Multiple PyBrain Output Modules

Question

I would like to train a network with several output levels.

in->hidden->out 1
          ->out 2

Is it possible? If so, how do I set up the datasets and trainer for training.

Answer 1

When you study sharing your output to have multiple SoftMax areas, you can use PartialSoftmaxLayerprovided by PyBrain.

Note that it is limited to slices of the same length, but its code may inspire you if you need a custom output layer:

https://github.com/pybrain/pybrain/blob/master/pybrain/structure/modules/softmax.py

Answer 2

No. You can have several hidden layers, for example

in → hidden 1 → hidden 2 → out

( ).

, , , , , .

Answer 3

, , .

in --> hidden --> concatenate([out1, out2])

, , , , .

, , , , - , .

Potential solutions to this problem may include defining a custom error metric (for example, using the option of weighted quadratic error or weighted cross-entropy) and / or standardizing the two output datasets so that they exist on a common scale.