EndGame Databases

[Rein Halbersma]

Rein, i don't make use of the symmetry concept so far.
Based on profiling I found out that the index function is consuming quite some time during the database generation, so i try to keep it as simple as possible.

Hi Bert, how did you implement your index function? Did you copy it from the 1994 paper of the Chinook team "Solving retrograde analysis ...."? In particular, do you incrementally update piece configurations to calculate the index of a board configuration?

Rein

[ildjarn]

There are some errors in the maximum pages. I have the full databases at work, will generate the maximum positions on tuesday (I have a day off on monday). From what I remember, the maximum length is OK, but the positions given aren't.

Here are the 3x3 stats, all other stats are correct on Michel's page.

http://sanguis.xs4all.nl/dammen/endgame/

[TAILLE]

There are some errors in the maximum pages. I have the full databases at work, will generate the maximum positions on tuesday (I have a day off on monday). From what I remember, the maximum length is OK, but the positions given aren't.

Here are the 3x3 stats, all other stats are correct on Michel's page.

http://sanguis.xs4all.nl/dammen/endgame/

In the maximum pages you can see the following position

White to move
My view is that this position is not a winning position. Do you agree ?
Gérard

[steenslag]

There are some errors in the maximum pages. I have the full databases at work, will generate the maximum positions on tuesday (I have a day off on monday). From what I remember, the maximum length is OK, but the positions given aren't.

Here are the 3x3 stats, all other stats are correct on Michel's page.

http://sanguis.xs4all.nl/dammen/endgame/

Link does not work, not even with "www." inserted. Domainname does resolve to an IP-adress; just can't find anything on it.

[BertTuyt]

Rein, i use the index function which was developed by Michael Grimminck.
Because I use the databases distributed by Harm Jetten.
You can dig into the source when you download the file from the Dragon site.
I don't know on what work this index function was based.

Bert

[ildjarn]

http://sanguis.xs4all.nl/dammen/endgame/

Link does not work, not even with "www." inserted. Domainname does resolve to an IP-adress; just can't find anything on it.

My modem's having problems. I'll fix it tonight.

EDIT: moved the stats to another webserver: http://www.xs4all.nl/~sanguis/endgame/

[ildjarn]

In the maximum pages you can see the following position

White to move
My view is that this position is not a winning position. Do you agree ?
Gérard

Can you indicate where black could reach a draw? I can check Michel's database to see what it says.

[TAILLE]

In the maximum pages you can see the following position

White to move
My view is that this position is not a winning position. Do you agree ?
Gérard

Can you indicate where black could reach a draw? I can check Michel's database to see what it says.

1.13-36 35-40 2.33-28 40-45 3.26-17 04-15 4.17-06 15-29 5.41-47 29-07 6.47-38 07-12 7.38-20 12-08 8.20-03 08-35 9.03-21 35-49 10.21-08 49-35 11.08-02 35-40 12.02-30 40-07 13.30-43 07-12 14.43-25 12-21 15.25-03 21-16 16.36-13 16-07 17.13-02 07-18 18.02-16 18-01 19.03-17 01-29 20.16-02 29-18 21.02-24 18-36 22.24-47 36-18 23.17-21
and now not 23...18-07?? as indicated but
18-09!! 24.21-26 09-18 25.47-33 18-04 26.33-50 04-18 27.26-21 18-04
and black has managed to impose, just in time, 25 moves without any conversion !!!
Surely a very remarkable draw.
Gérard

[ildjarn]

In my opinion, for a theoretical evaluation of a position, the 25 move rule shouldn't be used.

[TAILLE]

In my opinion, for a theoretical evaluation of a position, the 25 move rule shouldn't be used.

Hi Ildjarn,

I see what you mean by theoretical evaluation but can you accept to build a db giving you a “theoretical” result instead of the “real” one, taking into account all the existing rules of the game?

Of course you can imagine changing the game rules but it is another question and I am not competent for that.

The question is then the following :
If you want to buy an endgame db and you have the choice between a “theoretical” db and a “real” db which one will you buy?

The proposed position is of course very interesting because it proves that you cannot deduce easily the “real” db from the “theoretical” db !

Gérard

[ildjarn]

In my opinion, for a theoretical evaluation of a position, the 25 move rule shouldn't be used.

Hi Ildjarn,

I see what you mean by theoretical evaluation but can you accept to build a db giving you a “theoretical” result instead of the “real” one, taking into account all the existing rules of the game?

Of course you can imagine changing the game rules but it is another question and I am not competent for that.

The question is then the following :
If you want to buy an endgame db and you have the choice between a “theoretical” db and a “real” db which one will you buy?

The proposed position is of course very interesting because it proves that you cannot deduce easily the “real” db from the “theoretical” db !

