how to write a testsuite?

[elcabesa]

Hi, i'd like to try implementing this paper: Simulating Human Grandmasters: Evolution and Coevolution of Evaluation Functions.
It's about genetic algorithm and evolution of eval function.

the first step to implement is to create a testsuite of fen position and bestmoves, do you know how to pick sample positions from a pgn file and store then them in a epd like files?

thank you all

[AlvaroBegue]

Just a couple of days ago I used pgn2epd for a similar purpose. It's for Windows, but I made it work under Linux using wine.

[elcabesa]

thank you, but how should I use the program?
I have just tried "pgn2epd.exe round4.pgn res.edp" but it doesn't seem to work....

[elcabesa]

understood.....
pgn2epd.exe <round4.pgn >res.edp

[Ralph Stoesser]

I read that the resulting evaluation function was stronger than Crafty's evaluation. So wouldn't it be a good idea to try such eval tuning with Crafty?

[elcabesa]

maybe you are rigth, but I'm trying to improve my engine

[velmarin]

Not if it's this:
If you use Fritz or Arena (guess whatever)

Simply copy position to clipboard

paste clipboard to file fen, or Notepad.

[Ralph Stoesser]

maybe you are rigth, but I'm trying to improve my engine

That's natural. Are you going to publish your results here on Talkchess?

[tpetzke]

Good luck!

I've read the paper also some time ago and took some ideas from it. Personally I don't like the GA algorithm it used. I know there are GA libraries available that do all the evolution, crossover, mutation stuff but I wasn't comfortable using them. I got a GA running using galib and even was able to solve some simple problems (like the all 1) but I feared the complexity to connect it to my chess program.

But using some kind of GA is much more fun than manual tuning even if less successful (don't know yet, just hearsay). But as I'm in for the fun I'm still playing around with the idea.

Thomas...