[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: [Bug-gnubg] explain the 'proper cube action' statistic?

From: sebalotek
Subject: Re: [Bug-gnubg] explain the 'proper cube action' statistic?
Date: Fri, 7 Jun 2013 11:39:25 -0700 (PDT)

milfor wrote:
> The correct decision is not to double. But if you have a weak oppenent
> that makes the error to pass with more than 15.2% probability in
> situations like that then it would be a good idea to double here.
> So the 15.2% could be called the probability of an opponents error you
> need to justify a double that would normally be wrong against a well
> playing opponent.
> It can be calculated from the other percentages shown: 0.71 / (3.98 +
> 0.71) * 100 = 15.1
> (I think the small discrepancy to 15.2 is due to rounding errors.)

Thank you for explaining this so clearly Milfor and sorry for the delay in

It kind of strikes me that this on screen information is not of that much
practical use (when playing against gnubg) as the chances of it making an
incorrect cube decision are fairly close to zero.

It could be useful against a human player over the board, but you are very
unlikely to have this percentage statistic to hand in a live match (where
simultaneously running gnubg would probably be seen as cheating ;).

Am I missing something: does anybody else have practical examples on how
they use this particular statistic to improve their game (either against
bots or humans)?

I might have said this before, but imho, in live play, there is no way other
than pure guesswork as to what the % chances of your opponent making an
incorrect cube decision are (with any degree of accuracy). 

Or is there any documentation for an average for humans at different skill
levels (beginner, casual, intermed, advanced, expert, etc) that we can

i.e. for myself, a casual player I think I get a massive 25% of all cube
decisions wrong. For others it may be 10%. Or 15%. Or 5%?


View this message in context: 
Sent from the Gnu - Backgammon mailing list archive at Nabble.com.

reply via email to

[Prev in Thread] Current Thread [Next in Thread]