Re: [Bug-gnubg] Re: Bug? Pruning On/Pruning Off gives significantly different results at 4ply

[Joseph Heled]

I am not sure if you are missing the point or me. Being "more accurate" may be nice but unless the training is repeated it is probably harmfull.

On Tue, Sep 1, 2009 at 9:03 AM, Michael Petch <address@hidden> wrote:

>From what I can tell the new sigmoid function when used with SSE2 is faster and potentially (although slightly) more accurate. This change only applies to SSE2 builds (Which is now standard with Max’s new SSE2 enabled Windows releases)

The bulk changes of interest for SSE2 builds can be seen here:

http://cvs.savannah.gnu.org/viewvc/gnubg/lib/neuralnetsse.c?root=gnubg&r1=1.14&r2=1.15

On 31/08/09 2:51 PM, "Michael Petch" <address@hidden> wrote:

Oystein made SSE2 changes including a sigmoid patch on Jun 21st for performance reasons. Releases after that are slightly varied in their evaluations from previous releases (Something both Max and I have seen)

On 31/08/09 2:40 PM, "Joseph Heled" <address@hidden> wrote:

"sigmoid has changed"