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

BGonline.org Forums

Calculating confidence values

Posted By: Maik Stiebler
Date: Sunday, 30 August 2009, at 3:02 p.m.

In Response To: Calculating confidence values (Tom Keith)

Suppose I know from experience that the NN eval of a position is within 0.05 of the real MWC 90% of the time. I would presumably divide that 90% equally among the 101 array elements corresponding to +-0.05 of the NN value and distribute the remaining 10% equally among the remaining 900 array elements? [If abs(i/1000-eval) <= 0.05, then A[i] = i*0.9/101 else A[i] = i*0.1/900.]

Yes, letting the evaluation set the prior seems like a good idea to me, and that would be a sensible way to do it. Could it possibly be refined by letting the NN estimate its uncertainty as well?

Alternatively, can I collect the rollout results first, see what the observed mean and deviations are, and use that information to initialize the array?

Don't use the same information for the prior and the updates. However, you could use old rollout information for the prior and update with a new rollout.

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.