NEWS

---

kumquat 1.1.0 (2026-06-22)

• Added kernel_weights() to weight perturbations by their distance from the point of interest using LIME's exponential kernel.
• fit_local_model() now labels surrogate predictions with the levels of the pred column when it is a factor or character vector. Previously the labels followed the order of class_names, so passing e.g. unique(data$class) could swap them. class_names is now only used for numeric (0/1) predictions.
• fit_local_model() gains a weights argument, and kumquat() gains weighted, kernel_width and dist_fun arguments to fit the local model with proximity weights.

• Added new test datasets and unit tests.
• Updated internal lime calculation methods.

kumquat 1.0.0

• Initial CRAN submission.