Search Algorithm

[Rein Halbersma]

Even I use the NULL-move only in specific situations you cannot say that the NULL-move concept is not use in Draughts can you?

Gerard, thanks for your reply.
I was not aware that you used the NULL-move.
So as you do (even in specific situations) use it, than my statement is not true obviously.
But so far I dont have information if also the others use this concept in general or as in your case in specific positions.

So please let me know...

Bert

Bert, I have written something about null move on this forum in the past, based on my own experimentations. See this thread: viewtopic.php?f=53&t=2349&start=21

I haven't any clear-clut conclusions yet. But so far the "double null move" technique by chess programmer Vincent Diepeveen (who also wrote a draughts program in the 90s) seems completely safe because it will automatically take care of zugzwang. It might not be as efficient as other draughts-specific reductions though.

Rein

[BertTuyt]

Rein, thanks for sharing. Maybe I should re-study the Stockfish code.
Im not sure if in the Stockfish search all possible reductions are implemented such as, NULL-Move, MCP and LMR.
At least Damage uses MCP and LMR ( +FHR), although I have to admit that sometimes an obvious move is not recognized. As the gain in ply is very good, in general the overall performance is better.
So far I dont have a clue (and as such also no solution) in which specific cases MCP and/or LMR should be omitted.

As also Ed pointed out, the trick is better move-ordering (and better recognition whether a node is CUT/ALL, I guess) ...
My problem is that I so far (next to TT-Move) , very much rely on the History-Table (think also Ed is doing so). And I get more and more signals that the History Table at sufficient depth just produces "random noise". Basically move-ordering should be based on evaluation principles, but that is easier said than done

Bert

[Alvaro]

Hi,
I had null move in my program Profound (spanish checkers) for some time.
But I was getting draws all the time against Triturador.
Then at some point I removed it, and implemented some forward prunning of my own.
Then Profound started to play better and winning.
My implementation of null move had verification search, so I guess any possible benefits where a little offset by the verification search time.
One thing I want to test is to allow 2 nulls in a row, also I suspect we can get away by allowing any number of nulls in a row.
But only tests can shed some light on it.

best regards,
Alvaro

[BertTuyt]

I'm testing a few small modifications in the conditions for LMR.
See below 79-games (small) match against Kingsrow.

It started with 66 Draws, than 1 Win for Kingsrow, and the other games also Draw.

Bert

[BertTuyt]

A recommendation for everyone with a large Endgame Database, buy a Solid State Drive (SSD).
In Holland now available the Samsung 840 EVO series 1 TeraByte, listed for 549 euro....
And if you wait they will certainly become cheaper..

See below link:

http://www.anandtech.com/show/7173/sams ... els-tested

Bert

[BertTuyt]

Interesting article: http://uk.hardware.info/reviews/4790/in ... or-servers

New Intel Xeon Ivy-Bridge for servers with 12-cores..

I have already seen announcements from Dell and HP launching new products based on this processor.
Top of the line a dual processor workstation (so 24 cores) with 512 GByte Memory.

But I'm afraid this product wont survive the local budget committee.

Bert

[Walter]

Interesting article: http://uk.hardware.info/reviews/4790/in ... or-servers

New Intel Xeon Ivy-Bridge for servers with 12-cores..

I have already seen announcements from Dell and HP launching new products based on this processor.
Top of the line a dual processor workstation (so 24 cores) with 512 GByte Memory.

But I'm afraid this product wont survive the local budget committee.

Bert

Bert,

We are free to dream here without budget committees

I would be interested to know what nice project you would actually embark on if you had such a machine.

By the way, forget SSD. I want a skyEagle system for my 12 piece egdb (http://www.skyera.com/products/skyeagle/overview/)!

Regards,
Walter

[BertTuyt]

So it seems quiet (on the Western front).
In the mean time I have added now DataBase functionality to the Damage GUI.
Although work in progress Im now able to read and process TurboDambase (.cgd) files.
See below Ribbon.

