hardlim returns 0.5 and 1 !!

I have a binarry classification problem with 0-1 classes. I used 3layered neural network with "hardlim" as the activation function of output layer and "tansig" for hidden and input layer. I am wonderfull why my predicted output contains 0.5 and 1 instead of 0-1 !!

Does MATLAB automatically normalizes data when we use it's standard function for NN? Or I should manually normalize data?

[Greg Heath]

Woefully insufficient information.

How many hidden layers??

% One is ALWAYS sufficient but may not be practical

Greg

[Morteza]

My network contains 3 layers:

1-one input layer (with 25 neurons) and tansigfunction.

2-one hidden layer (with 13 neurons) and tansigfunction.

3- one output layer (just one neuron). Activation function hardlim

My data set has about 44000 samples, 25 features, all data are non-negative. one output is 0-1 (binary). I used MATLAB nftool with advanced script tool. Also "trainlm" as training algorithm. However, the output layer is hardlim, but my estimated outputs are 0.5 and 1 instead of 0 and 1.

[Morteza]

Any suggestion?

[Hugo Fournier]

Same problem here. Have you found a solution?

[Morteza]

Unfortunately No. We can replace 0.5 by 0.

Answer 1

[Garrett Beard]

'net variable'.output.processFcns = {}

By default, the neural network class applies mapminmax to the output vector. Remove the normalization and you will get [0, 1] for outputs.