Is mapstd the same as Zcore ? what is the reverse function?

Hi All

for my work , I should apply zcore to my target matrix , that is a 4x1055 matrix , the zcore I apply is like :

[ tc meanT stdT] = zscore(tc)

and for rolling back I do :

an = meanT + stdT'* an';

I just wanted to know if the mapstd works the same and if the above is correct ?

thank you very much

[farzad]

I tried

an = meanT + stdT* tc';

and

an = meanT .+ stdT* tc';

but unsuccessful

Answer 1

[Star Strider]

They are not the same. The most notable difference is that mapstd operates on rows, while zscore operates on columns.

If you are using neural networks, use mapstd for consistency with other neural network functions.

[farzad]

thank you so much

just shall you please tell me , if it works on the rows ,then should the rows be the samples or it doesn't matter ?

can you please tell me , that how to roll back the mapstd after training the net ?

[Star Strider]

My pleasure.

It appears to operate on the rows of your data array.

According the mapstd documentation, you can reverse the mapping. See the documentation under ‘Definitions’ for details.

[farzad]

thank you very much

one point in the doc not clear :

x1 = [1 2 4; 1 1 1; 3 2 2; 0 0 0]
[y1,PS] = mapstd(x1)

Next, apply the same processing settings to new values.

x2 = [5 2 3; 1 1 1; 6 7 3; 0 0 0]
y2 = mapstd('apply',x2,PS)

what is PS (settings??) and y1 and y2 should be produced both for every mapping procedure or they are 2 separate methods?

[Star Strider]

The documentation is not clear on ‘PS’, and I haven’t looked at the mapstd code to see what it is. I assume it is a structure that contains the mapping information used to standardise the input, and necessary to reverse-map the results. The standardisation mapping needs to be done separately for every input and target, and the associated ‘PS’ saved for each (with a slightly different name) to reverse-map the associated result.

You will likely have to experiment with it with your data and network to thoroughly understand it in that context.

[farzad]

thank you for your comment

well I found some explanations here , but didn't get it precisely where proff,heath has responded

[farzad]

and there comes one question , why we can't know who has written this documentation so we could ask him directly what did you mean by that ??!!

[Star Strider]

My pleasure.

My impression (from having worked with neural nets on and off for the past 20 or so years, with my own FORTRAN code and with MATLAB) is that they are at least as much art as science.

[Greg Heath]

I am confused about whether or not you need clarification.

 help mapstd
 doc mapstd
 type mapstd

The latter should have the name of the coder. However, I can probably answer any further clear-cut questions.

Greg

[farzad]

thank you dear Greg , with the other tutorial, I am done , and got my answer ,thank you very much