causantic 0.2.0

package/dist/clusters/cluster-refresh.js

@@ -0,0 +1,187 @@
+/**
+ * LLM-based cluster description refresh.
+ * Uses Claude API to generate human-readable descriptions for clusters.
+ */
+import Anthropic from '@anthropic-ai/sdk';
+import { getConfig } from '../config/memory-config.js';
+import { getClusterById, upsertCluster, getStaleClusters } from '../storage/cluster-store.js';
+import { getChunksByIds } from '../storage/chunk-store.js';
+import { createLogger } from '../utils/logger.js';
+import { createSecretStore } from '../utils/secret-store.js';
+const log = createLogger('cluster-refresh');
+/**
+ * Rate limiter for API calls.
+ */
+class RateLimiter {
+    lastCall = 0;
+    minIntervalMs;
+    constructor(callsPerMinute) {
+        this.minIntervalMs = (60 * 1000) / callsPerMinute;
+    }
+    async wait() {
+        const now = Date.now();
+        const elapsed = now - this.lastCall;
+        if (elapsed < this.minIntervalMs) {
+            await new Promise((resolve) => setTimeout(resolve, this.minIntervalMs - elapsed));
+        }
+        this.lastCall = Date.now();
+    }
+}
+/**
+ * Cluster refresher for generating descriptions using Claude.
+ */
+export class ClusterRefresher {
+    client = null;
+    rateLimiter;
+    config = getConfig();
+    constructor() {
+        this.rateLimiter = new RateLimiter(this.config.refreshRateLimitPerMin);
+    }
+    /**
+     * Initialize the Anthropic client.
+     * Checks keychain for API key if not set in environment.
+     */
+    async getClient() {
+        if (!this.client) {
+            // Check if API key is in environment
+            if (!process.env.ANTHROPIC_API_KEY) {
+                // Try to load from keychain
+                const store = createSecretStore();
+                const storedKey = await store.get('anthropic-api-key');
+                if (storedKey) {
+                    process.env.ANTHROPIC_API_KEY = storedKey;
+                }
+            }
+            if (!process.env.ANTHROPIC_API_KEY) {
+                throw new Error('No Anthropic API key found. Set ANTHROPIC_API_KEY environment variable ' +
+                    'or run "causantic config set-key anthropic-api-key" to store in keychain.');
+            }
+            this.client = new Anthropic();
+        }
+        return this.client;
+    }
+    /**
+     * Refresh a single cluster's description.
+     */
+    async refreshCluster(clusterId, options = {}) {
+        const startTime = Date.now();
+        const { model = this.config.clusterRefreshModel, maxExemplars = 3, maxTokensPerChunk = 500 } = options;
+        const cluster = getClusterById(clusterId);
+        if (!cluster) {
+            throw new Error(`Cluster not found: ${clusterId}`);
+        }
+        // Get exemplar chunks
+        const exemplarIds = cluster.exemplarIds.slice(0, maxExemplars);
+        const exemplars = getChunksByIds(exemplarIds);
+        if (exemplars.length === 0) {
+            return {
+                clusterId,
+                name: cluster.name ?? 'Empty Cluster',
+                description: 'No exemplar chunks available.',
+                durationMs: Date.now() - startTime,
+            };
+        }
+        // Build prompt
+        const prompt = buildRefreshPrompt(exemplars, maxTokensPerChunk);
+        // Rate limit
+        await this.rateLimiter.wait();
+        // Call Claude API
+        const client = await this.getClient();
+        const response = await client.messages.create({
+            model,
+            max_tokens: 200,
+            messages: [{ role: 'user', content: prompt }],
+        });
+        // Parse response
+        const text = response.content[0].type === 'text' ? response.content[0].text : '';
+        const { name, description } = parseRefreshResponse(text);
+        // Update cluster
+        upsertCluster({
+            id: clusterId,
+            name,
+            description,
+        });
+        return {
+            clusterId,
+            name,
+            description,
+            durationMs: Date.now() - startTime,
+        };
+    }
+    /**
+     * Refresh all stale clusters.
