Thanks, seems to be the same nets I am using - so no change in 1.01 -Joseph
On Sat, 10 Feb 2024 at 12:18, Philippe Michel <philippe.mich...@free.fr> wrote: > On Fri, Feb 09, 2024 at 09:58:24PM +0100, Øystein Schønning-Johansen wrote: > > > Do you have the raw neural network weight files? > > The source file used by gnubg is in th CVS repository: > > https://cvs.savannah.gnu.org/viewvc/*checkout*/gnubg/gnubg/gnubg.weights?revision=1.4 > > I have uploded it in the format used by gnubg-nn at: > https://alpha.gnu.org/gnu/gnubg/nn-training/1.01/nets/ > > There are no training_data or benchmark subdirectories of 1.01, I used > those of 1.00. > > > Do you know why gnubg uses a binary weights file ? I suppose parsing > 100000 floating point numbers in text format used to be slow, but > nowadays it is barely noticeable, if at all, when starting the tty > interface and dwarved by the time the GUI takes to open. > > Maybe we should get rid of this? gnubg already can read the text file ; > if there is no gnubg.wd file where it expects to find the weights file > it will try to read a gnubg.weights if there is one in the same place. >