netrics

• The scaled distances now divide by the largest finite distance, and unreachable nodes stay infinite, with a message
• Fixed node_by_diversity() to not include ego when measuring alter diversity (closes #39)
• Nodes without alters now return NA
• Fixed twomodes in node_by_information(), net_by_spatial() etc by replacing to_multilevel() with manynet::to_onemode() to create square matrices
• Added membership= to net_by_diversity() for mean diversity by group weighted by size
• Fixed mode_by_eigenvector() erroring on multilevel networks (closes #23)
• Fixed about 40 measures, memberships, marks, and motifs erroring on, or miscounting, CSSs
• Now combine reports into locally aggregated structures
• Tie-level functions still return one value per report
• Added variant= to node_by_transitivity() for Barrat, Onnela, or Zhang-Horvath weighted clustering
• Weighted networks now use Barrat's coefficient by default; variant = "watts" keeps the unweighted result
• Improved node_by_flow()
• Fixed ignoring tie weights and erroring on two-mode networks
• Added from= and to= for the maximum flow from or to a given node
• net_by_adhesion() now reads tie weights as capacities, returning weighted minimum cut
• Signed networks now read by their positive ties alone
• Added net_by_divergence() for networks' Hamming, Jaccard, or portrait divergence
• ideal= takes a network or any manynet::create_*() or manynet::generate_*() function
• Falls back to portrait divergence where the networks' nodes differ
• Added net_by_linkdensity() for the partition density of a tie membership

Memberships

• Fixed nodetie memberships returning NA for groups beyond the 702nd
• Fixed single community algorithms erroring on a signed 'stocnet' (closes #32)
• Added node_in_faction() for the k groups, of any size, that come closest to separate cliques
• Uses a tabu search, as the UCINET 'Factions' routine does
• variant= scores the fit by "hamming" (default), "phi", or "modularity"
• Added search= to node_in_block() for searching by "tabu" (default) or "iterated"
• The tabu search lets positions take any size, where the iterated search held them near equal
• Partitions may therefore differ from earlier versions; pass search = "iterated" to reproduce them
• Added tie_in_community() for link communities (Ahn et al. 2010), which group ties rather than nodes (closes #20)
• A node belongs to every community one of its ties does, so nodes' communities can overlap
• Fixed twomodes in node_in_partition(), node_in_regular() etc by replacing to_multilevel() with manynet::to_onemode() to create square matrices
• Improved the equivalence functions to compare nodes' profiles once, rather than twice, thanks @Kaladani (closes #29)
• node_in_equivalence(), node_in_structural(), node_in_motif(), and node_in_automorphic() now cluster the proximities between nodes' profiles directly, as documented and as sna::equiv.clust() does
• Memberships may therefore differ from earlier versions; pass distance = "euclidean" to reproduce them
• Added proximity= to choose how profiles are compared, by default "pearson", from any measure in manynet::to_proximity(), e.g. "cosine", "ruzicka", or "overlap"
• distance= is now NULL by default; naming a distance compares nodes a second time on their dissimilarities, as before
• Unbounded proximities such as "count" or "crossmin" are subtracted from the largest proximity, with a message, rather than returning negative merge heights
• A square census is now compared as the profile matrix it is, rather than as a one-mode network
• node_in_regular() now clusters the regularity_*() similarity directly
• Deprecated cluster = "cosine" and cluster_cosine() in favour of proximity = "cosine"
• cluster_hierarchical() now also returns the proximities it clustered as $proximity
• Fixed node_in_automorphic() erroring with k = "elbow" on weighted or unconnected networks
• Nodes that cannot reach each other are now treated as one step further apart than the most distant nodes that can, with a message
• Improved node_in_roulette(), with some implications for node_in_block()
• Now uses an iterated local search to improve the partition, rather than a single pass of greedy moves
• Weak moves now kept only if they improve the partition
• Strong perturbation now restarts from the best partition
• Improved speed by several times
• Improved groups= to accept a vector of group sizes
• Added attribute= and balance= to mix groups by a node attribute
• Added decay= to discount ties from earlier waves or times, by default uniform
• Now stops with a message on a two-mode network
• Split the documentation of the non-hierarchical community algorithms into three pages, by the idea they share
• ?member_community_modular, ?member_community_spread, and ?member_community_partition replace ?member_community_non
• ?member_community now compares all the algorithms in one table

Methods

• Added search_tabu() and search_iterated() for finding membership that minimises a cost

Motifs

• Fixed net_x_stability() miscounting the ties that changed between networks
• == bound after *, so it counted every absent tie, including self-ties, as changed
• Now also counts undirected and weighted ties only once
• Fixed net_x_correlation() requiring a second network (closes #27)
• Now correlates consecutive waves of a longitudinal network
• Fixed net_x_change() and net_x_stability() on networks with missing ties
• Fixed node_x_path() on weighted networks to return shortest path lengths (Opsahl et al. 2010)
• Was inverting tie weights, so every dyad without a tie was infinitely far apart
• Two-mode weighted networks now return a square matrix
• Improved node_x_alters(), thanks @RWKrause (closes #39)
• Added direction= for out, in, reciprocated, or all (default) alters
• Added WeightedSum, each alter's value times the strength of the tie to it
• Alters with missing values are now left out of the summary
• Nodes without alters of known value now return NA, also if categorical
• Two-mode attributes on one mode only are now read at distance two
• Signed ties now count by their magnitude
• Added node_x_percolation() for overlapping communities by clique percolation (Palla et al. 2005) (closes #21)