+     */
+    async refreshStaleClusters(options = {}) {
+        const { maxAgeMs = 24 * 60 * 60 * 1000, onProgress } = options; // Default: 24 hours
+        const staleClusters = getStaleClusters(maxAgeMs);
+        const results = [];
+        for (const cluster of staleClusters) {
+            try {
+                const result = await this.refreshCluster(cluster.id, options);
+                results.push(result);
+                onProgress?.(results.length, staleClusters.length);
+            }
+            catch (error) {
+                log.error(`Failed to refresh cluster ${cluster.id}`, { error: error.message });
+            }
+        }
+        return results;
+    }
+    /**
+     * Refresh all clusters (force refresh).
+     */
+    async refreshAllClusters(options = {}) {
+        const { onProgress } = options;
+        const staleClusters = getStaleClusters(0); // Get all
+        const results = [];
+        for (const cluster of staleClusters) {
+            try {
+                const result = await this.refreshCluster(cluster.id, options);
+                results.push(result);
+                onProgress?.(results.length, staleClusters.length);
+            }
+            catch (error) {
+                log.error(`Failed to refresh cluster ${cluster.id}`, { error: error.message });
+            }
+        }
+        return results;
+    }
+}
+/**
+ * Build prompt for cluster description generation.
+ */
+function buildRefreshPrompt(exemplars, maxTokensPerChunk) {
+    const truncatedExemplars = exemplars.map((e) => {
+        const content = e.content;
+        // Rough token estimate: 4 chars per token
+        const maxChars = maxTokensPerChunk * 4;
+        if (content.length > maxChars) {
+            return content.slice(0, maxChars) + '\n...[truncated]';
+        }
+        return content;
+    });
+    return `Analyze these conversation excerpts from the same topic cluster and generate:
+1. A short name (2-5 words) that captures the main theme
+2. A brief description (1-2 sentences) of what this cluster is about
+
+Excerpts:
+${truncatedExemplars.map((e, i) => `--- Excerpt ${i + 1} ---\n${e}`).join('\n\n')}
+
+Respond in this exact format:
+Name: [short name]
+Description: [brief description]`;
+}
+/**
+ * Parse the LLM response to extract name and description.
+ */
+function parseRefreshResponse(text) {
+    const nameMatch = text.match(/Name:\s*(.+?)(?:\n|$)/i);
+    const descMatch = text.match(/Description:\s*(.+?)(?:\n\n|$)/is);
+    return {
+        name: nameMatch?.[1]?.trim() ?? 'Unnamed Cluster',
+        description: descMatch?.[1]?.trim() ?? text.trim(),
+    };
+}
+// Singleton instance
+export const clusterRefresher = new ClusterRefresher();
+//# sourceMappingURL=cluster-refresh.js.map

