[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Bug-gnubg] Question about higher plies
From: |
Jim Segrave |
Subject: |
Re: [Bug-gnubg] Question about higher plies |
Date: |
Tue, 27 Jul 2004 10:04:41 +0200 |
User-agent: |
Mutt/1.4.1i |
On Tue 27 Jul 2004 (19:23 +1200), Joseph Heled wrote:
>
> First, don't forget that the real expensive call is NeuralNetEvaluate (and
> friends). Would be nice to see some profiling info, but CalculateHalfInputs
> would be a distant second, I guess. The real time saver is the usage of my
> small nets. Unfortunately, it never went into the mainline since my eval.c
> and GNUGB's have diverged too much, so it is not trivial.
>
It's been a while since I built a profiling version and ran an
analysis, but as I recall, CalculateHalfInputs was the number 2 time
eater in a 2ply match analysis.
Is there a pointer to your eval.c with smallnets? Is it part of your
training software? (not that I'm making any commitemnet to do
anything, as time is currently a scarce resource)
--
Jim Segrave address@hidden