very happy! v1.18 of Pos won from GNU Chess!

[flok]

I still can hardly believe it but my chess program (http://www.vanheusden.com/pos/) Pos version 1.18 reproducably wins occasionally from GNU Chess!
Since my program uses none of the regular techniques as the Shannon algorithm or brute force in any kind I found this to be a major milestone.
Yes, it won before from Pulchess v0.2i but now that also wins from other software as well it starts to mean something
Me being exstatic is a bit of an understatement.

[hgm]

What exactly does "reproducibly occasionally" mean?

More importantly, how was this win found? If I commit a game where an opponent loses to book, only very few programs would randomize thoroughly enough to not occasionally play the same game again. So a program that would just print all the moves of that game in a fixed sequence, would occasionally win from that opponent (And forfeit all other games due to illegal moves).

Even a random mover would occasionally beat Rybka, as the probability it does all the right moves by accident is non-zero. (Just 10^-100 or so.)

So a much more important question is: can Pos now 'reproducibly' beat N.E.G., in the sense that the reult is reproducible, not the moves.

[flok]

What exactly does "reproducibly occasionally" mean?

Once in, say, 100 games played.

More importantly, how was this win found?

I let the two play against each other using autoplay with default parameters. My engine does not do pondering and a very little bit of multithreading (top cpu usage is around 130%) so GNUChess can use continue 1 and most of the time almost 2 cpu cores (it is an intel e6600 system). 2GB of ram available with my engine using 400MB.

If I commit a game where an opponent loses to book, only very few programs would randomize thoroughly enough to not occasionally play the same game again.

I'm not entirely sure what you mean but it lost after 36 moves so that is beyond the book. The logfile and then .pgn-file are available from my site by the way ( http://www.vanheusden.com/pos/examples/ ... os1.18.pgn and http://www.vanheusden.com/pos/examples/ ... os1.18.log)

Even a random mover would occasionally beat Rybka, as the probability it does all the right moves by accident is non-zero. (Just 10^-100 or so.)

True but my program wins around 1% of the games so I think that is beyond random chance, isn't it? Not sure, I'm not a statisticus.

So a much more important question is: can Pos now 'reproducibly' beat N.E.G., in the sense that the reult is reproducible, not the moves.

Yes. In the last 24 hours (no continues play, had it play during lunchbreak and such) I saw it won at least 3 times.[/url]

[hgm]

I still don't get it. Once every 100 games Pos beats GNU. So that is "occasionally". But now what does the "reproducibly" mean? That when it wins, ita lsways wins through the same game (= same sequence of moves)?

What time control do you use? I could not find it in the PGN. Seems you are playing GNU with depth limited to zero ply, as it overlooks mate in one. Are you sure you are not posting this exactly one week early?

Now the part about N.E.G. I don't understand at all. You also played Pos against N.E.G. for the last 24 hours? And it won only 3 times? If it did not win every time, the result is not reproducible. You only played it 3 times? Do you play 8-hour games? Or do you just don't look?

[Tony Thomas]

By reproducible I think he only meant that POS can now consistently beat GNU chess in every 100 game match were as it lost to Gnuchess previously in every single match.

[flok]

I still don't get it. Once every 100 games Pos beats GNU. So that is "occasionally". But now what does the "reproducibly" mean? That when it wins, ita lsways wins through the same game (= same sequence of moves)?

Reproducibly means that it was not caused by a spontenaous bitflip or so.

What time control do you use?

Non configured. GNU Chess default.

I could not find it in the PGN. Seems you are playing GNU with depth limited to zero ply, as it overlooks mate in one. Are you sure you are not posting this exactly one week early?

Oh darn! I knew it would be too good to be true. Overlooked that one.

Now the part about N.E.G. I don't understand at all. You also played Pos against N.E.G. for the last 24 hours? And it won only 3 times? If it did not win every time, the result is not reproducible. You only played it 3 times? Do you play 8-hour games? Or do you just don't look?

I'll let it run for a while without the ply-limit.