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

BGonline.org Forums

OT: DMP PRs vs. Cubeful PRs

Posted By: Henrik Bukkjaer
Date: Sunday, 1 July 2012, at 10:13 p.m.

In Response To: OT: DMP PRs vs. Cubeful PRs (Maik Stiebler)

>> Compare the error sizes of an opening 31 played 13/10 6/5 at 1a:1a, 2a:2a and 3a:3a.

That doesn't make sense...

31 played correct to the 5 point win more games, but it also wins a lot more gammons, and looses fewer gammons.

So when comparing that move at different scores, especially DMP with scores where gammons count both ways, you will naturally get different error sizes - simply because it's not such a big error a DMP, because you have the gammon part of the move eliminated by the score.

You don't even get the same "2nd best" move at those scores for an opening 3-1.

The cube-value-problem when normalizing, as referred to in the article, doesn't really affect things at DMP? Unless Rockwell meant that at other scores, you'll get artificially inflated error-rates because of this normalization, which you won't get at DMP?

No matter what, at DMP you don't have the difficult cube decisions, but you always do have all the easy non-contact and bear-off moves. One thing is error-sizes, but number of decisions is equally important.

Cube decisions is worth mentioning here, because you have a much tighter filter on those, as to which decisions counts towards your average error rate. Trivial cube decisions doesn't count, in contrast to trivial (unforced) checker plays.

-----

Maybe PR (error-rate) counld be improved a lot, by looking at the "weights":

Instead of using just 0 or 1 (for forced/unforced moves), the model could be improved by using 0.5-1 and 1-1.5 for moves which are partly forced (eg. entering from the bar with only one of the two rolled numbers open) or moves that have more playable alternatives than usual (doubles that are not partly forced).

Also the rule for when a move doesn't count (ie. weight = 0) could be expanded, to include trivial bear-off moves. Moves that takes off 2 checkers and leave no holes (or singles) afterwards, etc.

Think about the position where you have two checkers on your 1 and 2 point. You roll 21 and bear off two checkers. You get "full points" for making this play, because you could have made the worse play of 2/1 2/o...

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.