Hi, I searched the net for the ANN library, but I couldn't find it. Does anyone know where it can be downloaded?

best

the ANN-library contains a dll file and some pd-files. They should work, right? I mean, do I put it into wrong directories or what's the clue? It's really just the main objects [ann_mlp] and [ann_td] that don't load. Strange...

Hi Elden,

I am sorry but I really don't know the answer to these questions, I never used this library before, all I did was to search for it online and I found the source code. I honestly think you should ask IOhannes about it, since he is one of the creators of the library. Unless, of course, someone else in the forum is familiar with it. I am also not familiar at all with compilations on Windows, so there's that as well...

Cheers,
Gilberto

Yeah, ok.
I already contacted Davide and IOhannes. No answer, yet. Well, I'll try it with patience ...

Installed it now correctly. I had to cut - paste the content of the subfolders into the "extra" path. Works. I wonder what the maximum number of neurons is. I can't find any infos on that. I need really many of them. didn't get more than a hand full...

@elden i tried to compile the library a while ago too and had similar problems. great that you made it work for windows. if i understand it right the ann library is already compiled on this link, but i need to compile fann? http://puredata.info/Members/dmorelli/ann/ann/ which subfolders did you cut and copy into the "extra" path? the fann subfolders? and is it possible for you to upload a windows compiled version of fann?

Hello Jona, i'm sorry I cannot recall - it's about two years since that. I found a better way to do what I wanted, so the artificial neural net approach went obsolete for me pretty soon. I think you need to find out by trial and error.

Ml-Lib includs a neural network tool too. Works right away.

https://github.com/cmuartfab/ml-lib

@Hugo Legrand thanks a lot. i downloaded the file "ml-0.17.0-alpha-WIN32.zip" from this link: https://github.com/cmuartfab/ml-lib/releases somehow the windows build contains only 2 .pd objects (ml.svm-help and ml.zerox-help) but a lot of .dll´s. i added the library path to the search path and the name to start processes and pd prints can´t load library. the 2 objects are recognized but only half of the messages do work. so i am not sure if i do something wrong, or if the build does not work (for me). i tried it with the osx build, in comparison to windows build the library get loaded on startup but everything else seems the same.
this are the printed messages if i click every help-file message from ml.svm-help.pd:
ml.svm: Support Vector Machines based on the GRT library version 0.2.5 revision: 2-Jan-2017
error: ml.svm: messages with the selector 'bang' are not supported
error: ml.svm: function not implemented
error: ml.svm: messages with the selector 'bang' are not supported
error: ml.svm: messages with the selector 'normalise' are not supported
error: ml.svm: function not implemented
ml.svm: new input vector size, adjusting num_inputs to 4
error: ml.svm: class label must be a positive integer
error: ml.svm: messages with the selector 'predict' are not supported
error: ml.svm: messages with the selector 'estimates' are not supported
error: ml.svm: messages with the selector 'load' are not supported
error: ml.svm: messages with the selector 'save' are not supported

On mac we have access to the 0.18 2.0 alpha. It was released 7 days ago.
Maybe a new window version is on its way. You can wait or send a message to Jammie bullock.
By the way, when I start pd, it tells me it cannot load ML-lib, but all objects are availaible and fully operationnal. Except ml-dtree that instantly causes pd to desapear ^^ Seems like it's the same for you so
don't worry.

To know what messages you can use, send a "help" into the left inlet of any
of the object. Then look at the pd window. It works in the last release.
The neural network is the "ml-mlp" object.

Thanks. Good to know.

I just found out that the Ann library isn't compatible with PD0.43-Extended. But works fine with the last PD release (PD0.48).

And it's the same for ML-Lib.

Furthermore, the ann perceptron is more powerful: it can have multiple hidden layers.

@Hugo Legrand thanks, i tried again and finally i think i figured out how to load the libraries. the answers are already above. @gsagostinho s ann link http://sourceforge.net/projects/fann/files/fann/2.2.0/FANN-2.2.0-Source.zip/download works for me, i had to add the lib to path and start. and ml-lib was already running (only need to set the path), but the strange thing that irritated me is that half of the help file messages do not seem to work with the object (i colored those in the screenshot) but all messages that are printed when i send the help message seem to work. ann is a little bit better documented, but i think i will check out @beep.beep `s patch for trying to understand how the libraries work.

Hi there,

Yes, as you mentioned, the ml-lib library only has pd help files for the svm & zerox objects. And, judging by the comments next to those messages, they look more like tests than completed documentation. If you send the message |help( to those objects, they do post some usage instructions to the console, but that's about it as far as current info for pd users.

I think I'm personally leaning more towards working with the ann library for now—partly due to the lack of ml-lib documentation, and also because ml.mlp still seems a bit unstable on Mac.

+1 for the ann library. The fact that the accuracy of the training is displayed during the training also is a plus. Make it possible to detecte what part of the trainning causes issues.

That ml-lib instability is a pity, because this library includes so many powerful tools.
The guy who created it know is a senor engineer at google.