New RomiChess released (WBEC site)

[Tony Thomas]

Yes, the learning file from Marc Lacross still works, but the one from Wael is much better.

Mine was never intended to be used for anything. It was just the result of the 6000 fast games experiment I ran a few years ago to prove that Mike's learning concept was statistically efficient.

It could be nice to implement this mechanism into a modified Polyglot 1.4 so that Polyglot could benefit of some kind of learning mechanism as a book manager.

Marc

When your learn file was first released, Romi got an instant 60-80 point boost from it. Current versions no longer seem to get that kind of benefit. May be you could make another one that good and Romi could make it to Leo's 3rd division.

[Denis P. Mendoza]

Yes, the learning file from Marc Lacross still works, but the one from Wael is much better.

Mine was never intended to be used for anything. It was just the result of the 6000 fast games experiment I ran a few years ago to prove that Mike's learning concept was statistically efficient.

It could be nice to implement this mechanism into a modified Polyglot 1.4 so that Polyglot could benefit of some kind of learning mechanism as a book manager.

Marc

Now you're talking. you just read my mind!

I'm also curious if there's also some possibility to generate a learning file from a collection of strong, perfect pgn games and compare it with the usual learning file. It would take a long time to build a big learning file (6000 fast games for about ~30mb learning file as stated).

Or just include this as an add-on for Polyglot in relation to what Marc suggested. It's simply my wild imagination though .

[ozziejoe]

Well, the learning modeul works really well still, and helps romi hit "above her weight." Here is how it is doing against three engines that are i think over 200 pnts stronger. I am using the same starting position (1 e4, c6; 2; d4, d5; 3; c4!?)

I am going to keep adding stronger engines and maybe increase the thinking time as the learn file gets stronger



1: Romichess 65.0/111 ·····································
2: List512 24.0/37 1=1011111111=01110====101110=1=010100
3: Aristarch 4.50 16.0/37 11=1=1010011111=00==0=0=00000=1=000=0
4: Yace Paderborn 6.0/37 0=01010=0=0100000=00000=0=00000000000

[ozziejoe]

that's an excellent explanation Michael.

What happens to the learn file in the case of a draw? As i strengthen romi's opponents, I think a draw will become an increasingly good outcome

[Graham Banks]

[David Dahlem]

is there a discussion of how romi chess learning works? Can the algorithim be described in a paragraph or so?

There were lots of dicussions about how Romi's learning works. Maybe someone will provide the links. I only have time for a short explaination, right now. I will write something up to include with my next release.

The 'skinny' is, Romi combines two ideas in her learn file, 1)Pavlov's dog experiment and 2)monkey see monkey do.

First, Romi remembers every move of every game for 160 ply (IIRC) and

1) gives a small bonus to every move of the winning side and a small penalty to every move of the loosing side (Pavlov)

2) Romi plays back winning lines played by herself as well as those played by her opponents (Monkey see)

so, Romi will without 'thinking' play back your own lines against you if you are good enough to beat her or her lines untill you beat her at wich time the bonus/penalty will kick in to make Romi change her play untill she finds a way to win.

Great idea!! So the stronger the opponents, the faster Romi learns, is that right?

Regards
Dave

[ozziejoe]

Hi david,

I don't think the opponents can be too strong, initially, because romi will never win. It will just get overwhelmed and the feedback won't be good (e.g., it will play a decent opening line, but lose in the late middle game and so devalue the opening line).

I think you can start with slightly stronger oponents and build your way up. I started with aristarch and yace, and then added list, and than ruffian, and now pharaon.

I think you have to keep playing all the opponents , or you could get this scenario: romi figures out how to beat yace every time. Then it learns another line from ruffian, so it changes its line against yace and loses to yace.


So to get moves that are generally good you do have to have a variety of oponents and force romi to find lines that work well against everybody

[Michael Sherwin]

that's an excellent explanation Michael.

What happens to the learn file in the case of a draw? As i strengthen romi's opponents, I think a draw will become an increasingly good outcome

In the case of a draw, a very small penalty for all moves.

[Michael Sherwin]

Thanks Graham,

Romi has never looked so good!

[Tony Thomas]

Thanks Graham,

Romi has never looked so good!

I thought she was supposed to be a bad girl.