Add multi label content

## Due Date
*To be completed by:* YYYY-MM-DD

Expand on this description:
Get to know the range of each input feature, get to know its variance and mean. And get to know how features will correlate with the labels. If at all possible, make sure that you would be able to do what you want the model to do. Also try to know if some of the output labels are going to be corelated, as that may influence other choices down the line.

Next: Training is done in batches. Usually you want your batches to be balanced, so your model gets a somewhat acccurate view of the world in each training step. Imagine being blind and wanting to "see" an elephant. If all you ever touched were it's legs, you might have a rather wrong picture of it. Now, in multi label classification you have usually the problem that you have no way of ever balancing your training examples, as there will always be some labels that appear more often than others, so you will have to make sure that each batch at least has a single example for all labels. As far as I know DL4J doesn't yet have anything that will help you with that.

For your model, you should use either BinaryXENT Loss with Sigmoid activation or MultiLabel Loss with your choice of activation. Both of them can deal with multi label loss, but BinaryXENT expects uncorrelated outputs while MultiLabel Loss should be able to find correlations better.

You should be able to use Multi Class Evaluation methods here (evaluateROCMultiClass) to see if you are learning anything at all.

Credit to @treo

## Description
*Write a short description of what needs to be done.*

## Assignees
*Please ensure you have assigned at least one person to this issue. Include any authors and reviewers required.*

L’issue non indica alcun file di documentazione, test o punto di ingresso. Inizia chiarendo quale pagina debba trattare la classificazione multilabel e quali DL4J APIs o esempi rientrino nell’ambito; il lavoro è concluso quando il contenuto concordato è documentato nella posizione selezionata.

Scritto dal modello di indicizzazione a partire dal testo della issue.