> Since Gnubg is now over the plateau reached by TD training, > I wondered if a new bout of TD training on top of the > supervised training might be beneficial. Øystein and Joseph, > are you saying that you have already tried this, to no avail?
I've not tried. Maybe it works? Who knows? However, I believe you should reimplement the TD-algorithm. Do the selfplay, but update only the crached and contact, nets according to TD. (I think we should be satisfied with the race net) Joseph? What do you suggest for a TD training of a pretrained net? Trial and error? Start with something high like 1.0 and half this value when you see you're way to high? Try different learning rates? Don't waste time with to high learning rates. -Øystein ------------------------------------------------------------------- The information contained in this message may be CONFIDENTIAL and is intended for the addressee only. Any unauthorised use, dissemination of the information or copying of this message is prohibited. If you are not the addressee, please notify the sender immediately by return e-mail and delete this message. Thank you. _______________________________________________ Bug-gnubg mailing list [email protected] http://lists.gnu.org/mailman/listinfo/bug-gnubg
