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

BGonline.org Forums

Fast Rollouts

Posted By: Daniel Murphy
Date: Sunday, 6 December 2009, at 9:53 a.m.

In Response To: Fast Rollouts (neilkaz)

At http://www.bgonline.org/forums/webbbs_config.pl?noframes;read=54366 D. Montgomery wrote:

"Why bad evaluations don’t introduce bias

"Variance reduction relies on a program’s evaluator to estimate the luck of each roll. Many people fear that this makes variance reduction unreliable. But in fact, even if we have very poor evaluations, a variance reduced rollout will on average give us the same answer as a regular rollout. To see why this is so, we need just a little math.

...

"But what if the evaluations are really bad?

"Bad evaluations don’t bias the result. All variance reduction does is subtract the total estimated luck, a value that averages zero. Subtracting zero can't change the expected value of a rollout. No matter how poorly a program understands a position, luck estimates will still average zero, because of the way they are calculated, and so a variance reduced rollout will produce the same result as a regular rollout, on average."

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.