Damy has the 7 pieces draw+ database

[TAILLE]

Hi,

FYI Damy have just finished generating the 7 pieces draw+ database. This database needs 21,7Gb of storage and Damy took one week for the generation.
Of course I still need some additionnal work in order to take into account this database in the search algorithm.

An interesting position of the 4 pieces database is the following:


Draw+ : 16 plies before conversion

[BertTuyt]

Gerard, 1 week for a 7p DB is very fast. Did you built the draw+ DB from the scratch, or did you (partial) use the existing 7p DB.
In my case it took me months to built the 7p DB (but this was on 1 core only). But even with multiple cores I guess i needed at least 1 month.

Bert

[TAILLE]

Hi Bert,

Gerard, 1 week for a 7p DB is very fast. Did you built the draw+ DB from the scratch, or did you (partial) use the existing 7p DB.
In my case it took me months to built the 7p DB (but this was on 1 core only). But even with multiple cores I guess i needed at least 1 month.

Bert

I am able to use two computers: one with 12 cores and 24Gb RAM and the other with 8 cores and 12Gb RAM.
With only the first machine (12 cores and 24Gb RAM) I estimate I need between 9 and 10 days to generate the 7 pieces draw+ database.
(note: it is only the time necessary for the generating process; I do not include here the additionnal time taken by the verification process)
Of course I use the existing 7 pieces WLD database in order to eliminate all winning and losing positions.

[Alvaro]

Hi everybody
could you please explain what info does the "draw+ database" has?
Also how many bits are used for each position ?
Is it a single database or two separate ones?
Also what are the advantages over the WDL database?

best regards,
Alvaro

[TAILLE]

Hi everybody
could you please explain what info does the "draw+ database" has?
Also how many bits are used for each position ?
Is it a single database or two separate ones?
Also what are the advantages over the WDL database?

best regards,
Alvaro

Hi Alvaro,

The draw+ notion has been used for sevral years during the worldchampionship. With this notion, the resut of a player for a given game is one of the five following ones: lost, draw-, draw, draw+ or win. The goal of the draw+ database is only to be able to distinguish the draw-, draw and draw+ results.

what are the advantages over the WDL database?: if a competition does not use the draw+ notion then this database is unuseful otherwise this database allows to get the best "draw" result.

The question Also how many bits are used for each position ? is not clear. I can only say that it exists 5 five possible results for a given position.

The question Is it a single database or two separate ones? is very implementation dependant. In Damy implementation a database is always a "binary" database. In Damy implemenation the so called "WLD" database is made of two "binary" databases : the win/no win db and the lost/no lost db. To take into account the draw+ notion I only add a daw+/no draw+ db and a draw-/non draw- db.