How i could use too large data sets in neural networks ??

I'm using neural networks in making regression and I want to use it in training large data sets with output data more than the input >> in the below code I've got an input of [12*480] and the output is a [1*905] cell array each cell has a matrix of [12*3], each time I run the code it gives me an errors that "Output data size does not match net.outputs{2}.size", please help me for repairing this error, thanx in advance.

A=[];
for j=1:905
k=rand(12,3);
A=[A,{k}]
end
P= rand(12,480)'
net=newff(minmax(P), [20,5], {'tansig','purelin'},'trainlm');
net.trainParam.show = 50; 
net.trainParam.lr = 0.05;
net.trainParam.epochs =1000; 
net.trainParam.goal = 1e-3; 
Nets=[];
for bi=1:905
  T=A{:,bi}';
  net1=train(net,P,T);
  Nets=[Nets,{net1}];
end

at the end of the simulation the variable Nets should have 905 model each with [12*3] network

[Walter Roberson]

This will not solve your problem, but as a matter of efficiency you should recode as

NumGen = 905;

A=[];
for j=1:NumGen
k=rand(12,3);
A=[A,{k}]
end
P= rand(12,480)'
net=newff(minmax(P), [20,5], {'tansig','purelin'},'trainlm');
net.trainParam.show = 50; 
net.trainParam.lr = 0.05;
net.trainParam.epochs =1000; 
net.trainParam.goal = 1e-3; 
Nets=cell(NumGen,1);
for bi=1:NumGen
  T=A{:,bi}';
  net1=train(net,P,T);
  Nets{bi} = net1;
end

[H.k]

thanks for recoding.. but the error still exists "Output data size does not match net.outputs{2}.size"

[Walter Roberson]

As I indicated, you should recode in terms of newfit()

Answer 1

[Walter Roberson]

In https://www.mathworks.com/matlabcentral/answers/77741-newff-create-a-feed-forward-backpropagation-network-obsoleted-in-r2010b-nnet-7-0 Greg Heath says that the syntax of newff that you are using is "very obsolete"

You should consider recoding in terms of the newer function newfit()

[Greg Heath]

newFIT for curveFITting/regression

and

newPR for PatternRecognition/classification

are special cases of the generic

FeedForward function newFF.

ALL ARE OBSOLETE. The upgrades are

FITnet for curveFITting/regression

and

PATTERNnet for PatternRecognition/classification

are special cases of the generic

FEEDFORWARD function FEEDFORWARDnet.

For details of all 6 functions use the documentation commands

 help , doc and type

Hope this helps.

Greg

[Walter Roberson]

Thanks, Greg.

[H.k]

thanks, it works.