How to use trained network to predict or classify test data

[zhish...@gmail.com]

In the package and DEMO, for example, the 3D MINST demo, I only saw Train function to train a network, but did not see a Classify or Predict function to 

use the trained network net to classify or predict a test data.  

For example,  after I get  "net"  after calling Train function,  how do I call a function like Classify as MATLAB Deep Learning Toolbox does :

   Predicted_Class  =  Classify(trainedNet, TestData)   ?

Thank you!

David

[Hagay Garty]

Hi, you can read in the documentation file on how to classify a single sample. Search for 'classify a single sample' in section 5.b

For your convenience, you can use the below code:

Batch       = zeros([net.layers{1}.properties.sizeOut   1]);

Batch(:,:,:,:,1) = sample_to_test;

net = feedForward(net, Batch , 1);

[~,netClassification]  = max(squeeze(net.layers{end}.outs.activation));

fprintf('sample classified as %d\n', netClassification);

 

[steven...@gmail.com]

Yes, yes, it works very well!

Thank you!

How about a regression case ?

Steven

[Hagay Garty]

Happy to hear it works well :)

For regression, you can take the last layer outputs as your prediction. 

i.e res = net.layers{end}.outs.activation;