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

BGonline.org Forums

The statsig police

Posted By: Timothy Chow
Date: Monday, 19 October 2009, at 1:48 a.m.

In Response To: The statsig police (Maik Stiebler)

Maik Stiebler wrote:

My point is that your suggestion does not strike me as a particularly efficient allocation of resources, although it should be more efficient than the current method, if you put a decent limit on the maximum number of trials in your method. You really don't want to rollout a play decision until the end of times, even if the equity difference is very small, do you?

I don't understand your response. If the equity difference is small, say 1 millipoint, then following my prescription, one would stop the rollout when the j.s.d. was 1 millipoint. One wouldn't bother continuing until the j.s.d. dropped to 1/2.33 millipoints. So in some cases you would be doing less work than the current method, rather than more.

Also, although I said that I am posting this early in order to give people the opportunity to implement this suggestion early, my comments here also apply if you just want to do the first pass any old way, because you can also apply the square-root threshold when deciding how to allocate resources during the second pass.

One advantage of the square-root method over, say, a fixed j.s.d. threshold is that it is robust to the number of errors. Suppose one bot makes more errors than the other, but typically of smaller magnitude. With a square-root threshold, the size of the uncertainty depends only on the size of the final error total, not on how many errors were made or how big they were. This would not be true of a fixed-j.s.d. threshold.

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.