Matlab - newpnn basic question

Author

newpnn basic question

hong

2008-01-30, 9:35 am

I have simple question for newpnn.
Please help me on this issue. Thanks in advance,

Reproduction Steps: P = [DT GR NPHI PEF RHOB ]';
clear('V_Data');

T= ind2vec(FACIES');
T = full(T);
spread=4;
spread=it;
net=newpnn(P,T,spread);
net.layers{2}.transferFcn = 'softmax';
net.trainParam.epochs = 50;
net.trainParam.goal = 0.01;
net = train(net,p,T);
A=sim(net,P);
AC=vec2ind(A);

----
I got error
??? Error using ==> network.subsasgn at 464
Attempt to reference field of non-structure array.

Error in ==> PNZPNN at 35
net.trainParam.epochs = 50;

Greg Heath

2008-01-31, 5:27 am

On Jan 30, 10:02=A0am, "hong " <stangh...@yahoo.com> wrote:
> I have simple question for newpnn.
> Please help me on this issue. Thanks in advance,
>
> Reproduction Steps: P =3D [DT GR NPHI PEF RHOB ]';
> clear('V_Data');
>
> T=3D ind2vec(FACIES');
> T =3D full(T);
> spread=3D4;
> spread=3Dit;
> net=3Dnewpnn(P,T,spread);
> net.layers{2}.transferFcn =3D 'softmax';
> net.trainParam.epochs =3D 50;
> net.trainParam.goal =3D 0.01;
> net =3D train(net,p,T);

I don't have access to MATLAB on this computer.
However, I don't recall that newpnn has to be
trained. If I'm correct, just delete the last
4 commands.

However, you need to determine an appropriate
value for spread. Therefore
> A=3Dsim(net,P);
> AC=3Dvec2ind(A);
>
> ----
> I got error
> ??? Error using =3D=3D> network.subsasgn at 464
> Attempt to reference field of non-structure array.
>
> Error in =3D=3D> PNZPNN at 35
> net.trainParam.epochs =3D 50;

Greg Heath

2008-01-31, 5:27 am

On Jan 31, 12:49=A0am, Greg Heath <he...@alumni.brown.edu> wrote:
> On Jan 30, 10:02=A0am, "hong " <stangh...@yahoo.com> wrote:
>
>
>
>
> I don't have access to MATLAB on this computer.
> However, I don't recall that newpnn has to be
> trained. If I'm correct, just delete the last
> 4 commands.
>
> However, you need to determine an appropriate
> value for spread. Therefore

You will probably need to loop over a range
of spread values.
[color=darkred]
> A=3Dsim(net,P);

Then calculate error rates.

Hope this helps.

Greg

hong

2008-01-31, 10:19 am

Tahnks Greg Heath.
I looped spread with step 1. from 1 to 7
However, none of the results are satisfying, I use
accuracy...sum(counts of matched)/sum(total counts).
None of them is above 25 percent.
Is there any other way to test? Improve?
>
>
> Then calculate error rates.
>
> Hope this helps.
>
> Greg

Greg Heath

2008-01-31, 10:19 am

On Jan 31, 8:59=A0am, "hong " <stangh...@yahoo.com> wrote:
> Tahnks Greg Heath.
> I looped spread with step 1. from 1 to 7
> However, none of the results are satisfying,

That's probably because you don't know enough
about your data. For example,

1. Analyse the results of

[nvar ncases] =3D size(p)
rank(p)
size(t)
rank(t)
rank([p,t])
corrcoef([p,t])
tc =3D vec2ind(t)
for i =3D 1:nvar
plot(p(:,i),tc)%color code?
end
2. Color coded scatter plots on (pi,pj) planes
3. Color coded scatter plots on dominant PCA planes.
4. Repeat 1 -3 after standardizing p and t.
5. Try a linear and/or logistic model

etc.

Go to Google Groups and find my post on pretraining
advice

greg-heath pretraining advice

Hope this helps.

Greg
[color=darkred]
> accuracy...sum(counts of matched)/sum(total counts).
> None of them is above 25 percent.
> Is there any other way to test? Improve?
>

Sponsored Links