Include additional measures of drift

[sgreenbury]

Also consider adding:

• [x] Permutation test
  • [x] Test with GMLVQ 8 features pipeline
• [ ] Discrete KS test to cover all variable types
• [x] CSTest from the the sdv package for categorical data
• [x] Other non-parametric options such as Mann-Whitney U
• [x] Separate Chi-Square per feature for category-like fetaures
• [ ] Metrics based on e.g. propensity score matching from causal inference. Well-matched samples are typically considered to have per variable differences of Δz_mean < 0.1 where z_mean is standardized mean for a sample. Nice example with balance plots in step 4 here.
• [x] Refactor logistic detector to call sklearn directly and get typical ML scores: ROC score, APR, F1, etc
• [x] Normalized GMM scores: check how normalization can make likelihood based scores (e.g. here)

[myyong]

The first pass for this is in this branch.

[sgreenbury]

Closing for now given #20.