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

BGonline.org Forums

Lowest winning percentage which is too good to redouble

Posted By: Rick Janowski
Date: Thursday, 11 July 2013, at 10:18 a.m.

In Response To: Lowest winning percentage which is too good to redouble (Ray Kershaw)

To understand the nature and magnitude of the cube vig in a position it is ideal if the cubeless values are known as this is in effect the baseline datum. From knowing the cubeful and cubeless equities (and cubeful percentages), the actual cube efficiency, x, in any position can be estimated to fairly high accuracy. If a particularly strange or unexpected result occurs this is a very strong indicator of inaccurate cube action by the bot. An extreme example would be were x is calculated to be less than zero or greater than 1.00, taking it outside the dead-live window. However, the percentages from the cubeful rollout, particularly those for ND (in the case of Too Good decisions) should normally give a reasonable estimate of cubeless winning chances, provided they are fairly similar to the D/T percentages. In cases were the cube is clearly being mishandled yet checker play appears correct (as often happens in badly busted positions or other extreme positions outside of the neural net's normal experience) it is usually still possible to provide reasonable estimates of cubeful equities from cubeless to cubeful conversions, thus enabling proper cube action to be ascertained.

In your example and those of Mochy, Xavier and Bob, the ND percentages do appear to provide good estimates of cubeless winning chances. Moreover, in all four cases the cube efficiency back calculated from the rollout data appear particular low with values between 0.09 and 0.16. If the values were any higher the margin of the take good decision would be more pronounced. This leads me to believe that all four rollouts are to be trusted as being valid too good positions. If the bot is making any errors it is more likely to be cashing prematurely rather than playing on too late, in my view, which might increase the margin of the too good decision slightly.

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.