How can I use sigmoid layer at output for multilabel classification?

layers = [ ...
    sequenceInputLayer(11890)
    bilstmLayer(100,'OutputMode','last')
    fullyConnectedLayer(60)
    sigmoidLayer
    weightedClassificationLayer(classWeights)
]

I tried to use sigmoid activation function at output node for multilable classification, but it says "softmaxlayer is left out" whether classificationLayer is custom or not.

how to use sigmoid layer at output for classification?

[Ankit Pasi]

Hi,

Did you end up solving your issue? I have a similar problem where the network graph does not accept sigmoid as the final layer and throws random errors. Useless actually compared to pytorch and tensorflow..

[Firas Abou Dalla]

Did you solve this problem?

Answer 1

[Pratyush Roy]

The following link might be helpful:

https://www.mathworks.com/matlabcentral/answers/409466-how-can-i-use-a-sigmoid-output-layer-in-an-lstm-or-cnn-network

sigmoidLayer has been introduced in MATLAB 2020b. The link to the documentation is given below:

https://www.mathworks.com/help/deeplearning/ref/nnet.cnn.layer.sigmoidlayer.html

[Tobe freeman]

Thanks for these links, they were a useful step foward.

But I was not able to get too much further with them. The link to sigmoidLayer contains the following Tip:

"...To use the sigmoid layer for binary or multilabel classification problems, create a custom binary cross-entropy loss output layer or use a custom training loop.

>> Create a custom binary cross-entropy loss output layer or use a custom training loop

How? Also, keep in mind that if an expert provides you with a choice between two options they are likely signalling that neither of them actually work. But I digress.