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

BGonline.org Forums

How to roll out an entire GNU match?

Posted By: Rich Munitz
Date: Monday, 1 June 2009, at 9:39 p.m.

In Response To: How to roll out an entire GNU match? (Coolrey)

With all of the great things that have been done with GNU, it boggles my mind that it apparently still can't do what I consider to be an indespensible analysis feature of Snowie: analyze an entire match at "N-ply" and in the same pass roll out all errors larger than X.

I do relatively small rollouts (108g x Snowie 2ply) on errors and I don't care about statistical significance. My feeling is that I don't need to review any play for which EITHER the bot's eval or the short rollout agreed with my play. However I do want to review any play that BOTH the eval and the rollout disagreed with my play.

In my opinion, it is the time it takes for me to review the analysis that is more valuable than the time it takes the computer to perform the analysis (which it usually does while I am sleeping). I want to only look at what are likely to be real mistakes. I don't want to interrupt my review to do rollouts because I question the results. And I subscribe to innocent until proven guilty - I'd rather not be made aware of something I did that was really an error than to be led to believe that something I did correctly was wrong. Its not like I can't find mistakes in my play, so just focus on the ones I have high confidence are actually wrong.

I know from experience the value of avoiding this misinformation. 80+% of the time, my error rate goes down, often substantially from automatically rolling out all errors. I am not picking and choosing which ones to question, I do them all. The reason for this consistent drop in ER is that there are lots of positions the bot either does not understand well, or does not have the vision to see how the position will evolve. In many of these cases, the rollouts reverse the eval and confirm the correctness of my play. The average amount by which my ER drops is the amount by which the eval analysis would lead me astray. The last thing I want to be doing during study is to spend valuable time trying to rid myself of my good instincts based on misguided bot evaluations. Short, non-statsig rollouts of all errors eliminate the bulk of these misguided results and what's left is more trustworthy.

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.