DO i need to process the unit8 data when i use "trainautoencoder" function?

I want to use the SAE net to processing my image data. But my data is unit8 format,and should I use "mapminmax" funtion to make the data betwen[-1,1]? Thanks all.

Answer 1

[Massimo Zanetti]

The trainAutoencoder accepts inputs of type single, double (and cell).

So, just cast your data as double

A=double(A);

and then run the function.

[Liang Seacom]

Thanks,and should I use "mapminmax" funtion to make the data betwen[-1,1] like the normal neural networks?

[Walter Roberson]

SAE does not require that, and defaults to not doing that, according to the documentation.

That said... you would have been much further along by now if you had just gone ahead and tried both with and without the mapminmax .

[Liang Seacom]

I have tried the both ways,but the result is very different when i use the the SAE networks to predict the test data.So I have the question:need to process the data between[-1,1]?

[Walter Roberson]

Which way gives you better results for your purpose?

[Liang Seacom]

funny result,the data*10 is the best. DATA*10 > DATA > mapminmax(DATA);