Changelog • mlearning

In mlearning(), when data is not provided, model.args$data is now evaluated in the environment of the formula, if any, instead of parent.frame() (the previous behavior is used as default fallback).

Documentation is refactored using Roxygen2 and considerably enhanced.
All camelCase function names now have their equivalence in snake_case, e.g., mlRforest -> ml_rforest(), or confusionImage() -> confusion_image() in order to adapt to the coding preferences of the user.
GitHub repository modernized ans Hex logo added.

The mlRpart() function implements rpart::rpart() for using decision trees.

The description is extended.
A {pkgdown} site is added.

mlKnn() is implemented for K-nearest neighbors.
Several adjustments were required for compatibility with R 4.2.0 (it is not allowed any more to use vectors > 1 with || and &&).

When predict() was applied to an mlearning object build with full formula (not the short one var ~ .), if the dependent variable was not in newdata =, an error message was raised (although this variable is not necessary at this point). Bug identified by Damien Dumont, and corrected.

In mlSvm.formula(), arguments scale=, type=, kernel= and classwt= were not correctly used. Corrected.

In mlLvq() providing size = or prior = led to an lvq object not found message. Corrected.

Sometimes, data was not found (e.g., when called inside a {learnr} tutorial).
In mlearning(), data is forced with as.data.frame() (tibbles are not supported internally).
In the mlXXX() function, it was not possible to indicate something like mlLda(data = iris, Species ~ .). Solved by adding train = argument in mlXXX().
In summary.confusion() produced an error if more than one type = was provided.

NEWS.md file added. Repository moved to GitHub.