Asking for a favor

[Ozymandias]

An Arena opening book will be selected for the gauntlet engines to use in the final test. I could use some suggestions for that book.

Arena books aren't exactly at the top. Unless you have a particular reason for using that GUI, I would recommend using Fritz. Brain1.ctg won SCCT Book ChampionShip XVI (latest in the series), and it's free.

[fauzi]

If you are interested I can donate my opening book for this project, which is the strongest opening book for Arena out there, if you need also some specific options feel free to tell me and I can modify it to meet the needs.

Here is the link so you can read a little about it, and if you are interested reply or pm me here so I can send it to you.

https://gumroad.com/l/fzabk

Best wishes for your project and best regards,
Fauzi Akram

[Michael Sherwin]

I was thinking putting them in a big archive and uploading them to a file sharing site. I can PM you when they are uploaded.

Okay got the engines. It will take some time to set them all up and register them in Arena. I will use 100 engines for the test. The first run will be with learning disabled to get a baseline performance. The second run will be with 800 trained games from the initial starting position. Then with 12,000 trained games from Hert500.pgn and mostly from the Nooman 3 move test suite also 500 positions. An Arena opening book will be selected for the gauntlet engines to use in the final test. I could use some suggestions for that book.

That will be 600 games at 5'+1". Then if I get a recent Komodo I will run the test from the beginning to see if there is any difference in the quality of learning as opposed to learning against SF 8.

I did not realize that a whole new layer of complexity would manifest by adding a third phase using an opening book. using a weak book would skew the results one way and using a strong deep book would skew the results the other way. Using to wide of a book would require a proportionally larger amount of games as fewer games would produce a larger range of outcomes. I might have to recant on the third phase using a book because I'm only interested at this time in determining what difference 800 games of training makes against a gauntlet of engines from the original position. Then the difference that 12,000 games of training make. So for less random variation I think that the test should not be expanded beyond the opening position for now.

Thanks for the book suggestions though!