Re: [Bug-gnubg] Measuring performance levels

[Douglas Zare]

Quoting Morten Wang <address@hidden>:

> * Douglas Zare
> > So, how does gnubg fare on different settings?
> 
> the 2ply rollout finished during the night, here's the result:
> 
> Win   (g)     (bg)    L(g)    L(bg)   Cubeless        Cubeful
> 0.7742        0,0361  0,0007  0,0642  0,0001  +0,5209         n/a
> 0,0004        0,0005  0,0000  0,0003  0,0008   0,0010         n/a
> 
> the rest of the rollout settings were the same as the 1ply rollout
> (which is a cubeless "world class" setting).

I decided to stop a Snowie 3-ply 33% rollout. After 5544 trials, the cubeless 
equity was 0.546+-0.004. 0.1 3.8 78.7 -- 21.3 6.6 0.0. As an aside, Snowie 
interprets this as a 0.939 take. 

The odd conclusion is that gnu's static evaluations play this position better 
than Snowie's, but Snowie 3-ply plays better than gnu 2-ply. This might not be 
too surprising if the parts of the game in which they misplay are different, 
and the strengthening from 1-ply to 3-ply fixes most of Snowie's problems, but 
from 0-ply to 2-ply does not fix gnu's.

How about 1 closed out, extras on the 23, 10, and 7, versus 13 off with a blot 
on the 1? 

Douglas Zare