1.3.0

What's Changed

• Add option to define order for CI plots (see #39)
• Add support for force_mode with bayesian statistics (see #33)
• Add effect size in relation to next-best result to result data frame and output options to latex_table (see #36)
• Add posteriors in relation to next-best results for bayesian statistics (see #36)
• Add function to visualize posteriors and decisions as heatmaps (see #40, thanks @fabiancpl)
• Fix working with pivor tables (see #37)
• Fix string replacement targets in bayesian table header (see #34, thanks @e10e3)
• Fix the chart (see #25) and add posterior plots to flow charts for bayesian options

1.2.1

• Fix pandas warnings (see #28, #32, thanks @e10e3)
• Fix problems with scikit-learn example (see #29, thanks @e10e3)
• Fix crash when latex tables are generated multiple times (see #30)
• Fix crash when a named or MultiIndex is used (see #16)
• Add missing output for Bayesian analysis (see #31, thanks @e10e3)

1.2.0

• Extends the Bayesian analysis by exposing the random seed to enable reproducible results (see #21, thanks to @Daniel300000 for the idea an PR)
• Extends results with sample created as part of the Bayesian analysis to, e.g., enable visualizations (see #20, thanks to @Daniel300000 for the idea and PR)
• Fixes issue with wrongly ordered outputs when the textual description of significant results for two populations are created (see #19)
• Updated external dependencies and dropped support for Python 3.7

1.1.3

• Fixes #10, #14 regarding broken ordering in CD diagrams (thanks to gerald4, ZhongLIFR, vnmabus, AlexandreAbraham for their input and PavelKromer for the PR that led to the fix)
• Adds new feature for forcing to use parametric/non-parametric approach requested in #8

1.1.2

• Fixes #5: Replaced usage of deprecated median_absolute_deviation from scipi (thanks to d-diaz for fixing)
• Fixes #7: Fixes wrong ordering of p-vals of Shapiro-Wilk test in some corner cases (thanks to delsner for reporting)

1.1.1

This releases fixes a bug in the summary of bayesian analysis that was not correct, in case the results were all equal and/or inconclusive.

1.1.0

This release adds the ability to conduct bayesian analysis instead of frequentist analysis.

• Use bayesian analysis with autorank(data, approach='bayesian')
• Report generation for bayesian analysis
• New example for bayesian analysis

1.0.1

Some small fixes and improvements

• Minimal versions for matplotlib, scipy, statsmodels, and pandas defined
• Beautified string output of results
• Added a warning in case the matplotlib option text.usetex is True
• Fixed some typos

1.0.0

Feature complete and stable version of Autorank.

Future changes will be downward compatible.

0.3.2

This release contains the following changes:

• import is no simplified to from autorank import autorank
• Allow to define if ranking is ascending or descending