Changelog#

1.2.0#

Added Bayesian ranking and selection utilities
Fixed covariance computation in MarginalRanking and SimultaneousRanking
Wrote a faster version of MarginalRanking
Implemented low memory algorithm for pairiwse hypothesis testing
Sped up ppf and rvs methods of nonparametric by caching the CDF
Improved interpolation of nonparametric so the PDF will always be weakly positive

Added robust empirical Bayes confidence intervals from Armstrong et al., 2020 for bayes.Normal.
Added q-values for hypothesis testing with correct false discovery rates.
Implemented a faster algorithm for pairwise hypothesis testing.

Added nonparametric empirical Bayes
Collapsed all normal prior Bayesian estimators (classic, maximum likelihood, and James-Stein) into a single Normal class
Created a separate Improper class for Bayesian models with an improper prior
Moved projection confidence intervals from RQU to a separate condfidence_set.ConfidenceSet class
Created the confidence_set module
Added non-parametric, mixture, and joint Distributions
Added significance conditional analysis
Renamed RQU to the more expressive RankCondition

Tested on Python3.8
Added Holm-Bonferroni correction
Added ability to specify columns in ModeBase.from_csv
Fixed a column selection bug in ProjectionResults.conf_int
Fixed a bug in Bayes results base: allows for rank matrix when estimating one parameter
Fixed a bug in RQU.get_distributions; the projection CIs didn’t line up with the column indices in the previous version

Added a truncnorm distribution with exponential tilting
Added reconstruction plot methods for Bayesian models
Added utilities for Wasserstein distances as a measure of reconstruction error
Added robustness to linear empirical Bayes likelihood optimization
Added empirical Bayes fitting to minimize Wasserstein reconstruction error
Changed the prior covariance parameter for empirical and hierarchical Bayes to represent the prior standard deviation rather than the prior variance
Fixed a bug in rank matrix calculation