Re: [computer-go] Computer Go hardware

[Stuart A Yeates <s.yeates@xxxxxxxxxxxxxxxx>]

Frank de Groot wrote:

> There are so many ways to optimize those patterns' value and not being a
> mathematician, I sometimes stare at the screen (output values) for hours and
> suddenly I get an idea, interrupt the learning process, modify 2 lines,
> restart it and the results converge faster/higher..
I suspect that this is a very important point.

The rate of improvement (or the process of improvement) is in the long 
term more important than how good the engine currently is. While there 
are steady stream of new, reasonable, ideas for improvement it may very 
well make sense to favour speed of evaluating these new ideas over cpu 
and memory intensive training. Of course, you also need to be aware of 
which of the new ideas are going to scale well when you do use cpu and 
memory intensive training.

cheers
stuart
[whose engine isn't very good at all]
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/