SoftMax and newff

[killian Janod]

Good Evening,

I'm writting because I want to use a Softmax transformation for my Multilayered Perceptron. But I always have an error because there is no deriv function for softmax.

Can someone explain me why the Softmax doesn't have a deriv function and if there is a way to make it work with newff ?

Thanks in advance

[Evgeny Zuev]

Hi!
SoftMax unsupported backpropagation train process (MLP) and  have not derivative. SoftMax may use in Kohonen layer/network or LVQ network. Why you need it?

[Waat]

Thanks for your answer, you confirmed what I thought.
I wanted to redo an experiment a coworker did. He used a softmax layer to let the MLP's last layer infer a probability distribution.
I know he did it with pylearn2, but I wanted to use neurolab.
I think the derivation of the softmax layer is done that way : http://stats.stackexchange.com/questions/79454/softmax-layer-in-a-neural-network

I might look deeper into that later and look for a way to implement a deriv function for the Softmax-layer if I have enough time.

Thanks again.

[Evgeny Zuev]

Ok. I will see this later too

01 дек. 2014 г. 15:07 пользователь "Waat" <k.j...@gmail.com> написал:

--
Вы получили это сообщение, поскольку подписаны на группу "NeuroLab".
Чтобы отменить подписку на эту группу и больше не получать от нее сообщения, отправьте письмо на электронный адрес py-neurolab...@googlegroups.com.
Чтобы настроить другие параметры, перейдите по ссылке https://groups.google.com/d/optout.

[Evgeny Zuev]

I add deriv method to SoftMax in svn