Cupcake 1.1a

[Dan Honeycutt]

Cupcake 1.1a is available for anyone interested; find it at:

https://dl.dropbox.com/u/80776806/Cupcake_11a.rar

Version 1.1 adds a couple of search enhancements and quashes a bug or two so it should give a few elo improvement over version 1.0. The evaluation is basically the same. The "a" following the "1.1" stands for "ass" (mine, to be exact). That's where I pulled all the evaluation factors from.

I don't want to create unrealistic expectations, but there could be a Cupcake version 1.1b in the not too distant future. The "b" stands for "better". At least we hope it's better. Cupcake 1.1b is the same as 1.1a except the evaluation factors will be auto tuned. Adam Hair has been working on this for several weeks. He has reported some progress as well as some setbacks.

Adam's work could shed some light on the viability of auto tuning. (No pressure, Adam). The fun part of chess programming is experimenting with new concepts and ideas. The drudgery is running a thousand games to try and determine if your trapped bishop penalty should be a bit higher or a bit lower. IMHO auto tuning has the potential of reducing the drudgery while giving you more time for the interesting stuff - a win, win situation.

Best
Dan H.

[Sylwy]

Thank you, Maestro Dan !

SilvianR

[Rebel]

Good to see you active again Dan.

[Adam Hair]

Cupcake 1.1a is available for anyone interested; find it at:

https://dl.dropbox.com/u/80776806/Cupcake_11a.rar

Version 1.1 adds a couple of search enhancements and quashes a bug or two so it should give a few elo improvement over version 1.0. The evaluation is basically the same. The "a" following the "1.1" stands for "ass" (mine, to be exact). That's where I pulled all the evaluation factors from.

I found it to be ~30 Elo stronger at short time controls.

I don't want to create unrealistic expectations, but there could be a Cupcake version 1.1b in the not too distant future. The "b" stands for "better". At least we hope it's better. Cupcake 1.1b is the same as 1.1a except the evaluation factors will be auto tuned. Adam Hair has been working on this for several weeks. He has reported some progress as well as some setbacks.

If I manage to eliminate the errors due to the tester (me), then maybe some improvements can be found.

Adam's work could shed some light on the viability of auto tuning. (No pressure, Adam). The fun part of chess programming is experimenting with new concepts and ideas. The drudgery is running a thousand games to try and determine if your trapped bishop penalty should be a bit higher or a bit lower. IMHO auto tuning has the potential of reducing the drudgery while giving you more time for the interesting stuff - a win, win situation.

Best
Dan H.

The real bitch would be to run thousands of games just to find that it really does not matter if the trapped bishop penalty is a little higher or lower.

Auto tuning, in one form or another, does make things better for the author. It is a more efficient method for finding good values for your program, which leads to more time for working on the engine. Of course, in Dan's case auto tuning has not proven itself yet. Hopefully I can get him some useful results before he decides to retire from being an engine author again

[Graham Banks]

Cupcake 1.1a is available for anyone interested; find it at:

https://dl.dropbox.com/u/80776806/Cupcake_11a.rar...

Thanks Dan

[Sylwy]

Hi !

Being extremely interested in Java chess engines I did a short test: Cupcake 1.1a-BeRoChess 1.0. The hash for Cupcake was 64 MB (the hash difference comes from JVM) and 5'/move in ini file.
The final score was 3-1 for BeRoChess.
Not visible bugs yet.



Regards,
SilvianR