package/dist/clusters/cluster-refresh.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"cluster-refresh.js","sourceRoot":"","sources":["../../src/clusters/cluster-refresh.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,SAAS,MAAM,mBAAmB,CAAC;AAC1C,OAAO,EAAE,SAAS,EAAE,MAAM,4BAA4B,CAAC;AACvD,OAAO,EAAE,cAAc,EAAE,aAAa,EAAE,gBAAgB,EAAE,MAAM,6BAA6B,CAAC;AAC9F,OAAO,EAAE,cAAc,EAAE,MAAM,2BAA2B,CAAC;AAE3D,OAAO,EAAE,YAAY,EAAE,MAAM,oBAAoB,CAAC;AAClD,OAAO,EAAE,iBAAiB,EAAE,MAAM,0BAA0B,CAAC;AAE7D,MAAM,GAAG,GAAG,YAAY,CAAC,iBAAiB,CAAC,CAAC;AA0B5C;;GAEG;AACH,MAAM,WAAW;IACP,QAAQ,GAAG,CAAC,CAAC;IACb,aAAa,CAAS;IAE9B,YAAY,cAAsB;QAChC,IAAI,CAAC,aAAa,GAAG,CAAC,EAAE,GAAG,IAAI,CAAC,GAAG,cAAc,CAAC;IACpD,CAAC;IAED,KAAK,CAAC,IAAI;QACR,MAAM,GAAG,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QACvB,MAAM,OAAO,GAAG,GAAG,GAAG,IAAI,CAAC,QAAQ,CAAC;QACpC,IAAI,OAAO,GAAG,IAAI,CAAC,aAAa,EAAE,CAAC;YACjC,MAAM,IAAI,OAAO,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,UAAU,CAAC,OAAO,EAAE,IAAI,CAAC,aAAa,GAAG,OAAO,CAAC,CAAC,CAAC;QACpF,CAAC;QACD,IAAI,CAAC,QAAQ,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;IAC7B,CAAC;CACF;AAED;;GAEG;AACH,MAAM,OAAO,gBAAgB;IACnB,MAAM,GAAqB,IAAI,CAAC;IAChC,WAAW,CAAc;IACzB,MAAM,GAAG,SAAS,EAAE,CAAC;IAE7B;QACE,IAAI,CAAC,WAAW,GAAG,IAAI,WAAW,CAAC,IAAI,CAAC,MAAM,CAAC,sBAAsB,CAAC,CAAC;IACzE,CAAC;IAED;;;OAGG;IACK,KAAK,CAAC,SAAS;QACrB,IAAI,CAAC,IAAI,CAAC,MAAM,EAAE,CAAC;YACjB,qCAAqC;YACrC,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,iBAAiB,EAAE,CAAC;gBACnC,4BAA4B;gBAC5B,MAAM,KAAK,GAAG,iBAAiB,EAAE,CAAC;gBAClC,MAAM,SAAS,GAAG,MAAM,KAAK,CAAC,GAAG,CAAC,mBAAmB,CAAC,CAAC;gBACvD,IAAI,SAAS,EAAE,CAAC;oBACd,OAAO,CAAC,GAAG,CAAC,iBAAiB,GAAG,SAAS,CAAC;gBAC5C,CAAC;YACH,CAAC;YAED,IAAI,CAAC,OAAO,CAAC,GAAG,CAAC,iBAAiB,EAAE,CAAC;gBACnC,MAAM,IAAI,KAAK,CACb,yEAAyE;oBACzE,2EAA2E,CAC5E,CAAC;YACJ,CAAC;YAED,IAAI,CAAC,MAAM,GAAG,IAAI,SAAS,EAAE,CAAC;QAChC,CAAC;QACD,OAAO,IAAI,CAAC,MAAM,CAAC;IACrB,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,cAAc,CAClB,SAAiB,EACjB,UAA0B,EAAE;QAE5B,MAAM,SAAS,GAAG,IAAI,CAAC,GAAG,EAAE,CAAC;QAC7B,MAAM,EAAE,KAAK,GAAG,IAAI,CAAC,MAAM,CAAC,mBAAmB,EAAE,YAAY,GAAG,CAAC,EAAE,iBAAiB,GAAG,GAAG,EAAE,GAC1F,OAAO,CAAC;QAEV,MAAM,OAAO,GAAG,cAAc,CAAC,SAAS,CAAC,CAAC;QAC1C,IAAI,CAAC,OAAO,EAAE,CAAC;YACb,MAAM,IAAI,KAAK,CAAC,sBAAsB,SAAS,EAAE,CAAC,CAAC;QACrD,CAAC;QAED,sBAAsB;QACtB,MAAM,WAAW,GAAG,OAAO,CAAC,WAAW,CAAC,KAAK,CAAC,CAAC,EAAE,YAAY,CAAC,CAAC;QAC/D,MAAM,SAAS,GAAG,cAAc,CAAC,WAAW,CAAC,CAAC;QAE9C,IAAI,SAAS,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAC3B,OAAO;gBACL,SAAS;gBACT,IAAI,EAAE,OAAO,CAAC,IAAI,IAAI,eAAe;gBACrC,WAAW,EAAE,+BAA+B;gBAC5C,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;aACnC,CAAC;QACJ,CAAC;QAED,eAAe;QACf,MAAM,MAAM,GAAG,kBAAkB,CAAC,SAAS,EAAE,iBAAiB,CAAC,CAAC;QAEhE,aAAa;QACb,MAAM,IAAI,CAAC,WAAW,CAAC,IAAI,EAAE,CAAC;QAE9B,kBAAkB;QAClB,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,SAAS,EAAE,CAAC;QACtC,MAAM,QAAQ,GAAG,MAAM,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC;YAC5C,KAAK;YACL,UAAU,EAAE,GAAG;YACf,QAAQ,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC;SAC9C,CAAC,CAAC;QAEH,iBAAiB;QACjB,MAAM,IAAI,