Adapt evaluation to the different data types

[csala]

The current evaluate function does not distinguish between the different field types and, most importantly, includes the id fields in the evaluation.

Id fields are not modeled and can be very different between the sampled and the real data, which means that they can produce a huge and meaningless impact on the produced scores.

The evaluate function should be changed to expect a metadata argument as input and adapt the evaluation to the different data types.