Re: [Bug-gnubg] html export (2)

[Joern Thyssen]

On Mon, Jul 29, 2002 at 04:15:19PM +0200, olivier croisille wrote
> A few further comments
> 
> 
> 1. It's been a long time since I thought about this : It is very pleasant to 
> be ranked "world class", and it IS deserved sometimes (could those long 
> studying hours prove to pay eventually?? :-))) ... BUT ... in my humble 
> opinion GNU is too kind, as far as evaluation is concerned, with the 
> miserable human players we are :
> 
> consider file attached : 7 pointer where I handled the cube close to perfect 
> but with a -rather unusual I must say :-)))- frightening total of 5 blunders 
> and 5 bad moves for checker play (by my standards, that is, respectively 
> -0.100 and -0.050). And yet, GNU gives me a WC rating in the end (see match 
> stats), which is absolutely not deserved IMO, as a 7 pointer is a relatively 
> short match. True world class is much more demanding than that.

Yes, but you gave you a total euqiyt of -1.341 on 135 moves,
corresponding to 0.0099 pr. move. This is precisely enough to give you
the world class rating.

I would be happy if I played with an error rate of 0.01 :-) My error
rate is typically something like

error rate = (n + 1 ) * 0.01, 

where n is the number of beers I've been drinking.

Note that the chequer play rating obtained by dividing the total equity
given up with the number of *unforced* moves.

I've plans to implement something similar for the cube decisions. In
that particular match you have 77 cube decisions, but many of these are
"forced", i.e., very easy decisions. For example, in game 4 you have 20
cube decisions, but many of these are trivial as your equity is very
close to -1 far far away from a double.

> 3. file, moves 5,7,9 : Red tries the "trick" to gain a free point by 
> delaying his double when odd away-1 away and reaching a gammonish position 
> when Black could hesitate taking (wrong hesitation if he wins more than he 
> get gammoned BTW). GNU indicates "missed double" in export eventhough equity 
> are aboslutely equal as REd can double any time in the game.

Only the 3 leading digits are shown. Perhaps there is a small difference
on the fourth decimal.

Anyway, gnubg's algorithm is to double if equity for doubling >= equity
for no double and >= equity for double pass, so gnubg thinks you should
double.

So, it counts as a missing double, but since the error is 0.000, it'll
not count in the error rate pr cube decision, and hence the cube
decision rating and the overall rating will be unaffected.

Jørn

PS! Can you compress attached files next time or publish them on a
web-site, please! Some of the list subscribers have modem connection :-(

-- 
Joern Thyssen, PhD
Vendsysselgade 3, 3., DK-9000 Aalborg, Denmark
+45 9813 2791 (private) / +45 2077 2689 (mobile) / +45 9633 7036 (work)