GACR,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,CAAC,CAAC,QAAQ,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC;QACtE,MAAM,EAAE,IAAI,EAAE,WAAW,EAAE,GAAG,oBAAoB,CAAC,IAAI,CAAC,CAAC;QAEzD,iBAAiB;QACjB,aAAa,CAAC;YACZ,EAAE,EAAE,SAAS;YACb,IAAI;YACJ,WAAW;SACZ,CAAC,CAAC;QAEH,OAAO;YACL,SAAS;YACT,IAAI;YACJ,WAAW;YACX,UAAU,EAAE,IAAI,CAAC,GAAG,EAAE,GAAG,SAAS;SACnC,CAAC;IACJ,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,oBAAoB,CACxB,UAAkD,EAAE;QAEpD,MAAM,EAAE,QAAQ,GAAG,EAAE,GAAG,EAAE,GAAG,EAAE,GAAG,IAAI,EAAE,UAAU,EAAE,GAAG,OAAO,CAAC,CAAC,oBAAoB;QAEpF,MAAM,aAAa,GAAG,gBAAgB,CAAC,QAAQ,CAAC,CAAC;QACjD,MAAM,OAAO,GAAoB,EAAE,CAAC;QAEpC,KAAK,MAAM,OAAO,IAAI,aAAa,EAAE,CAAC;YACpC,IAAI,CAAC;gBACH,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,cAAc,CAAC,OAAO,CAAC,EAAE,EAAE,OAAO,CAAC,CAAC;gBAC9D,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBACrB,UAAU,EAAE,CAAC,OAAO,CAAC,MAAM,EAAE,aAAa,CAAC,MAAM,CAAC,CAAC;YACrD,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,GAAG,CAAC,KAAK,CAAC,6BAA6B,OAAO,CAAC,EAAE,EAAE,EAAE,EAAE,KAAK,EAAG,KAAe,CAAC,OAAO,EAAE,CAAC,CAAC;YAC5F,CAAC;QACH,CAAC;QAED,OAAO,OAAO,CAAC;IACjB,CAAC;IAED;;OAEG;IACH,KAAK,CAAC,kBAAkB,CAAC,UAA0B,EAAE;QACnD,MAAM,EAAE,UAAU,EAAE,GAAG,OAAO,CAAC;QAC/B,MAAM,aAAa,GAAG,gBAAgB,CAAC,CAAC,CAAC,CAAC,CAAC,UAAU;QACrD,MAAM,OAAO,GAAoB,EAAE,CAAC;QAEpC,KAAK,MAAM,OAAO,IAAI,aAAa,EAAE,CAAC;YACpC,IAAI,CAAC;gBACH,MAAM,MAAM,GAAG,MAAM,IAAI,CAAC,cAAc,CAAC,OAAO,CAAC,EAAE,EAAE,OAAO,CAAC,CAAC;gBAC9D,OAAO,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;gBACrB,UAAU,EAAE,CAAC,OAAO,CAAC,MAAM,EAAE,aAAa,CAAC,MAAM,CAAC,CAAC;YACrD,CAAC;YAAC,OAAO,KAAK,EAAE,CAAC;gBACf,GAAG,CAAC,KAAK,CAAC,6BAA6B,OAAO,CAAC,EAAE,EAAE,EAAE,EAAE,KAAK,EAAG,KAAe,CAAC,OAAO,EAAE,CAAC,CAAC;YAC5F,CAAC;QACH,CAAC;QAED,OAAO,OAAO,CAAC;IACjB,CAAC;CACF;AAED;;GAEG;AACH,SAAS,kBAAkB,CAAC,SAAwB,EAAE,iBAAyB;IAC7E,MAAM,kBAAkB,GAAG,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE;QAC7C,MAAM,OAAO,GAAG,CAAC,CAAC,OAAO,CAAC;QAC1B,0CAA0C;QAC1C,MAAM,QAAQ,GAAG,iBAAiB,GAAG,CAAC,CAAC;QACvC,IAAI,OAAO,CAAC,MAAM,GAAG,QAAQ,EAAE,CAAC;YAC9B,OAAO,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,QAAQ,CAAC,GAAG,kBAAkB,CAAC;QACzD,CAAC;QACD,OAAO,OAAO,CAAC;IACjB,CAAC,CAAC,CAAC;IAEH,OAAO;;;;;EAKP,kBAAkB,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,eAAe,CAAC,GAAG,CAAC,SAAS,CAAC,EAAE,CAAC,CAAC,IAAI,CAAC,MAAM,CAAC;;;;iCAIhD,CAAC;AAClC,CAAC;AAED;;GAEG;AACH,SAAS,oBAAoB,CAAC,IAAY;IACxC,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,wBAAwB,CAAC,CAAC;IACvD,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,kCAAkC,CAAC,CAAC;IAEjE,OAAO;QACL,IAAI,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,IAAI,iBAAiB;QACjD,WAAW,EAAE,SAAS,EAAE,CAAC,CAAC,CAAC,EAAE,IAAI,EAAE,IAAI,IAAI,CAAC,IAAI,EAAE;KACnD,CAAC;AACJ,CAAC;AAED,qBAAqB;AACrB,MAAM,CAAC,MAAM,gBAAgB,GAAG,IAAI,gBAAgB,EAAE,CAAC"}

package/dist/clusters/hdbscan/cluster-extraction.d.ts

@@ -0,0 +1,26 @@
+/**
+ * Cluster extraction from condensed tree.
