NEWS

landgraph 0.0.4 (2026-10-30)

Documentation now matches terradish. nu is an effective-information value that the user supplies, not a SNP, locus, or feature count. The help for cov_from_biallelic() explains the count-standardized scale and why terradish divides by N[1] only with equal sample sizes. The help and the vignette describe the "gower" default for diagonal and how missing data and NA coding are handled.
dist_from_cov() keeps the dimnames of its input and drops every attribute except diagonal and centered, so terradish can still flag a "within" diagonal.
fst_from_biallelic() names its result by rownames(Y) and accepts N in any form accepted by cov_from_biallelic(). A locus-length N vector was previously recycled down the columns and misaligned; that case now gives different (correct) values. No other results change.
dist_from_biallelic() gains N = NULL and passes ... to cov_from_biallelic().
deme_graph() requires k to be a single positive whole number for neighbours = "knn", and documents its coords element.
deme_graph() objects are no longer described as interchangeable with terradish fitting graphs, and edge_gradient() no longer cites a terradish directed model that does not exist.
DESCRIPTION cites the implemented methods, and citation("landgraph") reports the installed version.

Default to the Gower diagonal for grouped genetic covariance. This preserves site centering and avoids mixing covariance and within-population variance scales. Explicit within and legacy auto choices remain available.

Record covariance diagonal construction and centering metadata without changing numerical values. Distinguish pooled allele-frequency centering with unequal sample sizes and centering before diagonal replacement.
Clarify rare-variant weighting, unequal sampling variance, and the use of Gower covariances for Wishart models. FST ratio estimates are not Wishart responses.

Initial release. Shared landscape-genetic primitives extracted from terradish so that landscape-genetic network methods stand on a common, dependency-light base.
Graph: deme_graph() builds a lightweight deme/landscape graph (vertex coordinates + undirected edge list) from coordinates, with Delaunay, k-nearest- neighbor, or lattice (rook/queen) adjacency. The result is class c("landgraph", "terradish_graph").
Genetic covariance / distance: cov_from_biallelic(), cov_from_genetic_data(), fst_from_biallelic(), dist_from_cov(), dist_from_biallelic().
Directional edge covariates: edge_gradient() (gradient of a scalar potential) and edge_flow() (projection of a vector flow field, carrying a curl component).
cov_from_biallelic() applies the per-locus pooled frequency correctly per column (fixes a column-major recycling error that affected multi-deme covariances).