Improvements:
- Rich console mode no longer suppresses warnings, and warnings are now handled in a nicer way and can be toggled with SynthEval's new
show_warningsargument.
Bug fixes:
- The FIO metric would fail when only a single outcome variable was provided, due to a mistaken variable reassignment.
- The MIA metric would fail when the size of the synthetic data was less than the size of the holdout data. We made a catch because keeping a large holdout set may be intended.
- PCA is not really meant to work with fewer rows than columns; for the PCA metric, this would cause an error from sklearn. We made a catch and now produce a warning when this is the case.