Learning option

[Catherine]

Hi all Programmers,

It's seem that add learning option to program is more used now, Scan, Dragon, Plus500, Aurora Borealis use it!!!!
Is it like a database? I mean for example, Dragon play against Scan a thousand games and learn them, the target is it to not do the same errors or to load in a file the winning way used by Scan ?
Or , is it a kind of A.I?
I read on http://www.ffjd.fr/forum/sujet.php?id_sujet=60545 that Scan played several games against itself with the learning option to optimize his strenght.
Another question for you; what it is the file for learning option of Scan. Can you confirm that playing several games against itself or another program permit to Scan to learn?
Thanks, Catherine.

[Birdy]

That's a very good question.

Catherine, please check your P.M.

[Fabien Letouzey]

Is it like a database? I mean for example, Dragon play against Scan a thousand games and learn them, the target is it to not do the same errors or to load in a file the winning way used by Scan ?
Or , is it a kind of A.I?
I read on http://www.ffjd.fr/forum/sujet.php?id_sujet=60545 that Scan played several games against itself with the learning option to optimize his strenght.

It's more than just a database, as it extracts positional knowledge from the games.

Another question for you; what it is the file for learning option of Scan.

There is no such option because the learning is "offline" (also in Dragon). This means that the learning only takes place once under the supervision of the programmer; this is sometimes called the "training" phase. Then the program uses the extracted knowledge without learning anymore.

What you are suggesting, online learning = adapting the knowledge after every game, also exists but is not used in current draughts programs. Both ways have pros and cons.

Can you confirm that playing several games against itself or another program permit to Scan to learn?

Yes but offline. This is probably not what you expected.

Fabien.