From transformations and normalization to encoding and transformation

We are moving from the current approach to one where we have:

• Transformation: where we can have log10, tofloat, and also normalizations!! The user is the responsible of providing them in the correct order, but by default default normalizations are added as the last element. NONE of this change the dimension.
• Encoding: in this case, we change the dimension, from parametric to ML space. For the moment, one-hot and label-smoothing.