Hi Davide, a small bug report... I just forgot to send a "create" message before sending a "train" message to nn and pd crashed. I think it should be like with "train-on-file it", which gives an error message. marius.
Hi Marius, Thank you for the report! I've simply forgotten to check if the net is initialized before calling the train procedure... I will fix it as soon as I can and put a 0.03 version on my site (temporarly, as soon as I get access to cvs as a developer I'll put my external in the ann folder with a working linux makefile or autoconf). I think I will also add a second outlet for the mean square error, usefull while training on-the-fly, i don't like to share the same outlet for mse reporting and for network's output...
BTW: thank you for your article on FFT+PD, that's why I passed from csound to pd !
----- Original Message ----- From: "Marius Schebella" marius.schebella@chello.at To: "Davide Morelli" info@davidemorelli.it; pd-list@iem.at Sent: Tuesday, May 10, 2005 6:01 PM Subject: not initialized nn crash
Hi Davide, a small bug report... I just forgot to send a "create" message before sending a "train" message to nn and pd crashed. I think it should be like with "train-on-file it", which gives an error message. marius.
From: "Davide Morelli"
BTW: thank you for your article on FFT+PD, that's why I passed from csound to pd !
I think this this paper is pretty out of date in the meantime. It is only still online because of my melancholic character. I should do an update sometime. Actually, I think MillerŽs book has the true facts about FFT and Pd... http://crca.ucsd.edu/~msp/techniques.htm marius.