Changes in version 2024.6.19

- fix ubsan issue by only doing &vec[0] if there is at least one element.

Changes in version 2023.6.15

- if(requireNamespace("penaltyLearning")) in examples.
- rm Suggests lattice.
- examples for min AUM and max AUC line search.

Changes in version 2023.6.14

- setDTthreads(1) in aum_linear_model_cv example for CRAN.

Changes in version 2023.5.12

- aum_linear_model/cv defaults to maxIterations="min.aum".

Changes in version 2023.4.7

- aum_line_search now accepts maxIterations="min.aum" or "max.auc".

Changes in version 2023.4.4

- removed some code to check faster on CRAN.

Changes in version 2023.4.3

- reformat NEWS, for CRAN.
- faster examples for aum_linear_model_cv and aum_line_search.

Changes in version 2023.3.31

- Line search C++ code review, exclude speed and line search vignettes from CRAN.

Changes in version 2023.2.21

- aum_diffs_penalty now correctly handles denominator="rate" (previously there could be problems if there are more examples in error table than in pred.name.vec).

Changes in version 2023.2.15

- Ties are now handled correctly in line search. aum_line_search$line_search_result contains new columns intersections and intervals, which are greater than 1 if there are any ties. Also new column q.size contains the number of step sizes in the queue at each iteration.

Changes in version 2023.2.3

- aum_line_search$line_search_result contains new columns aucAtStep and aucAfterStep.

Changes in version 2022.12.8

- bugfix in aum_diffs_penalty(denominator="rate"), which previously incorrectly normalized for each example, but now correctly normalizes based on the total error over all examples.

Changes in version 2022.12.7

- aum_linear_model and aum_linear_model_cv with plot and predict methods.

Changes in version 2022.12.1

- aum_line_search function with plot method.

Changes in version 2022.2.7

- Add arXiv link to DESCRIPTION, clarify outputs in aum_diffs.

Changes in version 2022.2.3

- Remove un-necessary C++ code, just keep aum_sort and interface.

Changes in version 2022.1.27

- rename test file.

Changes in version 2021.9.23

- aum_sort.cpp: fix read out of bound when err_N=1, use std::sort instead of qsort.

Changes in version 2021.3.9

- vignette comparing logistic regression and other loss functions to aum minimization.

Changes in version 2021.3.2

- Use qsort (standard C) instead of qsort_r (not standard).

Changes in version 2021.2.20

- error checking for min.lambda values input to aum_diffs_penalty.
- new aum_sort_interface C++ function (faster), older function renamed to aum_map_interface, separate source and header files.

Changes in version 2021.2.16

- aum supports names for predictions (copied to row names of derivative_mat).

Changes in version 2021.2.15

- more C++ error checking / tests.
- vignette comparing speed with penaltyLearning::ROChange.

Changes in version 2021.2.14

- aum_diffs, aum_diffs_binary, aum_diffs_penalty for creating error diffs data frame required for input to aum.
- aum_errors for converting aum_diffs to canonical error functions (which start at fp=0 and end at fn=0). plot.aum_diffs uses this to show a default plot of the error functions.
- fn.not.zero example data taken from feaure-learning-benchmark.

Changes in version 2021.2.12

- First draft.