NEWS

gcf 0.1.1 (2026-09-30)

Fixed platform-dependent geocomplexity features (*_gc_k23 / *_P_gc). spdep::knearneigh() switches to a kd-tree search when the optional dbscan package is installed, and the two searches break ties at the k-th neighbour distance differently. On gridded data such ties are common, so the neighbour sets (and hence the geocomplexity values) depended on the machine; this made the reference tests fail on some CRAN check flavours. The geocomplexity neighbours now always use spdep's brute-force search, whose ties go to the lower row index, which is also what the paper's results were computed with. Data without distance ties (for example the bio_grid case study) are unaffected.
New test pinning the tie-breaking rule of the geocomplexity neighbours.
Now requires spdep >= 1.1-7 (the first version with the use_kd_tree argument of knearneigh()).

Initial CRAN release.
GCF variable generation: gcf_field(), with step functions gcf_psi() (11 spatial-pattern operators over buffer radii), gcf_zx() (buffer-wise neighbourhood quantiles) and gcf_reduce() (functional reduction to the X/P/D candidate variables).
Variable selection: gcf_select() (random forest impurity importance with spatial-block stability resampling and group voting) and gcf_blocks(). gcf_select() seeds the random number generator only when seed is supplied (seed = 1 reproduces the paper) and restores the caller's generator state on exit.
Datasets: sim_grid (paper simulation, 900 grid cells) and bio_grid (south-western Australia plant species richness, 6229 grid cells).
Vignette: "The GCF workflow: from spatial variables to better predictions".
Implements Song (2026) doi:10.1080/13658816.2026.2729719; feature generation and selection reproduce the paper's reference implementation to machine precision.