BrainNetTest 0.2.2

• Clarified that global_test() uses permutation inference and does not assume standard-normal scaling when its normalization constant is one.
• New function global_test() runs the permutation test for a difference between populations on its own: it returns the observed statistic T, the proportion of label permutations with a smaller statistic (the p-value), and the permutation null distribution, as an object of class "global_test" with a print() method. Until now this test was only available as the first step of identify_critical_links(). It takes the populations, n_permutations and an optional seed; the normalisation constant a is fixed at 1, since the p-value does not depend on it. The statistic is on the scale of compute_test_statistic(), and with the same populations, n_permutations and seed the p-value is exactly the one identify_critical_links() records. The input must be binary, as for identify_critical_links().
• The permutation machinery shared by global_test() and identify_critical_links() (edge-wise decomposition of T and the matrix-product null) now lives in one set of internal helpers. Results are unchanged.
• identify_critical_links() now returns an object of class "critical_links" with print(), summary() and plot() methods, so the result reports itself instead of having to be picked apart by hand. print() gives the populations compared, the global-test result and the most significant critical edges; summary() adds the call, the settings the analysis ran with, and the node-level ranking from get_critical_nodes(); plot() is a method interface to plot_critical_edges() and takes the original populations as its second argument.
• The returned object gained three components alongside the existing three: p_value (the permutation p-value of the global test on the unmodified data, which was previously computed and discarded), n_edges and settings, plus the matched call. critical_edges, edges_removed and modified_populations keep their names, positions and contents, so existing code that extracts them is unaffected.
• Corrected the DOI given for Fraiman and Fraiman (2018) throughout the package. The previous value, 10.1038/s41598-018-21688-0, resolves to an unrelated article; the correct one is 10.1038/s41598-018-23152-5.
• Count arguments are validated. generate_category_graphs(0.7) used to fail with attempt to select less than one element in integerOneIndex, generate_random_graph(0.7) and generate_community_graph(n_nodes = 0.7) returned a degenerate 0 x 0 matrix, and a fractional n_graphs was silently truncated. These now raise an error that names the offending argument. The same applies to batch_size, n_permutations, alpha and a in identify_critical_links().
• identify_critical_links() now checks that populations is a list of non-empty lists of equally sized square binary matrices, as its documentation has always required. Weighted matrices were previously passed to fisher.test(), which rounded the resulting counts and returned a meaningless ranking.
• compute_edge_pvalues() no longer subscripts out of bounds for single-node networks.
• compute_test_statistic() returns an unnamed scalar; it used to carry the name of the first population. rank_edges() resets the row names of its output, so the ranking reads 1, 2, 3, … rather than the positions the edges occupied before sorting.
• Internal simplifications following review feedback: lengths() in place of sapply(x, length), and Reduce("+", x) / length(x) in place of an accumulation loop in compute_central_graph().
• .Rbuildignore now excludes the manuscript sources, so they are no longer included in the source tarball.

BrainNetTest 0.2.1

• compute_edge_pvalues() now clamps every returned p-value to the valid probability range [0, 1]. On platforms built without extended (long double) precision, exact tests such as fisher.test() can return a value fractionally greater than 1 due to floating-point rounding, which caused a test failure under CRAN’s noLD check. (Reported by the CRAN team.)
• Added a regression test that mocks the underlying test to return an out-of-range p-value, so the clamping is verified on every platform rather than only on noLD builds.

BrainNetTest 0.2.0

• Removed plot_graph_with_communities() and plot_graphs_grid() to streamline the API. The recommended plotting function is plot_critical_edges(), which produces a multi-panel visualisation of the analysis results.
• Removed unused Suggests: ggplotify, gridExtra.
• Removed unused imports: grDevices::rainbow, graphics::legend.

BrainNetTest 0.1.0

• Initial CRAN submission.
• Implements the L1-distance ANOVA test for populations of brain networks of Fraiman and Fraiman (2018) doi:10.1038/s41598-018-23152-5.
• Fast permutation procedure for identifying critical edges via a prefix-sum decomposition of the test statistic, reducing complexity from O(K * B * |E| * m) to O(B * |E| * m).
• Helpers to generate synthetic community-structured graphs and to visualise brain networks with communities.