- Ensure we only use quantiles that are part of the features - Added link to doc describing the algorithm
2.1.0
- Added support for features with zero variance - Improved performance and less memory usage
2.0.0
- Switched to the procedure described in Blume, 2012 but normalized by a feature's stddev and extended to multiple quantiles - Improved quantile estimation - Now returning `pandas.DataFrame`s consistently
1.3.0
- allow for a generic model type - performance improvements - extend and enhance tests
1.2.0
- made featureimpact Python3 ready
1.1.0
- it is now assumed that predict() does not change its input - performance improvements - improved documentation