+ * Supports EOM (Excess of Mass) and Leaf methods.
+ */
+import type { CondensedTree } from './types.js';
+/**
+ * Extract clusters using the specified method.
+ *
+ * @param tree Condensed cluster tree.
+ * @param method Selection method: 'eom' (default) or 'leaf'.
+ * @returns Array of selected cluster IDs.
+ */
+export declare function extractClusters(tree: CondensedTree, method?: 'eom' | 'leaf'): number[];
+/**
+ * Assign labels to points based on selected clusters.
+ *
+ * @param tree Condensed cluster tree.
+ * @param selectedClusters Selected cluster IDs.
+ * @returns Labels for each point (-1 = noise).
+ */
+export declare function assignLabels(tree: CondensedTree, selectedClusters: number[]): number[];
+/**
+ * Get cluster stability scores.
+ */
+export declare function getClusterStabilities(tree: CondensedTree): Map<number, number>;
+//# sourceMappingURL=cluster-extraction.d.ts.map

package/dist/clusters/hdbscan/cluster-extraction.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"cluster-extraction.d.ts","sourceRoot":"","sources":["../../../src/clusters/hdbscan/cluster-extraction.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,KAAK,EAAE,aAAa,EAAqB,MAAM,YAAY,CAAC;AAEnE;;;;;;GAMG;AACH,wBAAgB,eAAe,CAC7B,IAAI,EAAE,aAAa,EACnB,MAAM,GAAE,KAAK,GAAG,MAAc,GAC7B,MAAM,EAAE,CAyBV;AAgJD;;;;;;GAMG;AACH,wBAAgB,YAAY,CAC1B,IAAI,EAAE,aAAa,EACnB,gBAAgB,EAAE,MAAM,EAAE,GACzB,MAAM,EAAE,CAwBV;AA2CD;;GAEG;AACH,wBAAgB,qBAAqB,CACnC,IAAI,EAAE,aAAa,GAClB,GAAG,CAAC,MAAM,EAAE,MAAM,CAAC,CAUrB"}

package/dist/clusters/hdbscan/cluster-extraction.js

@@ -0,0 +1,232 @@
+/**
+ * Cluster extraction from condensed tree.
+ * Supports EOM (Excess of Mass) and Leaf methods.
+ */
+/**
+ * Extract clusters using the specified method.
+ *
+ * @param tree Condensed cluster tree.
+ * @param method Selection method: 'eom' (default) or 'leaf'.
+ * @returns Array of selected cluster IDs.
+ */
+export function extractClusters(tree, method = 'eom') {
+    if (tree.nodes.size === 0 || tree.root === -1) {
+        return [];
+    }
+    // Get all cluster nodes (id >= numPoints)
+    const clusterNodes = [];
+    for (const [id, node] of tree.nodes) {
+        if (node.isCluster && id >= tree.numPoints) {
+            clusterNodes.push(node);
+        }
+    }
+    if (clusterNodes.length === 0) {
+        return [];
+    }
+    // Compute stability for all cluster nodes
+    computeStability(tree, clusterNodes);
+    if (method === 'leaf') {
+        return extractLeafClusters(clusterNodes);
+    }
+    return extractEOMClusters(tree, clusterNodes);
+}
+/**
+ * Compute stability for cluster nodes.
