Tools for Post-Processing Class Probability Estimates

Models can be improved by post-processing class probabilities, by: recalibration, conversion to hard probabilities, assessment of equivocal zones, and other activities. 'probably' contains tools for conducting these operations.


Travis buildstatus Codecov testcoverage Lifecycle:experimental

Introduction

probably contains tools to facilitate activities such as:

  • Conversion of probabilities to discrete class predictions.

  • Investigating and estimating optimal probability thresholds.

  • Inclusion of equivocal zones where the probabilities are too uncertain to report a prediction.

Installation

You can install probably from CRAN with:

install.packages("probably")

You can install the development version of probably from GitHub with:

devtools::install_github("topepo/probably")

Examples

Good places to look for examples of using probably are the vignettes.

  • vignette("equivocal-zones", "probably") discusses the new class_pred class that probably provides for working with equivocal zones.

  • vignette("where-to-use", "probably") discusses how probably fits in with the rest of the tidymodels ecosystem, and provides an example of optimizing class probability thresholds.

News

probably 0.0.2

Bug fixes

  • A failing test relying on the R 3.6 change to sample() has been corrected.

  • An rlang warning in threshold_perf() has been fixed.

  • A small R 3.1 issue with vctrs has been fixed.

probably 0.0.1

  • First release

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.

install.packages("probably")

0.0.3 by Davis Vaughan, 9 days ago


https://github.com/tidymodels/probably/


Report a bug at https://github.com/tidymodels/probably/issues


Browse source code at https://github.com/cran/probably


Authors: Max Kuhn [aut] , Davis Vaughan [aut, cre] , RStudio [cph]


Documentation:   PDF Manual  


GPL-2 license


Imports dplyr, generics, rlang, tidyselect, vctrs, yardstick

Suggests covr, ggplot2, knitr, parsnip, rmarkdown, rsample, testthat


See at CRAN