neural network - design tool Options

[miri]

Hi everybody,

I'm writing my diploma thesis about neural networks. Now I'm wondering about how to design one. I found a tool called "Matlab" (http://www.mathworks.de/) but it's too expensive for me (yes, I know about the version for students, but unfortunately that's impossible to buy for me).
Does anyone of you know another tool which costs less and is also useful to design neural networks?

[lucid_dream]

scilab is freeware matlab

[miri]

scilab is freeware matlab

yeah, but as far as i can see the neural network toolbox only exists for unix/linux:
http://www.scilab.org/contrib/displayContr....php?fileID=166

[lucid_dream]

write up the neural network computations in C, C++, or your favorite programming language, and then use scilab to plot results. What neural net architectures are you investigating?

[maximus242]

Why not just code it by hand? C++ is a perfect language to write neural networks in, and its cheap. If you dont know or want to learn C++ then you could opt for Python which is free, easy to understand and simple. Granted you wont get the same processing power, but if your not into computer science then its a easy alternative.

[miri]

Well, first of all coding it by myself takes time I don't have and then I'm quite new to that subject.
So I thought I could design the network with MatLab, generate the code and do all the rest.

[lucid_dream]

the Matlab Neural Network toolbox is not that great. The documentation is nice though, and is worth looking at to learn more about neural net architectures. But the toolbox otherwise leaves much to be desired. You'd be better off writing your own script in scilab, which utilizes almost the same syntax as matlab.

I know a few people who write neural network programs in matlab, myself included, but I don't know anyone who actually uses the toolbox for that since it's straightforward to write your own program. If you understand the neural network architecture and the basics of matlab/scilab syntax, then there shouldn't be any problems writing a simple script to implement the neural network. And that can be done in under an hour even for novices.

[miri]

Okok, I'll give it a try.
Thanks very much for your help. I would have never found SciLab otherwise.

[maximus242]

It only takes 16 lines of code to write a basic neural network in Python.

[lucid_dream]

It only takes 16 lines of code to write a basic neural network in Python.

To do it in Python, you need additional Python extensions, like Numpy and probably something for plotting, don't you?

[Rick]

I googled "free neural network" (without the quotes) to discover several, including

http://www.jooneworld.com/

http://www.sigma-research.com/bookshelf/rtthinks.htm

[lucid_dream]

Joone looks promising. Take a look at the samples starting on page 112 of the Joone Help Manual

[maximus242]

It only takes 16 lines of code to write a basic neural network in Python.

To do it in Python, you need additional Python extensions, like Numpy and probably something for plotting, don't you?

Um depends on how you do it, but it only took me one night to learn python, its one of the easiest real programming languages (were not talking HTML which simply presents data in a human readable way). It is rediculously easy to read anyone elses code, so if you look in the google code seach and find some NN python code, its really easy to understand it (unlike C++ which is like walking through a mine field). If you need extensions they usually only take a minute to install and have easy to use documentation, plus you can integrate it with a GUI and have yourself a mini civilization (you could play god)

Which is a lot more fun than basic mechanical neural network building, unless you can get some program to ouput the NN code into C++ because Python and C++ are extremly easy to integrate.

[D.R]

About of how simple is your program I think is better to be faster, C++ is faster a is not based on a instruction by instruction microprosessor system, is a parallel system using interpretated languages will delay results however it always depends to the application.


Regards.

[D.R]

I am still thinking is better a closest machine language like C, C++ or even ASM for programming an Artificial Neural Network

[bob55]

nice info....
thankx
=============bopb

[Yocttar]

I am still thinking is better a closest machine language like C, C++ or even ASM for programming an Artificial Neural Network

I belive you are wrong about the ASM proposal, since most C and C++ compilers nowday make much faster code then a human can (they have a vast systematic view).