Gérard

If you want to build 'real' databases, you shouldn't just build databases with the 25 move rule in your mind in the forward play, but also in the historical play. A position in which the 10 previous moves were king moves must be solved differently than the exact same position in which the previous 15 moves were king moves. In other words, a 'real' database is i.m.o. impossible to build.

[Ed Gilbert]

If you want to build 'real' databases, you shouldn't just build databases with the 25 move rule in your mind in the forward play, but also in the historical play. A position in which the 10 previous moves were king moves must be solved differently than the exact same position in which the previous 15 moves were king moves. In other words, a 'real' database is i.m.o. impossible to build.

It's true that you cannot assign draw values to positions in your db with DTC less than 50 because their WLD value can depend on the path you took to get to them. But any position with a DTC value greater than 50 could be assigned as a draw. The combination of the database and your engine could then correctly navigate towards a drawn solution, even in positions where your db might say its a loss. I did not do anything about this 50 ply rule in my db, and I do not think it is worth troubling over. The number of DTC > 50 positions in the 7-piece db is small, and the probability of playing into one of these in a game is pretty close to 0.

-- Ed

[ildjarn]

The number of DTC > 50 positions in the 7-piece db is small, and the probability of playing into one of these in a game is pretty close to 0.

And the probability that players actually play these kind of endgames perfectly is even smaller.

For computer-tourneys the 25-move rule would be useless anyway, because all programs use databases nowadays (or is that forbidden during a computer tourney? Not likely but you never know...)

[TAILLE]

If you want to build 'real' databases, you shouldn't just build databases with the 25 move rule in your mind in the forward play, but also in the historical play. A position in which the 10 previous moves were king moves must be solved differently than the exact same position in which the previous 15 moves were king moves. In other words, a 'real' database is i.m.o. impossible to build.

It's true that you cannot assign draw values to positions in your db with DTC less than 50 because their WLD value can depend on the path you took to get to them. But any position with a DTC value greater than 50 could be assigned as a draw. The combination of the database and your engine could then correctly navigate towards a drawn solution, even in positions where your db might say its a loss. I did not do anything about this 50 ply rule in my db, and I do not think it is worth troubling over. The number of DTC > 50 positions in the 7-piece db is small, and the probability of playing into one of these in a game is pretty close to 0.

-- Ed

Hi Ed, Ildjarn
I don't see your point. Why a real db is impossible to build?

My view is that it is very easy to take into account the 25 moves rule and, in addition it is faster for the generation process itself ! So why introduce some errors in the db saying that the concerned positions remains exeptionnal ?

If you have just made 15 moves without any conversion you are in fact in the db for 15 moves.
The point is the following : the moment you enter the db you use in fact only the "real WLD db" then, after a non conversion move, you have to take into account the "DTC db" with witch you can take into account the historical play.

Starting with the following position

After 1.13-36 the real WLD db tells us that 1…35-40! is the only move to obtain the draw then after 2.33-28 the real WLD db tells us that 40-45! is again the only move to obtain the draw. After 3.26-17 we reach a key point. The WLD real db tells us that the position is lost for black (witch looks like a contradiction) but, due to the fact that the previous move was not a conversion move, we have here to use the DTC db. This DTC db tells us that it remains 49 plies before the next conversion and the main engine can then conclude that black can obtain the draw by playing 04-15!!
After 4.17-06 15-29 5.41-47 29-07 6.47-38 07-12 7.38-20 12-08 8.20-03 08-35 9.03-21 35-49 10.21-08 49-35 11.08-02 35-40 12.02-30 40-07 13.30-43 07-12 14.43-25 12-21 15.25-03 21-16 16.36-13 16-07 17.13-02 07-18 18.02-16 18-01 19.03-17 01-29 20.16-02 29-18 21.02-24 18-36 22.24-47 36-18 23.17-21
though the WLD db continues to tell us that the position is lost for black, the DTC db tells us that it remains 9 plies before the next conversion and that means that the position is a draw due to the existence of the 41 last moves without any conversion. 23…18-09!! is the only move to obtain this draw.

Do you think there is something wrong here ? With a good use of “real” WLD db and DTC db it seems to me that it is possible to build a “real” db.
Gérard

[TAILLE]

The number of DTC > 50 positions in the 7-piece db is small, and the probability of playing into one of these in a game is pretty close to 0.

And the probability that players actually play these kind of endgames perfectly is even smaller.

For computer-tourneys the 25-move rule would be useless anyway, because all programs use databases nowadays (or is that forbidden during a computer tourney? Not likely but you never know...)

Just a theretical question for computer tournament. Let's suppose that the position

white to move
is reach at the exact moment we have to stop the game and to ask the referee for the result (typically 75 moves have been played).
What will be the decision of the referee seeing that different db give different results ? I do not know, but I guess he will let the game continue waiting for a clear win or an application of the 25 move rule ?
What will you do yourself as referee ?
Gérard