NEWS

BayesPPR 0.2.0.9000

Fixed corrupted basis bookkeeping when resuming a fit whose last draw contained a categorical ridge function (bppr_resume() failed outright).
Fixed stale cached basis functions in predict.bppr(), which silently returned incorrect predictions when consecutive draws shared the same number of ridge functions but used different features for a categorical ridge.
Fixed fitting failures when X contains a constant column and ncol(X) <= 3.
bppr() now accepts a data frame for X, as documented.
bppr_pca() and predict.bppr_pca() no longer fail when parallel::detectCores() returns NA, and now validate par_type.
pca_setup() now bounds n_pc by min(nrow(Y), ncol(Y)) instead of the maximum, and handles prop_var = 1.
predict.bppr_pca() no longer fails when newdata has a single row.
Added validation of idx_use, newdata dimensions, w_feat, w_n_act, and missing values in X/y.