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

BGonline.org Forums

15552-game rollout of 41Z-65R-43

Posted By: Timothy Chow
Date: Sunday, 27 December 2009, at 6:47 p.m.

In Response To: 15552-game rollout of 41Z-65R-43 (Chuck Bower)

Chuck Bower wrote:

Truncated rollouts trade statistical uncertainties for systematic uncertainties, the latter which are nearly impossible to quantify. Why should that make one feel better about truncated rollouts?

Full rollouts also contain systematic uncertainty, assuming your goal is to get at the truth of the position as opposed to what the bot thinks. The trade is not between systematic uncertainty and statistical uncertainty, but uncertainty and computation time. It is a common mistake for people to mentally compare truncated vs. full rollouts assuming unlimited computation time. This is never the case in the real world. The more realistic comparison fixes the amount of computation time. If you have, say, 1000 CPU hours to spare, do you spend it on full rollouts or truncated rollouts? The answer is not a priori clear. It can only be determined by empirical investigation.

Nack says that for opening rollouts, his experience demonstrates that truncated rollouts are more reliable for a fixed amount of computation time. It could have turned out the other way, of course. What I'm arguing is that you can't determine the answer a priori based on general observations about unbiasedness and so forth.

I believe it has been argued that due to dependency between candidates in rollouts that the 'j.s.d.' value is OVER-stated.

I don't think that this is at all clear. I haven't worked this through to the bitter end but my preliminary calculations suggest that there are factors working in both directions.

My guess is that strictly speaking, equities and variances are all undefined because there's a nonzero probability of hitting one of those "weird" positions with provably undefined equity. But even if you somehow work around that problem, it's going to be very difficult to demonstrate that estimated j.s.d. values are usually overestimates.

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.