Does anyone know if the training dataset for the ML weights shipped with #gnubg at git.savannah.gnu.org/cgit/gnub is available from anywhere?

And, if not, is the #FSF going to exclude gnubg from #GNU? (It would be the right thing to do, short of obtaining and shipping the dataset as gnubg's preferred form of modification, under a #freesoftware license.)

@zacchiro I agree it is a problem worth discussing. However, I believe the famous ancestor of gnubg (td-gammon) was trained by self-play. Wikipedia seems to confirm, but I'm not expert enough on neural networks to judge if any magic/human input is needed to reproduce the weights. Indeed, anything that (heavily) uses random generators during the development is potentially problematic.

Follow

@mmonga

Where is the problem?
Record random seeds.
At worst, record the training.

@zacchiro

Sign in to participate in the conversation
Qoto Mastodon

QOTO: Question Others to Teach Ourselves
An inclusive, Academic Freedom, instance
All cultures welcome.
Hate speech and harassment strictly forbidden.