MATLAB Central Newsreader - save net after training

save net after training

Mike — Fri, 18 Jul 2008 07:02:53 -0400

Hi

I have studied Sample Training Session in neural network toolbox.
Codes are:

load choles_all
%reset random number generator. Otherwise, one gets different results
every
%time.
rand('state',0); randn('state',0);
[pn,meanp,stdp,tn,meant,stdt] = prestd(p,t);
[ptrans,transMat] = prepca(pn,0.001);
[R,Q] = size(ptrans)
iitst = 2:4:Q;
iival = 4:4:Q;
iitr = [1:4:Q 3:4:Q];
val.P = ptrans(:,iival); val.T = tn(:,iival);
test.P = ptrans(:,iitst); test.T = tn(:,iitst);
ptr = ptrans(:,iitr); ttr = tn(:,iitr);

net = newff(minmax(ptr),[5 3],{'tansig' 'purelin'},'trainlm');
[net,tr]=train(net,ptr,ttr,[],[],val,test);

plot(tr.epoch,tr.perf,tr.epoch,tr.vperf,tr.epoch,tr.tperf)
legend('Training','Validation','Test',-1);
ylabel('Squared Error'); xlabel('Epoch');

an = sim(net,ptrans);
a = poststd(an,meant,stdt);
for i=1:3
figure(i)
[m(i),b(i),r(i)] = postreg(a(i,:),t(i,:));
end

I have some questions:
1/ Why do we need test data? Aren't test data the same as validation
data?
2/I think one needs to save the training results. Then if I have some
newData, I can use the trained "net" to simulate by using "sim". How
to save "net"?

Thank you in advance.

Mike

Re: save net after training

Greg Heath — Fri, 18 Jul 2008 09:19:03 -0400

On Jul 18, 3:02=A0am, Mike <Sulfate...@gmail.com> wrote:
> Hi
>
> I have studied Sample Training Session in neural network toolbox.
> Codes are:
-----SNIP
> I have some questions:
> 1/ Why do we need test data? =A0Aren't test data the same as validation
> data?

No.

The Design set is composed of
a. the Training subset used to estimate weights and thresholds
b. The Validation subset used to determine when to stop training

Error estimates using training and/or validation data will be biased
because both are used to design the net.

The Test set is used to obtain an unbiased estimate of the
generalization error (i.e., error on nondesign data).

> 2/I think one needs to save the training results. =A0Then if I have some
> newData, I can use the trained "net" to simulate by using "sim". =A0How
> to save "net"?

Sorry, I forgot. Before searching for the answer, try something
like

save net

Hope this helps.

Greg

Re: save net after training

Greg Heath — Fri, 18 Jul 2008 09:22:45 -0400

On Jul 18, 5:19=A0am, Greg Heath <he...@alumni.brown.edu> wrote:
> On Jul 18, 3:02=A0am, Mike <Sulfate...@gmail.com> wrote:
>
> > Hi
>
> > I have studied Sample Training Session in neural network toolbox.
> > Codes are:
> -----SNIP
> > I have some questions:
> > 1/ Why do we need test data? =A0Aren't test data the same as validation
> > data?
>
> No.
>
> The Design set is composed of
> a. the Training subset used to estimate weights and thresholds
> b. The Validation subset used to determine when to stop training
>
> Error estimates using training and/or validation data will be biased
> because both are used to design the net.
>
> The Test set is used to obtain an unbiased estimate of the
> generalization error (i.e., error on nondesign data).
>
> > 2/I think one needs to save the training results. =A0Then if I have som=
e
> > newData, I can use the trained "net" to simulate by using "sim". =A0How
> > to save "net"?
>
> Sorry, I forgot. Before searching for the answer, try something
> like
>
> save net

whoops!

save filename net

produces filename.mat

Hope this helps.

Greg