what is difference between fprop and upward_pass ?

[HyunHo Kim]

I use Softmax in MLP.

When I test with MLP.fprop(), upward_pass() function not found in Softmax class.

upward_pass() is implemented just in Autoencoder and RBM.

Although upward_pass() is not in Softmax class, fprop() is in this.

and I think that we can make upward_pass() by copying fprop().

Is it right??

What is difference between fprop and upward_pass()??

[Devon Hjelm]

Are you trying to use Softmax in RBM or Autoencoders? There's a specific Softmax layer in the DBM version of RBM. Autoencoders IIRC specify the nonlinearity as a string.

I'd be careful about trying to insert an MLP Layer into another model: that layer may be missing other critical features for the model. Maybe eventually the idea of Layer and nonlinearity will become more ubiquitous and one wont have to worry about a MLP Layer or RBM layer or whatever...

[Panagiotis Sourtzinos]

Hello Kim,

have you find the answer to your question? 

I am trying to load a pretrained MLP and I am facing the same problem as you did back then check the link (https://groups.google.com/forum/#!topic/pylearn-users/-67yFG5sS_U).

Thanks in advance

Panos