Adding mixed type in privacy metrics silently fails with no feedback

[npatki]

Problem Description

If I accidentally include mixed types in a Privacy Metric calculation, the metric returns a nan with no indication as to whether there was an error or what made the computation fail.

Using the same demo data from the User Guide, I tried to run the following code:

>>> NumericalLR.compute(real_data, synthetic_data
        key_fields=['second_perc', 'mba_perc', 'gender'],
        sensitive_fields=['degree_perc'])
nan

(This is happening because gender is a categorical column being used in a Numerical metric. It correctly works if I remove gender.)

Expected Outcome

In sdv-dev/SDMetrics#134, I filed an issue about supporting mixed types.

However even if we don't support mixed types, we should validate the inputs and throw a nice, descriptive Error to the user.