Neural Network with large data set

[Rui Dias]

Hey guys, I'm doing a neural network for college that has 10 inputs, 9 outputs and 1 million lines of dataset. I already noticed that the dataset doesn't influence the number of hidden layers, so I'm just using 10 and I have some questions if you guys don't mind answering. Does supervised or unsupervised influence on the time it takes to find the solution, I didn't quite understand the difference between them. Out of the learning rules, which one should I use (backpropagation, backpropagation with momentum, etc)? I read somewhere, there was a way to split a large set into smaller sets and test which is the best, for faster learning (not like crossvalidation, if I understood correctly, crossvalidation,takes small test sets, not small training sets). I can't find this option in Neuroph. Is there any advice at all you can give me, so that I can make this work? I mean it's kind of frustrating, I used a program to make that data which contains almost every possibility of inputs/outputs but I can't make the algorithm work.

[Zoran Sevarac]

If you have target outputs use supervised, otherwise use unsupervised
Use backprop with momentum
Yes there is option, right click data set then 'create training and test subset'