David Charte

Provide an API or similar which specifies all possible argument values for each parameter. - [ ] Layers - [ ] Networks - [ ] Autoencoder models - [ ]...

Currently, the evaluated metric shows up as `[[2]]` in the result.

Allow to tie encoder's and decoder's weights

When defining a network through an integer vector, decide on layer activations instead of leaving all as "linear"

Use `fit_generator` to dynamically augment data

https://www.icml.cc/2012/papers/910.pdf

`evaluate` functions are always verbose. [keras docs](https://www.rdocumentation.org/packages/keras/versions/0.3.8/topics/evaluate)