fullseye

graph_edge_consensus — GRAPH population op

• データ種: table → table
• 呼び出し: import fullseye as fs; fs.ledger.graph_edge_consensus(adjs, ordered=False, n_null=0, swaps_per_edge=5, seed=0) (実装を直接呼ぶなら import graphinv; graphinv.graph_edge_consensus(adjs, ordered=False, n_null=0, swaps_per_edge=5, seed=0)、台帳から引くなら opsgraph.get("graph_edge_consensus"))

使い方

How many individuals share each connection — K wiring matrices on one node order.

adjs is either a list of K square matrices or a table {name: matrix} (its insertion order is the individual order), all on the same node order (align the cell names first; a node absent from an individual is a zero row and column). Weights (synapse counts) are kept for the synapse shares; presence is > 0. Self-loops are dropped.

Returns a dict with:

• k (K), n, names, edges_per_individual and union (edges seen in at least one individual)
• occupancy_hist — h[c] = number of edges present in exactly c individuals, c = 0..K (h[0] is always 0: only the union is counted)
• jaccard — the K x K pairwise overlap |Ei ∩ Ej| / |Ei ∪ Ej|
• synapse_share — for each c, the fraction of all synapses (summed over individuals) on edges of occupancy c (with a 0/1 input: the edge share)
• with ordered=True (individuals listed in developmental order): the union is split by its presence pattern along the order into stable (present in all), added (absent, then present to the end: 0..01..1), lost (1..10..0) and flicker (anything else). The four sum to union.
• with n_null >= 2: every individual is rewired independently by degree-preserving swaps (each node’s in/out degree kept exactly, checked per sample, fail-closed) and the occupancy histogram is re-counted. Reports occupancy_null_mean / occupancy_null_sd per c and shared_all_ratio (edges present in all K, observed over the null mean; inf when the null never shares one, so shared_all_null_max is reported too): the chance level of “the same connection in every animal” given only each animal’s degrees.

Identities (the tests are built on these, not on fitted numbers): sum_c c * h[c] == sum_i |E_i|; sum_c h[c] == union; K identical inputs give h[K] == union and every Jaccard == 1; the Jaccard matrix is symmetric with a unit diagonal; stable + added + lost + flicker == union; synapse_share sums to 1.

Raises ValueError: adjs not a list / table; fewer than 2 individuals; matrices of different sizes; any refusal of :func:graph_degree_summary (non-square, negative, non-finite, string / bool / complex / masked input); an individual with no edges; n_null equal to 1 (no spread) or negative; swaps_per_edge < 1.

参考(サンプルデータ・文献)

• サンプルデータ カタログ(DL URL / ライセンス) — 2-D は skimage.data(BSD/public)+ 合成、3-D は実データ源(Stanford/PDS 等)の DL URL。
• 演算子の来歴・参考文献 — この op 族の元になった研究/手法の出典。
• アルゴリズムの正典(著者・年)と用途は上記ファミリ使い方ガイドに記載。

実行できる例(この op を実際に呼ぶ検証済みサンプル)

• poc_connectome_across_decades — py -3.11 examples/poc_connectome_across_decades.py
• poc_connectome_across_worms — py -3.11 examples/poc_connectome_across_worms.py

型が繋がる次の op(table を入力に取れる)

graph_core_persistence · tree_morphometry · tree_sholl · tree_run_length

同カテゴリ(population)

graph_strength_growth · graph_core_persistence

---

Provenance: graphinv.py — GRAPH operator registry. この per-op ノートは tools/opdocs.py md が自動生成(手編集しない)。

© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.

This site is open source. Improve this page.