Re: [computer-go] strange results

[Aidan Karley <aidan_karley@xxxxxxx>]

In article <6.0.1.1.0.20050511215210.01e20fa0@xxxxxxxxxxxxxxxxx>, Tom 
Cooper wrote:
> On the face of it, this seems to contradict the rules of probability.  Is
> there some reason why this might be happening, or at least why it is not
> as strange as it at first seems?
>
       Lots of possible, plausible scenarios. 
       Consider the circumstance that you're measuring the distance between 
Land's End (Cornwall, UK) and Nova Scotia (east Canada) to an instrumental 
accuracy of 0.5cm (2-sigma), and repeating the measurement at 5 year 
intervals. You'd get a series of measurements with a 1 cm 95% confidence 
interval, where the discrete measurements don't even overlap, even though 
your instruments are fine and your analysis is fine. The underlying reason 
is that the model on which you base your analysis is missing an important 
factor (the spreading of the Atlantic - I'm a geologist <G>).
       Following this analogy, a plausible missing factor might be that one 
of the two programs is learning the repertoire of the other as they play - 
so there is a directional trend in the inter-program strength difference. 
Actually you could get the same effect by having one of the programs work 
through a library of joseki and discard the ones it doesn't have success 
with, so it might not strictly be a relationship between two particular 
programs, but between one program and the rest of the world. 
       
-- 
 Aidan Karley,
 Aberdeen, Scotland,
 Location: 57°10' N,  02°09'  W (sub-tropical Aberdeen), 0.021233
 Written at Wed, 11 May 2005 22:08 +0100


_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/