+ * Stability = sum over member points: (point_lambda_death - cluster_lambda_birth)
+ *
+ * For a cluster to have positive stability, points must "live" within it
+ * (join at lambdaBirth or later, persist until the cluster dies)
+ */
+function computeStability(tree, clusters) {
+    // First, for each leaf cluster, find its directly associated points
+    // A point belongs to a leaf cluster if its lambdaDeath equals the cluster's lambdaBirth
+    // Then for parent clusters, accumulate stability from children plus
+    // any points that joined directly (not via children)
+    // Build a map from cluster to its member points
+    const clusterToPoints = new Map();
+    for (const cluster of clusters) {
+        clusterToPoints.set(cluster.id, new Set());
+    }
+    // Assign each point to the deepest cluster it belongs to
+    // A point belongs to a cluster if its lambdaDeath is at or below the cluster's lambdaBirth
+    for (let i = 0; i < tree.numPoints; i++) {
+        const pointNode = tree.nodes.get(i);
+        if (!pointNode)
+            continue;
+        const pointLambda = pointNode.lambdaDeath;
+        if (pointLambda <= 0 || !isFinite(pointLambda))
+            continue;
+        // Find the cluster(s) this point belongs to
+        // It belongs to a cluster if it joined (lambdaDeath) at or after cluster's lambdaBirth
+        for (const cluster of clusters) {
+            // Point joined this cluster if it was born at this cluster's birth
+            // (i.e., the cluster formed from a component containing this point)
+            if (Math.abs(pointLambda - cluster.lambdaBirth) < 1e-10) {
+                clusterToPoints.get(cluster.id).add(i);
+            }
+        }
+    }
+    // Compute stability for each cluster
+    for (const cluster of clusters) {
+        const points = clusterToPoints.get(cluster.id);
+        let stability = 0;
+        for (const pointId of points) {
+            const pointNode = tree.nodes.get(pointId);
+            // Stability contribution = lambda_death - lambda_birth of cluster
+            // But for these points, lambda_death === cluster.lambdaBirth
+            // So we need to think about this differently
+            // Actually, stability is about how long a point persists in a cluster
+            // A point joins at lambda = cluster.lambdaBirth
+            // It leaves when the cluster dies at cluster.lambdaDeath
+            // But we store point's lambdaDeath as when it joined a cluster
+            // The correct formula is:
+            // For each point in cluster: (min(lambdaDeath_cluster, lambda_point_left) - lambdaBirth_cluster) * 1
+            // Since we're dealing with lambdas (1/distance), higher lambda means closer/denser
+            // Let's use: stability = sum of (point's persistence in this cluster)
+            // A point persists from lambdaBirth to lambdaDeath of cluster (or until it left)
+            const clusterDeath = isFinite(cluster.lambdaDeath) ? cluster.lambdaDeath : 0;
+            const persistence = cluster.lambdaBirth - clusterDeath;
+            if (persistence > 0) {
+                stability += persistence;
+            }
+        }
+        cluster.stability = stability;
+    }
+    // For non-leaf clusters, we might need to add stability from points that
+    // joined after children merged but before the cluster died
+    // This is handled above since we check all points against all clusters
+}
+/**
+ * Extract clusters using Excess of Mass (EOM) method.
+ * Bottom-up selection: compare node stability vs children total.
+ */
+function extractEOMClusters(tree, clusters) {
+    // Sort by lambdaBirth descending (process children before parents)
+    const sorted = [...clusters].sort((a, b) => b.lambdaBirth - a.lambdaBirth);
+    // Track subtree stability (sum of selected children OR own stability)
+    const subtreeStability = new Map();
+    for (const cluster of sorted) {
+        if (cluster.children.length === 0) {
+            // Leaf cluster: select it
+            subtreeStability.set(cluster.id, cluster.stability);
+            cluster.selected = true;
+        }
+        else {
+            // Non-leaf: compare own stability vs children's subtree stability
+            let childrenTotal = 0;
+            for (const childId of cluster.children) {
+                const childStability = subtreeStability.get(childId) ?? 0;
+                childrenTotal += childStability;
+            }
+            if (cluster.stability >= childrenTotal) {
+                // This cluster is more stable - select it, deselect children
+                subtreeStability.set(cluster.id, cluster.stability);
+                cluster.selected = true;
+                for (const childId of cluster.children) {
+                    const child = tree.nodes.get(childId);
+                    if (child)
+                        child.selected = false;
+                }
+            }
+            else {
+                // Children are more stable - keep them selected
+                subtreeStability.set(cluster.id, childrenTotal);
+                cluster.selected = false;
+            }
+        }
+    }
+    return clusters.filter((c) => c.selected).map((c) => c.id);
+}
+/**
+ * Extract clusters using Leaf method.
