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

BGonline.org Forums

Role of truncs, variance of full rollouts

Posted By: Daniel Murphy
Date: Saturday, 19 June 2010, at 5:56 p.m.

In Response To: Role of truncs, variance of full rollouts (Nack Ballard)

Thanks to all of you for your rollouts and comments. Gloss:

A 2k trial, truncated, had bar/21 ahead of bar/22 24/23 by 0.005. The CI's (95% Confidence Intervals) for each play were presumably quite small, due to truncation. A 5k trial, full, had bar/21 ahead by 0.017, with CI's of 0.012 for each play. A 20k trial, full, had bar/21 ahead by 0.007, with CI's of 0.005 and 0.006. In this case, the 2k trial agrees closely with the 20k trial and the 5k trial is odd man out. And in conclusion:

"It didn't have to work out that way. It often doesn't. But overall the predictive value of the 1k trunc and the 5k full seem to be about equal for early game rollouts, in my experience. So, early game rollout enthusiasts, please understand that the old truncated results I drudge up have value. "

Ok. I agree they have value. I don't think anyone argues that short truncated rollouts are either always reliable or always useless. Whatever data are the best we have are just that. Perhaps there is merely a difference of emphasis or value placed on data which are the best we have but are not the best we could have (in this case, there's lots of variance even in the 20k trial). In assigning value, one emphazises that short truncated rollouts tend to be accurate. Another emphasizes that it's difficult to know which of the short trunctated rollouts are especially accurate until longer, more rigorous rollouts supersede them.

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.