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

BGonline.org Forums

RO

Posted By: Chuck Bower
Date: Tuesday, 3 July 2007, at 2:08 p.m.

In Response To: RO (Stick)

The problem with SW (OK, one problem) is that it doesn't show you a cube-adjusted uncertainty. If you do live-cube it does show one, and always for the cubleless numbers. There is probably a canonical factor (which GNU-bg experimentation would find) that you can multiply the cubeless uncertainty by to get the cube-adjusted.

GNU-bg's uncertainties are pretty solid (AFAIK) except for a very small number of trials. I think 36 trials should give you an accurate uncertainty. IMO it helps to stop after a multiple of 36, until you get up to (at least) a few hundred trials, though.

I believe there are some situations where SW in particular UNDERestimates the errors. One example is the Bagai conjecture: cube rollouts comparing double+take vs. no double, because of the parallel dice + plays. The reason this is a stronger effect for SW than GNU-bg (my claim) is that GNU-bg ends its rollouts when D+P but since SW is rolling out (effectively) cubeless it doesn't stop until one side bears off 15 checkers.

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.