+ * Select only leaf nodes (finest granularity).
+ */
+function extractLeafClusters(clusters) {
+    const leaves = [];
+    for (const cluster of clusters) {
+        if (cluster.children.length === 0) {
+            leaves.push(cluster.id);
+            cluster.selected = true;
+        }
+        else {
+            cluster.selected = false;
+        }
+    }
+    return leaves;
+}
+/**
+ * Assign labels to points based on selected clusters.
+ *
+ * @param tree Condensed cluster tree.
+ * @param selectedClusters Selected cluster IDs.
+ * @returns Labels for each point (-1 = noise).
+ */
+export function assignLabels(tree, selectedClusters) {
+    const labels = new Array(tree.numPoints).fill(-1);
+    if (selectedClusters.length === 0) {
+        return labels;
+    }
+    // For each selected cluster, find all points that belong to it
+    for (let clusterLabel = 0; clusterLabel < selectedClusters.length; clusterLabel++) {
+        const clusterId = selectedClusters[clusterLabel];
+        const clusterNode = tree.nodes.get(clusterId);
+        if (!clusterNode)
+            continue;
+        // Find all points in this cluster (recursively through children)
+        const points = getAllClusterPoints(tree, clusterId);
+        for (const pointId of points) {
+            if (labels[pointId] === -1) {
+                labels[pointId] = clusterLabel;
+            }
+        }
+    }
+    return labels;
+}
+/**
+ * Get all points that belong to a cluster (including via children).
+ */
+function getAllClusterPoints(tree, clusterId) {
+    const points = [];
+    const clusterNode = tree.nodes.get(clusterId);
+    if (!clusterNode || !clusterNode.isCluster) {
+        return points;
+    }
+    // Find all points with lambdaDeath matching this cluster or any ancestor
+    const clusterLambda = clusterNode.lambdaBirth;
+    for (let i = 0; i < tree.numPoints; i++) {
+        const pointNode = tree.nodes.get(i);
+        if (!pointNode)
+            continue;
+        // Point belongs to this cluster if:
+        // 1. Its lambdaDeath >= cluster's lambdaBirth (joined at or after cluster formed)
+        // 2. Its lambdaDeath <= cluster's lambdaDeath (left at or before cluster died)
+        const clusterDeath = clusterNode.lambdaDeath;
+        const pointLambda = pointNode.lambdaDeath;
+        if (pointLambda >= clusterDeath && pointLambda <= clusterLambda + 1e-10) {
+            points.push(i);
+        }
+    }
+    // Also recurse into children
+    for (const childId of clusterNode.children) {
+        const childPoints = getAllClusterPoints(tree, childId);
+        for (const p of childPoints) {
+            if (!points.includes(p)) {
+                points.push(p);
+            }
+        }
+    }
+    return points;
+}
+/**
+ * Get cluster stability scores.
