Enh/nn/cnn whale make network structure more flexible
The dimensions of the inner layers were hardcoded in the CNNWhale:create_net_structure()
method, forcing the images to have a certain shape (60 x 20 pixels).
I made changes (commit 54e0fafd) to calculate these dimensions from the input_shape
instance attribute, so now images of any size should work (issue #4 (closed))
Other aspects of the network structure are still predefined (e.g.: the number of convolutional layer and the number of channels in each of these layers).
We could probably make all of this more flexible. Alternatively, one could simply inherit from the class and override this method with a custom structure.
I'll leave the WIP tag for now, while we decide which approach to take.