How to set manually training and test data for training a neural network

Hello All,

I am making some experiments with mathlab neural network toolbox. Usually to train a neural network i give some training and test data and the net object takes care of dividing the data. However what i need to do is divide the data by myself and set the training and test data explicitly in the net object.

If it is needed some aditional explanation please tell me. Any help would be appreciated.

Thanks.

[Geoff]

From memory, if you look at the help for 'net', it tells you how to do this. You can specify which ranges are used for training, validation and testing. I don't have the neural network toolbox anymore, so I can't be more useful than that.

[Eduardo]

Geoff thanks for your answer,

In the help I get the following:

" NET.inputs{i}.range

This property defines the minimum and maximum values of the input data

used to configure the input, or zero if the input is unconfigured.

"

However what I need is to set directly the test and training data. Again thanks for your suggestion.

Answer 1

[Geoff]

Read this:

http://www.mathworks.com/help/nnet/ug/divide-data-for-optimal-neural-network-training.html

You want the divideind option.

[Geoff]

Incidentally, I found that page by googling "matlab neural net data division".

MatLab Answers is great, but often Google is better ;-)

[Eduardo]

Thanks!, I think you point me out in the right way. This is helpful because I need to add external data to the training set and see if it helps in decreasing error (at least a little bit) in the test set.

Answer 2

[Greg Heath]

It is probably easier to use 'divideblock'

Hope this helps.

Greg