What to play

[FeikeBoomstra]

From DNC 2008: Schwartzman - Heusden


Black on move: Ron played 14-20 and eventually lost the game. BoomstraDam suggests 13-18 with a very small disadvantage. Other suggestions?

[wellnesswrotter]

B:W24,29,35,36,37,38,48:B4,13,14,15,16,17,26,27

Schwarzman - Heusdens, DNC Open NK 2008.

[Ed Gilbert]

Kingsrow thinks both 13-18 and 14-20 are strong for black. After this short sequence of moves:

[FEN "B:W24,29,35,36,37,38,48:B4,13,14,15,16,17,26,27."]
13-18 2. 35-30 17-21 3. 37-32

now kingsrow sees a database win. If black lost the game with 14-20 it must have been from a later mistake.

-- Ed

[FeikeBoomstra]

Interesting, this is the entire game, where did it go wrong for Ron?

http://www.damweb.nl/tho/index.php?page=showgame&p=1-8

[TAILLE]

Hi,
A white man is missing on 28 square
Gérard

[FeikeBoomstra]

Sorry, you are right, I editted the position, but I did something wrong.

[Ed Gilbert]

In this position:

B:W24,28,29,35,36,37,38,48:B4,13,14,15,16,17,26,27.

Kingsrow plays 4-9 and sees database draw very quickly. If I force it to play 13-18 it eventually sees database draw after about 5 minutes of searching.

-- Ed

[TAILLE]

Hi,

Damy chooses also 04-09 but without demonstrating the draw as Kingsrow did.
Just a question Ed. Does the demonstration use 8-9-10 pieces db or do you used only 7 pieces db ?

Gérard

[Ed Gilbert]

Gerard,

The previous result was with the 9pc db. I tried it again using only 7 pieces (and fewer) and it sees database draw in about 30 seconds, at depth 21. I let it continue to depth 25 and at each iteration it continued to show db draw.

-- Ed

[TAILLE]

Hi Ed.

Gerard,

The previous result was with the 9pc db. I tried it again using only 7 pieces (and fewer) and it sees database draw in about 30 seconds, at depth 21. I let it continue to depth 25 and at each iteration it continued to show db draw.

-- Ed

I would have been happy to compare Damy and Kingrow on such problem but it appears very difficult because Damy and Kingsrow do not use the same algorithm (MTD(f) vs MTD(f)-best).

The problem here is the following. Trying the test value -0,300 Damy find that 04-09 is better than -0,300 and Damy finds in the other hand that 13-18 is less than -0,300. Because it is too hard to see that 13-18 is in fact better than -0,300 Damy does not try to obtain a better value.
It does not harm in a real game where the deal is only to find the better move but it is completely inefficient for giving the result of a position.
By the way is Kingsrow able to demonstrate the draw after 13-18 with only the 7 pieces endgame db ?

Gérard

[Ed Gilbert]

Gerard,

I tried 13-18 with the 7pc db and it actually found db draw faster, in 2 minutes. I'm not sure what this really means. Remember that a db draw score, as you well know, does not mean that the position is guaranteed to be a draw (as it does with a db win or db loss). The search score is a mixture of db scores and heuristic scores propagated from the leaf nodes, and so a db draw score means that anyplace a db draw was compared to a heuristic score the heuristic score was inferior and the db draw was propagated. I did notice that the search speed in nodes/sec was a lot higher using only 7 pieces here. That may be true in general, I just don't do these comparisons that often to remember. Sometimes the larger dbs help kingsrow win a game or salvage a draw in a weak position, and other times they are a detriment because they slow it down.

I understand what you mean about mtd-best and not being able to obtain an exact score. Here's a suggestion: why don't you add some control to your user interface that lets you select either mtd or mtd-best as the search algorithm? The difference is only a relatively small amount of code at the root level of the search. I think you could conditionally use either algorithm without too much difficulty. Then you could turn on the ability to get exact scores when you are doing position analysis where that info is useful, and keep the mtd-best for tournament games.

-- Ed

[wellnesswrotter]

Question about the FEN notation.
You keep putting a point after the FENstring.
But is that protocol? Not according to the Wikipedia page.

[Ed Gilbert]

Question about the FEN notation.
You keep putting a point after the FENstring.
But is that protocol? Not according to the Wikipedia page.

Is there a documented FEN standard for draughts? I have not seen one. I use the same format as Martin Fierz's CheckerBoard program because that is an informal standard due to its popularity.

-- Ed

[wellnesswrotter]

According to

http://en.wikipedia.org/wiki/Portable_Draughts_Notation

The FEN is:
the initial position of the checkers board. This is used to record partial games (starting at some initial position). It is also necessary for some draughts variants where the initial position is not always the same as traditional chess. If a FEN tag is used, a separate tag pair "SetUp" must also appear and be have its value set to "1".
A position can be stored by the FEN tag:

[SetUp "1"]
[FEN "[Turn]:[Color 1][[K][Square number][,]...]:[Color 2][[K][Square number][,]...]"]

Examples:

[FEN "B:W18,24,27,28,K10,K15:B12,16,20,K22,K25,K29"]
[FEN "B:W18,19,21,23,24,26,29,30,31,32:B1,2,3,4,6,7,9,10,11,12"]


But in the PDN examples later on the page they do use a dot at the end of te the string...

Standards!

[TAILLE]

Remember that a db draw score, as you well know, does not mean that the position is guaranteed to be a draw (as it does with a db win or db loss). The search score is a mixture of db scores and heuristic scores propagated from the leaf nodes, and so a db draw score means that anyplace a db draw was compared to a heuristic score the heuristic score was inferior and the db draw was propagated

I understand now what you mean by "db draw".
I do not use this "db draw" notion in Damy bur I use what I could call a "clear demonstrated draw" which imply no kind of doubt on the result.
For Damy the "clear demonstrated draw" information is stored in the hashtable in order to be able to stop the tree exploration when the position is reached.
For Damy a "db draw" is simply a position evaluated to 0,000;

Can you clarify the usefulness of this "db draw" notion for Kingsrow ?

I understand what you mean about mtd-best and not being able to obtain an exact score. Here's a suggestion: why don't you add some control to your user interface that lets you select either mtd or mtd-best as the search algorithm? The difference is only a relatively small amount of code at the root level of the search. I think you could conditionally use either algorithm without too much difficulty. Then you could turn on the ability to get exact scores when you are doing position analysis where that info is useful, and keep the mtd-best for tournament games.

Yes Ed, your are right it is not difficult. My intention was to write a special algorithm in order to search the real result of a position : win, loss or "clear demonstrated draw". The program run until the correct result is found or when the operator decide to stop manually the exploration.
Gérard