I how to divide target data for feed forward network.

I have solved lorenz equation and got value of x,y,z. Then I have divided x ,y,z data into inputX, inputY, inputZ data and target into targetX,targetY,targetZ data. All input and target data are separate matrix of size 1*20000.

Now I have to train the neural network by dividing each target data into trainInd,valInd,testInd of size 1*15000, 1*2000 and 1*3000 respectively. please tell me how to do this.

Answer 1

[Greg Heath]

I don't believe a 20K dataset is needed to train/val/test the net. I recommend designing with smaller multiple sets of data and testing on the rest. The amount of design data can be increased until successful.

 size(input) = [3 N]
 size(target) = [3 N ]

Hope this helps

Greg

[uday]

I have to do one step ahead prediction using feed forward neural network for this I have 3*20000 data of lorenz series solved using ode45 solver on time step of 0.01. how to do one step ahead prediction using this dat by using feed forward back propagation neural net.

[Greg Heath]

Except for the size of your data set it looks like a straightforward example for narnet.

 help narnet; doc narnet 

 help closeloop; doc closeloop

 Also, search for posted code

 greg narnet closeloop