How to create an output layer consisting of a single output neuron using patternnet?

Hi!

I have a 2-class classification problem, for which I want to use an output vector that consists of only two values, 1 and 0. When I use patternnet for creating the network and then train it using train function, all the patterns are classified to only one class, giving me only 50% accuracy (I have equal patterns from both classes). Is there any solution? Thanks.

[Andrew Stirn]

I have the same issue! I tried removing 'removeconstantrows' but without success. I can get some improvement by changing over to 'trainlm'. 'trainscg' worked with this same data and produced good results in 2014b. For me, thats the only work around. Can we get a Mathworks response?

[Andrew Stirn]

To be clear this issue only exists for me with 2015a.

[Greg Heath]

Write MATHWORKS directly via mathworks.com.