amosblomqvist
f5edc7f0a5
Louvain only ever found the reasoning chains (the graph is disconnected by nature — every connected component became its own cluster) and edge-orphans could never join anything. Clusters are now RELATED TOPICS: propositions grouped by meaning. - Embedder (src/server/embeddings.ts): in-process transformers.js v4 running EmbeddingGemma-300M q8 (MTEB 68.4, ~300MB one-time download), MRL-truncated to 512d, "title: none | text:" prefix on 100% of inputs, L2-normalized so cosine = dot. Model/dim mismatch voids the ledger (full re-embed). Injected at the composition root — the agent layer never imports the server layer (architecture law). - Ledger (data/graph/embeddings.json): model + dim + vectors + the persisted cluster partition. Reclustering runs at the END of every embedding batch (post-merge hook, both front doors) — never per read (the agglomerative pass is ~1.5s at 338 nodes). - Clustering (src/core/semantic-clusters.ts): hand-rolled average-linkage agglomerative over cosine (npm HDBSCAN is an abandoned alpha). Stragglers stay singletons by default; attachStragglers is the all-in toggle. Threshold 0.60 swept on the real graph: 26 topics, 74/90 edge-orphans land in real topics, 10 stragglers. - Consumers (snapshot, settle, namer, replay) read the persisted partition — one cluster concept everywhere; core-survival label matching unchanged (20/26 topics kept labels). - Louvain retired: detectCommunities + graphology* deps deleted (graphology-louvain had leaked into the client bundle since the clusters wave); scripts/bench-graph.ts (dead FA2 bench) removed. - scripts/embed-graph.ts: backfill/cold-start (338 nodes in 150s including the one-time model download).
86 lines
3.7 KiB
TypeScript
86 lines
3.7 KiB
TypeScript
/**
|
|
* Community detection tests (vision §1, DECIDED 2026-09-13) — pure
|
|
* structural Louvain over follows+supports, deterministic via seed.
|
|
* Assertions are on SHAPE and MEMBER SETS, never community indices
|
|
* (indices reshuffle; member sets are what labelings key to).
|
|
*/
|
|
import { strict as assert } from "node:assert";
|
|
import { describe, test } from "node:test";
|
|
import { matchClusterLabelings, MIN_CLUSTER_SIZE, type Community } from "../src/core/clusters.ts";
|
|
import { clusterHue } from "../components/cluster-geometry.ts";
|
|
import type { ClusterLabeling } from "../src/core/contracts.ts";
|
|
|
|
const ids = (n: number, prefix = "n") => [...Array(n).keys()].map((i) => `${prefix}${i}`);
|
|
|
|
/** A clique on the given ids. */
|
|
const clique = (nodeIds: string[]) =>
|
|
nodeIds.flatMap((a, i) => nodeIds.slice(i + 1).map((b) => ({ from: a, to: b })));
|
|
|
|
const labeling = (id: string, core: string[], createdAt = "2026-09-01T00:00:00.000Z"): ClusterLabeling => ({
|
|
id,
|
|
label: `label-${id}`,
|
|
core,
|
|
createdAt,
|
|
});
|
|
|
|
describe("matchClusterLabelings", () => {
|
|
test("claims a labeling while at least half its core survives in the community", () => {
|
|
const community: Community = { members: ["a", "b", "c", "d", "e"] };
|
|
const records = [labeling("clu_1", ["a", "b", "x", "y"])]; // 2 of 4 survive — exactly half
|
|
const [claim] = matchClusterLabelings([community], records);
|
|
assert.equal(claim?.id, "clu_1");
|
|
// One fewer survivor: under half → relabel.
|
|
const [dead] = matchClusterLabelings([{ members: ["a", "c", "d", "e", "f"] }], records);
|
|
assert.equal(dead, null);
|
|
});
|
|
|
|
test("single successor: a split core keeps its label in only one community", () => {
|
|
const records = [labeling("clu_1", ["a", "b", "c", "d"])];
|
|
const left: Community = { members: ["a", "b", "z"] }; // 2 of 4 — half
|
|
const right: Community = { members: ["c", "d", "w"] }; // 2 of 4 — half
|
|
// Tie: larger overlap wins; equal → both eligible, greedy takes first come.
|
|
const claims = matchClusterLabelings([left, right], records);
|
|
const claimed = claims.filter((c) => c !== null);
|
|
assert.equal(claimed.length, 1);
|
|
assert.equal(claimed[0]?.id, "clu_1");
|
|
});
|
|
|
|
test("greedy: the best-overlapping community wins the record", () => {
|
|
const records = [labeling("clu_1", ["a", "b", "c", "d"])];
|
|
const weak: Community = { members: ["a", "b", "q", "r", "s"] }; // 2/4
|
|
const strong: Community = { members: ["a", "b", "c", "x", "y"] }; // 3/4
|
|
const claims = matchClusterLabelings([weak, strong], records);
|
|
assert.equal(claims[0], null);
|
|
assert.equal(claims[1]?.id, "clu_1");
|
|
});
|
|
|
|
test("one record per community; newest record breaks ties", () => {
|
|
const oldR = labeling("clu_old", ["a", "b"], "2026-09-01T00:00:00.000Z");
|
|
const newR = labeling("clu_new", ["a", "b"], "2026-09-02T00:00:00.000Z");
|
|
const community: Community = { members: ["a", "b", "c"] };
|
|
const [claim] = matchClusterLabelings([community], [oldR, newR]);
|
|
assert.equal(claim?.id, "clu_new");
|
|
});
|
|
|
|
test("no labelings: every community unclaimed; empty communities array ok", () => {
|
|
assert.deepEqual(matchClusterLabelings([{ members: ["a", "b"] }], []), [null]);
|
|
assert.deepEqual(matchClusterLabelings([], [labeling("clu_1", ["a"])]), []);
|
|
});
|
|
|
|
test("MIN_CLUSTER_SIZE is 3 (the render/naming threshold)", () => {
|
|
assert.equal(MIN_CLUSTER_SIZE, 3);
|
|
});
|
|
});
|
|
|
|
describe("clusterHue", () => {
|
|
test("is stable per id and spreads", () => {
|
|
// Same id → same hue, always; different ids → usually different hues.
|
|
assert.equal(clusterHue("cluster-a"), clusterHue("cluster-a"));
|
|
const hues = new Set<number>();
|
|
for (let i = 0; i < 40; i++) hues.add(clusterHue(`cluster-${i}`));
|
|
assert.ok(hues.size > 20, `expected spread, got ${hues.size} distinct hues for 40 ids`);
|
|
for (const h of hues) assert.ok(h >= 0 && h < 360);
|
|
});
|
|
});
|
|
|