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

BGonline.org Forums

Real Position #1

Posted By: Taper_Mike
Date: Saturday, 19 January 2013, at 3:37 a.m.

In Response To: Real Position #1 (Bob Koca)

–

I think it depends on your purpose.

If all you want to know is whether XGR++ can find the “right” play, then your approach is certainly correct. When XGR++ is wrong, all you may care about is by how much, and that is the error statistic you have championed. For this purpose, I want onboard your train.

But consider another possibility. Suppose Ken is interested to use XGR++ to evaluate a series of potential reference positions. In each, let’s say XGR++ is wrong (compared to the rollout), and declares that the rollout’s top play is worse by 0.10 than the play XG picks. Further, suppose that the rollouts always put XG’s play in second place, trailing by only 0.0001. By your standard, Ken should be satisfied that XGR++ has done a top-rate job. Its error was only 0.0001. Ken can then rely on XGR++, and duly commit his reference positions to memory, certain the second best play is a big whopper, when in actuality it is essentially tied.

For this latter purpose, it seems clear that the relative difference I suggested is a far better measure of the utility of XGR++. Whenever it matters to Ken just how far back the second-best play is he should prefer measuring the swing rather than the error.

So, you can see, it depends on your purpose.

Mike

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.