How to save neural network

I have used neural network to the datasets and now if i trained network one time and now i want to apply this network to different datasets then how i can apply? I want to do this because i want to compare how efficient the transferability? I am eager to find solution.

Answer 1

[Greg Heath]

clear all, close all, clc;

[x,t] = simplefit_dataset;

[I N ] = size(x)

[O N ] = size(t)

xtrn = x(1:2:N);

ttrn = t(1:2:N);

MSEtrn00 = var(ttrn',1)   % Reference MSEtrn

xval = [];               % No val data for Early Stopping

tval = [];

xtst = x(2:2:N);        % For post training testing

ttst = t(2:2:N);

MSEtst00 = var(ttst',1)  % Reference MSEtst

H = 4                    % Minimized by trial & error

net = fitnet(H);

net.divideFcn = '';     % No automatic data division

net.trainParam.goal = 0.01*MSEtrn00;   % Want training R^2 >= 0.99

[net tr Ytrn Etrn] = train(net,xtrn,ttrn);

ytrn = net(xtrn);

etrn = ttrn-ytrn;

MSEtrn =mse(etrn)

check1 = max(abs(ytrn - Ytrn))

check2 = max(abs(etrn - Etrn))

check3 = max(abs(MSEtrn - tr.perf(end)))

NMSEtrn = MSEtrn/MSEtrn00   % Normalized

R2trn      = 1-NMSEtrn      % R^2

% Save for use with other data

net01 = net;

save net01                 % Save

whos net net01             % Both in workspace

pause(5)

clear net net01           % Cleared from workspace

whos                      % Proof

%Retrieve and use

load net01

ytst = net01(xtst);

MSEtst = mse(ttst-ytst)

R2tst = 1-MSEtst/MSEtst00

Hope this helps

Thank you for formally accepting my answer

Greg

[AMINU]

Hi greg,

I am trying to see how these codes work, but it shows that 'maxabs' is an undefined function. please help

thanks

[Walter Roberson]

Possibly http://www.mathworks.com/matlabcentral/fileexchange/37075-maxabsx-returns-the-maximum-absolute-value-of-x

[Greg Heath]

Close. I had defined a similar function which wasn't quite as general: max(abs(x)) instead of max(abs(x(:))