NARX Time series Neural Network

Hi,

I have a cell array that contains 2000 separate time series of data. The arrays in the cells are of different length, ranging from about 1 to 20 days, 1 reading per day. Its 2000 patients with readings taken over a number of days and I am trying to predict one value from this

I think want to make each of the time series the same length so that I can try to do time series prediction which is what this says: href = ""<http://webcache.googleusercontent.com/search?q=cache:bdrNPdyY_i0J:www.mathworks.com/help/toolbox/nnet/ug/bss36ff-1.html+&cd=3&hl=en&ct=clnk</a>>

I have tried to do it using this:

xxx = catsamples(getelements(big,1),'pad')

but I keep getting the error X{1,2} and X{1,1} have different numbers of rows.

which is true, but I thought that using the above would make them the same length and merge them, so that I can feed the 2000 instances in as inputs to a neural network.

[monsie]

Are you using the GUI? There you can select that your input is rows instead of columns. That's how NN deals with it.

[Mark]

I have, but the problem is for each patient I have varying numbers of rows of data, some may have 5 rows, others may only have 2, others may have 20. All rows contain the same data columns.

[Mark]

I might be being really dumb, but...If you do it in the gui, how do you tell it that this is one patients set of records, and this is the next patients is what I'm wondering. At the moment I get the feeling its taken all 12000 rows in total and then stuck them in without differentiating - I haven't told it where one patient ends and the next starts. When I have a cell array, its not showing up in the inputs data source when I go in the gui

Answer 1

[Greg Heath]

Since your output is one value, it is not a time series problem.

If that value is a classification into 1 of c categories, use patternnet.

Otherwise use fitnet.

Unfortunately, the data for each net has to have the same number of input variable rows. Therefore you can make 20 different data bases... one with 20 inputs, one with 19 inputs(including those with 20 inputs with the last input deleted) etc.

Hope this helps.

Thank you for officially accepting my answer (;>)

Greg