Next to that I wanted to extend the standard search capabilities as provided by TurboDambase.
For this reason Im now developing a Draughts Query Language which I named DQL, and which I based on the previous work of the Chess Query Language (CQL).
With the Search Dialog it is now possible to write (open/save) DQL scripts to "data mine" the Game Database.

I'm now working to finalize the initial DQL-interpreter, and next to that Im writing a DQL document which I want to share will all, to verify if there is a need for additional language elements and to get further/more suggestions..

Bert

[BertTuyt]

Herewith the DB_Info Dialog after opening the Mega2005.cgd file....

Bert

[BertTuyt]

To get familiar with CQL I extracted some info from several webpages and included all in a separate .doc.
For those people interested attached the CQL.doc file.
As stated before I will write a (draft) DQL.doc document.

Bert

[Piet Bouma]

So it seems quiet (on the Western front).
In the mean time I have added now DataBase functionality to the Damage GUI.
Although work in progress Im now able to read and process TurboDambase (.cgd) files.
See below Ribbon.

Next to that I wanted to extend the standard search capabilities as provided by TurboDambase.
For this reason Im now developing a Draughts Query Language which I named DQL, and which I based on the previous work of the Chess Query Language (CQL).
With the Search Dialog it is now possible to write (open/save) DQL scripts to "data mine" the Game Database.

I'm now working to finalize the initial DQL-interpreter, and next to that Im writing a DQL document which I want to share will all, to verify if there is a need for additional language elements and to get further/more suggestions..

Bert

Nice developments Bert.
In Toernooibase I can "deliver" .pdn files.
Is there a possibility to "data mine" a .pdn file of almost 125.000 games of Toernooibase?
(First I must see if I can make this large file....)
And what about "on-line" data mine such a large file with some webservices?
Would that be possible?

[BertTuyt]

Nice developments Bert.
In Toernooibase I can "deliver" .pdn files.
Is there a possibility to "data mine" a .pdn file of almost 125.000 games of Toernooibase?
(First I must see if I can make this large file....)
And what about "on-line" data mine such a large file with some webservices?
Would that be possible?

Piet, it is not a huge effort to modify some routines in such a way that also .pdn files can be processed (only think analyzing a .pdn will take more time compared with the condensed/compressed TurboDambase file).
If you could create such a large file, I can certainly test it.
As I have a dropbox, I can share a directory with you where we could locate the .pdn (as email wont work).

If you are interested I can also share with you (and others) the new GUI with the Database options for further testing.
Where I definitely need support/help is in the data mine options you want to have included.
As already stated in a previous post, I think there is a need for far more complex search queries than the options supported by TurboDambase (although i definitely very much like and appreciate TurboDambase).

To offer this via web services, in alter stage, is also something which (to my opinion) can be realized.

Bert

[Krzysztof Grzelak]

Bert and which year you are using bases TurboDambase in the program Damage. I still have so request to you and you can make the program available for testing.

Krzysztof Grzelak.

[BertTuyt]

The base DQL framework seems to work now.
Still with a limited language base, but not so difficult to expand.
The code was not yet optimized for speed (and certainly not error and fool proof), so TurboDambase is still faster.
In the given example it took DQL 4.5 second to find all (23513) "match games".

See below the TurboDambase mask for a specific outpost on 24 pattern
( black man 13 , 15, 20, 25 ; empty 19 , 30, 35 ; white man 24 , 29)
and also the DQL code.

So far I use M (capital ) for white man, m for black man en e for empty square (but that is easy to change...
Both TurboDambase and Damage DQL find in the Mega2005.cgd file with 316794 games 23513 match games.

Bert

[Rein Halbersma]

Hi Bert,

Very ambitious project to make a draughts equivalent to CQL! Are you planning on supporting the :shift: keyword of CQL? That is something that is not possible in TurboDambase and would really make a difference.

Rein