problem of using patternnet in Neural network toolbox

8 views (last 30 days)
I use the pattern recognition network (patternnet) to train a set of data, whose dimension is 7100 and there have 2000 data points. The output is a 2*2000 vector. In other words, there have two classes. I set “hiddenLayerSize” as 50, after running the program, the systems gives the following error message? Is that due to the setup of “hiddenLayerSize” or any other issues?
??? Out of memory. Type HELP MEMORY for your options.
Error in ==> perfs_jejj>calc_Y_trainPerfJeJJ at 89
JJy = Jwb_y * Jwb_y';
Error in ==> perfs_jejj>singlecalc at 58
[Y,trainPerfy,trainN,JEy,JJy] = calc_Y_trainPerfJeJJ(net,data,fcns);
Error in ==> perfs_jejj at 9
[trainPerfy,trainN,valPerfy,~,testPerfy,~,JEy,JJy] = singlecalc(net,data,fcns);
Error in ==> trainlm>train_network at 199
[perf,vperf,tperf,je,jj,gradient] = nntraining.perfs_jejj(net,data,fcns);
Error in ==> trainlm at 113
[net,tr] = train_network(net,tr,data,fcns,param);
Error in ==> network.train at 107
[net,tr] = feval(net.trainFcn,net,X,T,Xi,Ai,EW,net.trainParam);
Error in ==> clusteringSOM at 147
[net,tr] = train(net,inputs,targets);

Accepted Answer

Greg Heath
Greg Heath on 20 Nov 2011
use the pattern recognition network (patternnet) to train a set of data,
whose dimension is 7100 and there have 2000 data points. The output
is a 2*2000 vector. In other words, there have two classes. I set
“hiddenLayerSize” as 50, after running the program, the systems gives
the following error message? Is that due to the setup of
“hiddenLayerSize” or any other issues?
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
For an I-H-O node topology,
size(p) = [ I Ntrn ] % Input
size(t) = [ O Ntrn ] % Output
Neq = Ntrn*O % No. of training equations
Nw = (I+1)*H+(H+1)*O % No. of unknown weights
For accurate and stable weight estimates when training to convergence require that
Neq >> Nw
or
1 <= H << (Neq-O)/(I+O+1)
1 <= H << (2000*2 -2)/(7100+2+1) = 3998/7103 = 0.56
Which is impossible.
Therefore, drastically reduce the input dimension I.
Hope this helps.
Greg

More Answers (0)

Community Treasure Hunt

Find the treasures in MATLAB Central and discover how the community can help you!

Start Hunting!