|
| From: | Philippe Michel |
| Subject: | Re: [Bug-gnubg] Possible error in SSE sigmoid calculation |
| Date: | Sat, 31 Dec 2011 14:57:50 +0100 (CET) |
| User-agent: | Alpine 2.00 (BSF 1167 2008-08-23) |
On Thu, 22 Dec 2011, Joseph Heled wrote:
I am not sure which nets and which sigmoid implementation is used by gnubg now - but note that my nets were trained against an approximation. It does not matter that much if you use the exact definition - what is important is using a "valid pair", that is a net and the sigmoid implementation it was trained with. In the past, breaking this meant a serious drop in move quality. I suggest someone checks this ....
As far as I understand, the non-SSE sigmoid approximation in gnubg is the same as the one in gnubg-nn (the default code in the latter, with SIGMOIDNEW undefined). With the SSE code in gnubg now fixed, the sigmoid implementation is the same everywhere.
| [Prev in Thread] | Current Thread | [Next in Thread] |