Hi list, I just finished a little external I wrote to use FANN classes in PD. It is about multi layered neural networks.
I would like to know who is the author of the ann_som external, I think he could be interested in it (for ann_mlp).
If you want to have a look at my external here it is: https://puredata.org/Members/dmorelli/nn-0.01/download With sources, examples and a compiled for win32.. It should compile on linux and osx also but I didn't try (you'll need to add paths to fann src and libs).. if you do it please send me a makefile and I'll add it.
Best, Davide.
Davide Morelli wrote:
Hi list, I just finished a little external I wrote to use FANN classes in PD. It is about multi layered neural networks.
I would like to know who is the author of the ann_som external, I think he could be interested in it (for ann_mlp).
its me.
i haven't tried your code yet, but if you do not mind, i would highly appreciate if you would like to include your external into ann.
btw, since i haven't done anything to ann within the last 4 years (or so), i wouldn't even mind if you join the development of ann (or take it over...)
the only thing is, that it should remain onder GnuGPL.
mfg,.a.dr IOhannes
IOhannes,
I agree the best place for my external is inside ann (that is why I posted to the list) and I also agree that it should remain GnuGPL. I will just rename it from "nn" to "ann_mlp", is it ok for you?
It is the first time I put something in CVS so please be patient and help me not messing up your code: my external needs fann so it will need a different makefile from ann and ann_som, I think is better to put ann_mlp in a subfolder than ann and ann_som, something like externals\ann\ann_mlp Am I right? Or is it better to change the makefile and put everything in the externals\ann\src directory?
I am honored to share an external with you, by the way thank you for GEM, zexy and the external howto (I wrote my external only thanks to it!).
My external has one major problem: the training process. It is tricky to fill a file with training patterns but the real problem is that when I train the network using the training file I halt pd until the function returns (see train_on_file function) which can take minutes. This slows down the pc. Is there a way to do it in an asynchronous way? Something like creating a thread? Is out there an example of external using threading?
Davide.
-----Messaggio originale----- Da: pd-list-bounces@iem.at [mailto:pd-list-bounces@iem.at] Per conto di IOhannes m zmoelnig Inviato: domenica 8 maggio 2005 19.09 A: Davide Morelli Cc: PD list Oggetto: Re: [PD] New external on neural networks
Davide Morelli wrote:
Hi list, I just finished a little external I wrote to use FANN classes in PD. It is about multi layered neural networks.
I would like to know who is the author of the ann_som external, I think he could be interested in it (for ann_mlp).
its me.
i haven't tried your code yet, but if you do not mind, i would highly appreciate if you would like to include your external into ann.
btw, since i haven't done anything to ann within the last 4 years (or so), i wouldn't even mind if you join the development of ann (or take it over...)
the only thing is, that it should remain onder GnuGPL.
mfg,.a.dr IOhannes
PD-list@iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list
-- No virus found in this incoming message. Checked by AVG Anti-Virus. Version: 7.0.308 / Virus Database: 266.11.3 - Release Date: 03/05/2005
Davide Morelli wrote:
IOhannes,
I agree the best place for my external is inside ann (that is why I posted to the list) and I also agree that it should remain GnuGPL. I will just rename it from "nn" to "ann_mlp", is it ok for you?
perfect.
It is the first time I put something in CVS so please be patient and help me not messing up your code: my external needs fann so it will need a different makefile from ann and ann_som, I think is better to put ann_mlp in a subfolder than ann and ann_som, something like externals\ann\ann_mlp Am I right? Or is it better to change the makefile and put everything in the externals\ann\src directory?
i think it would be best to create an autoconf that checks for fann and then creates one single makefile accordingly.
My external has one major problem: the training process.
as georg has pointed out, the best thing (imo) would be to support online learning. and create a file-interface to read back an already trained nn, so you would need the cpu-sucking learn_from_file() only once...
like creating a thread? Is out there an example of external using threading?
i'd rather not go for threads, as they are a pain in the whereever you want. crossplatform would become much harder.
I am honored to share an external with you,
oioi btw, georg, of course you are invited too...
mfg.asd.r IOhannes
like creating a thread? Is out there an example of external using threading?
i'd rather not go for threads, as they are a pain in the whereever you want. crossplatform would become much harder.
well, i don't think that the equation "threads == evil", although it seems to be quite common among pd developers.
cheers ... tim
Just an observation,
Whats the status of the "transparent" threading in FLEXT for externals?
B.
IOhannes m zmoelnig wrote:
i'd rather not go for threads, as they are a pain in the whereever you want. crossplatform would become much harder.
I am honored to share an external with you,
oioi btw, georg, of course you are invited too...
mfg.asd.r IOhannes
PD-list@iem.at mailing list UNSUBSCRIBE and account-management -> http://lists.puredata.info/listinfo/pd-list