Binary Classification with patternnet - wrong output

Hi together,

what i want to do, is to train a neural network (patternnet) classificator. Inputs are the size 10000x7, so 10.000 samples and 7 inputs.

Output is logical (10.000x1).

My problem is, if i use the patternnet function with a HiddenLayerSize > [], my output is no longer binary, than in a range between [0,1].

Here is my code:

    hiddenLayerSize = [10,10,10];
    net = patternnet(hiddenLayerSize, method);
      net = configure(net, in_train', out_train');
      
      net.input.processFcns = {'removeconstantrows','mapminmax'};
      net.output.processFcns = {'removeconstantrows','mapminmax'};
      
      net.trainParam.goal = gl;
      net.divideFcn = divider;  % Divide data randomly
      net.divideMode = divMod;  % Divide up every sample
      
      net.divideParam.trainRatio = x/100;
      net.divideParam.valRatio = y/100;
      net.divideParam.testRatio = z/100;
      
      net.performFcn = pfFcn; 
      
      net.trainParam.epochs = epoch;
      
      net.Layers{:}.transferFcn = transfer; 
      
      net = train(net,in_train',out_train'

What can i do?

Thanks :)

Answer 1

[Anshika Chaurasia]

Hello Vincent,

It is my understanding that after training the neural network, the ouput is in range between [0,1] and you want output as logical i.e., {0,1}.

For getting logical output you could apply threshold i.e., if output >= 0.5 then output = 1 or else output = 0.

% Train the Network
[net,tr] = train(net,inputs,targets);
% Test the Network
outputs = net(inputs);

logical_output = zeros(1,10000);
logical_output(outputs >= 0.5) = 1; % applying threshold

[Vincent Kelber]

Hi Anshika,

yes you are absolutely right. And that is one oportunity, yes. But the goal i want to achieve (to simplify the model) is not covered with that. Is there no other possibility to do that inside the model?

Thanks, Vinc