[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums

New GV article

Posted By: Timothy Chow
Date: Tuesday, 15 December 2009, at 9:34 p.m.

In Response To: New GV article (Jason Lee)

The other thing to be aware of, from a mathematical point of view, is that there are certain assumptions behind the claims of statistical significance (e.g., that we can use a Gaussian approximation) that are at best asymptotically true; i.e., they're true only if we do "enough" trials. How much is "enough" is not completely precise, but one should apply some common sense here. Common sense would suggest that 108 trials is not a lot of data. It's true that you can do some statistical tricks like variance reduction to squeeze as much as possible out of the data, but the bottom line is that there's a limit to how much information you can really squeeze out of those 108 trials. The fewer the number of trials, the more you're really just relying on the accuracy of the bot's evaluation function.

I've personally settled on 1296 as a minimum number of trials for any position. Others may feel comfortable with a lower number of trials, but 108 really does seem pretty low.

Messages In This Thread

 

Post Response
Your Name:
Your E-Mail Address:
Subject:
Message:

If necessary, enter your password below:

Password:
 

 

[ View Thread ] [ Post Response ] [ Return to Index ] [ Read Prev Msg ] [ Read Next Msg ]

BGonline.org Forums is maintained by Stick with WebBBS 5.12.