geraldg@gmx.de

I'm working on a simulation of artificial neural networks ins Max. After some weeks work, the patch is working, but because of the 256 member limit for lists, it is very constricted... Has anyone an idea how to calculate few calculations with a huge number of participants (up to 10 thousands floats, all doing the same calculations +,-,*, / )

how-to-handle-lists-with-much-more-than-256-members

Quote: geraldg@gmx.de wrote on Thu, 05 April 2007 12:49

----------------------------------------------------

> I'm working on a simulation of artificial neural networks ins Max. After some weeks work, the patch is working, but because of the 256 member limit for lists, it is very constricted... Has anyone an idea how to calculate few calculations with a huge number of participants (up to 10 thousands floats, all doing the same calculations +,-,*, / ) 

> jitter? java? split lists? i don't know!

This hits us all the time with Spectral Synthesis. The options are Javascript, Jitter, Java, or externals programmed in C.

The choice has to do with money, skill, realtime perfomance needs... I would have a look in "listpackage.help" which outlines the java list operators that are included with Max.

Quote: geraldg@gmx.de wrote on Thu, 05 April 2007 12:49
----------------------------------------------------
> Hi!
>
> I'm working on a simulation of artificial neural networks ins Max. After some weeks work, the patch is working, but because of the 256 member limit for lists, it is very constricted... Has anyone an idea how to calculate few calculations with a huge number of participants (up to 10 thousands floats, all doing the same calculations +,-,*, / ) 
>
> jitter? java? split lists? i don't know!
>
> Thanks very much....
>
> GErald
----------------------------------------------------

This hits us all the time with Spectral Synthesis.  The options are Javascript, Jitter, Java, or externals programmed in C.

The choice has to do with money, skill, realtime perfomance needs...  I would have a look in "listpackage.help" which outlines the java list operators that are included with Max.

i might be misunderstanding how neural networks work, but i was thinking that it seems similar to FM synthesis matrices, where you have a number of oscillators all connected together, with varying degrees of connectivity. maybe you could make a massive signal matrix.....

maybe you could train a network of oscillators to speak welsh

i might be misunderstanding how neural networks work, but i was thinking that it seems similar to FM synthesis matrices, where you have a number of oscillators all connected together, with varying degrees of connectivity. maybe you could make a massive signal matrix.....
  maybe you could train a network of oscillators to speak welsh


Just for the fun of it, here's an abstraction that hacks lists into Max-sized bites, so you can do something. It then reassembles them:

Just for the fun of it, here's an abstraction that hacks lists into Max-sized bites, so you can do something.  It then reassembles them:

mzed, your patch is working fine, but the problem is, that i cant get longer lists out of th larray object, so i wold have to hange all my get messages and so on, so I decidet to try it with jitter.

The only problem is, that I have a very basic experience in jitter....

Maybe you can answer me a few questions before?

1. is there ANY internal LIMIT to the jit.matrix ;-))

2. to query rows and collumns, do I have to define submatices for the given parameters then?

thanx for your tips and patches!
mzed, your patch is working fine, but the problem is, that i cant get longer lists out of th larray object, so i wold have to hange all my get messages and so on, so I decidet to try it with jitter.
The only problem is, that I have a very basic experience in jitter.... 

1. is there ANY internal LIMIT to the jit.matrix ;-))
2. to query rows and collumns, do I have to define submatices for the given parameters then?

Have you looked at IRCAM's FTM environment? It's designed to handle matrices and could have very long lists.

Other than that, as suggested, Java externals would do that okay. They're fairly simple to code too.

What are the ANN's doing? Sounds an interesting project

Thanks! But it seems, the FTM is only running under Mac OS!?

For now i rebult it, using Jitter, it was not too complicated, but must say, that finding errors in Jitter matrices is not too easy...

But it is something like working. We'll see how it will grow.

I'm doing this project for my diploma in July, and it is just an experiment to figure out the range of possibilities for the use of ANN in live music environment...

Thanks! But it seems, the FTM is only running under Mac OS!?
For now i rebult it, using Jitter, it was not too complicated, but  must say, that finding errors in Jitter matrices is not too easy...
But it is something like working. We'll see how it will grow.

How to handle lists with much more than 256 members?