[Top][All Lists]
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Bug-gnubg] Detecting SSE at runtime
From: |
Øystein Johansen |
Subject: |
Re: [Bug-gnubg] Detecting SSE at runtime |
Date: |
Mon, 10 Oct 2005 21:20:09 +0200 |
User-agent: |
Mozilla Thunderbird 1.0.6 (Windows/20050716) |
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Jon Kinsey wrote:
> I did notice a call to NeuralNetEvaluate in FindBestMoveInEval(). Which
> doesn't use the 128 equivalent - does anyone know if this is correct?
Yes, but....
It's correct because FindBestMoveInEval calls the smaller pruning neural
nets. These net's has only 5 hidden nodes. I don't think that's worth
vectorizing.
However... What if it was 8 hidden nodes? Or 12? I guess a vectorized 8
hidden node neural net would be even faster that a 5 hidden node net
unvectorized. Is feasible to retrain the pruning nets to have 8 hidden
nodes instead of 5?
Will it gain anything at all? Is it worth the effort?
Joseph? Any comments?
- -Øystein
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.4 (MingW32)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org
iD8DBQFDSr7p6kDTFPhwyqYRAlDdAJ49eVFzwlMilj0UpEHSYIdKLcipEgCfVvmR
3u8DpK9NSuLSMsLW9QbX3f4=
=/AO3
-----END PGP SIGNATURE-----