woldouby position, W:W25,27,28,30,32,33,34,35,37,38:B12,13,14,16,18,19,21,23,24,26. Does Damy now see a draw there?

-- Ed

Yes Ed, the Woldouby position is quite easy for Damy. With the relevant bdd already in the cache, Damy needs 2 seconds to prove the draw.

[BertTuyt]

Gerard, thats amazing...
Do you use a 7P or 8P DB?

Bert

[TAILLE]

Gerard, thats amazing...
Do you use a 7P or 8P DB?

Bert

Hi Bert,
I use of course the 8p db. Anyway with my previous algorithm I was not able to resolve completly the positions given above so it was a very good news for me to see such inexpected results.