+ */
+export function getClusterStabilities(tree) {
+    const stabilities = new Map();
+    for (const [id, node] of tree.nodes) {
+        if (node.isCluster && id >= tree.numPoints) {
+            stabilities.set(id, node.stability);
+        }
+    }
+    return stabilities;
+}
+//# sourceMappingURL=cluster-extraction.js.map

package/dist/clusters/hdbscan/cluster-extraction.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"cluster-extraction.js","sourceRoot":"","sources":["../../../src/clusters/hdbscan/cluster-extraction.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAIH;;;;;;GAMG;AACH,MAAM,UAAU,eAAe,CAC7B,IAAmB,EACnB,SAAyB,KAAK;IAE9B,IAAI,IAAI,CAAC,KAAK,CAAC,IAAI,KAAK,CAAC,IAAI,IAAI,CAAC,IAAI,KAAK,CAAC,CAAC,EAAE,CAAC;QAC9C,OAAO,EAAE,CAAC;IACZ,CAAC;IAED,0CAA0C;IAC1C,MAAM,YAAY,GAAwB,EAAE,CAAC;IAC7C,KAAK,MAAM,CAAC,EAAE,EAAE,IAAI,CAAC,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;QACpC,IAAI,IAAI,CAAC,SAAS,IAAI,EAAE,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;YAC3C,YAAY,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC1B,CAAC;IACH,CAAC;IAED,IAAI,YAAY,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC9B,OAAO,EAAE,CAAC;IACZ,CAAC;IAED,0CAA0C;IAC1C,gBAAgB,CAAC,IAAI,EAAE,YAAY,CAAC,CAAC;IAErC,IAAI,MAAM,KAAK,MAAM,EAAE,CAAC;QACtB,OAAO,mBAAmB,CAAC,YAAY,CAAC,CAAC;IAC3C,CAAC;IAED,OAAO,kBAAkB,CAAC,IAAI,EAAE,YAAY,CAAC,CAAC;AAChD,CAAC;AAED;;;;;;GAMG;AACH,SAAS,gBAAgB,CAAC,IAAmB,EAAE,QAA6B;IAC1E,oEAAoE;IACpE,wFAAwF;IAExF,oEAAoE;IACpE,qDAAqD;IAErD,gDAAgD;IAChD,MAAM,eAAe,GAAG,IAAI,GAAG,EAAuB,CAAC;IAEvD,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,eAAe,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,EAAE,IAAI,GAAG,EAAE,CAAC,CAAC;IAC7C,CAAC;IAED,yDAAyD;IACzD,2FAA2F;IAC3F,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,SAAS,EAAE,CAAC,EAAE,EAAE,CAAC;QACxC,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QACpC,IAAI,CAAC,SAAS;YAAE,SAAS;QAEzB,MAAM,WAAW,GAAG,SAAS,CAAC,WAAW,CAAC;QAC1C,IAAI,WAAW,IAAI,CAAC,IAAI,CAAC,QAAQ,CAAC,WAAW,CAAC;YAAE,SAAS;QAEzD,4CAA4C;QAC5C,uFAAuF;QACvF,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;YAC/B,mEAAmE;YACnE,oEAAoE;YACpE,IAAI,IAAI,CAAC,GAAG,CAAC,WAAW,GAAG,OAAO,CAAC,WAAW,CAAC,GAAG,KAAK,EAAE,CAAC;gBACxD,eAAe,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAE,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;YAC1C,CAAC;QACH,CAAC;IACH,CAAC;IAED,qCAAqC;IACrC,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,MAAM,MAAM,GAAG,eAAe,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,CAAE,CAAC;QAChD,IAAI,SAAS,GAAG,CAAC,CAAC;QAElB,KAAK,MAAM,OAAO,IAAI,MAAM,EAAE,CAAC;YAC7B,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,OAAO,CAAE,CAAC;YAC3C,kEAAkE;YAClE,6DAA6D;YAC7D,6CAA6C;YAE7C,sEAAsE;YACtE,gDAAgD;YAChD,yDAAyD;YACzD,+DAA+D;YAE/D,0BAA0B;YAC1B,qGAAqG;YACrG,mFAAmF;YAEnF,sEAAsE;YACtE,iFAAiF;YAEjF,MAAM,YAAY,GAAG,QAAQ,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC;YAC7E,MAAM,WAAW,GAAG,OAAO,CAAC,WAAW,GAAG,YAAY,CAAC;YACvD,IAAI,WAAW,GAAG,CAAC,EAAE,CAAC;gBACpB,SAAS,IAAI,WAAW,CAAC;YAC3B,CAAC;QACH,CAAC;QAED,OAAO,CAAC,SAAS,GAAG,SAAS,CAAC;IAChC,CAAC;IAED,yEAAyE;IACzE,2DAA2D;IAC3D,uEAAuE;AACzE,CAAC;AAED;;;GAGG;AACH,SAAS,kBAAkB,CAAC,IAAmB,EAAE,QAA6B;IAC5E,mEAAmE;IACnE,MAAM,MAAM,GAAG,CAAC,GAAG,QAAQ,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,GAAG,CAAC,CAAC,WAAW,CAAC,CAAC;IAE3E,sEAAsE;IACtE,MAAM,gBAAgB,GAAG,IAAI,GAAG,EAAkB,CAAC;IAEnD,KAAK,MAAM,OAAO,IAAI,MAAM,EAAE,CAAC;QAC7B,IAAI,OAAO,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAClC,0BAA0B;YAC1B,gBAAgB,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,EAAE,OAAO,CAAC,SAAS,CAAC,CAAC;YACpD,OAAO,CAAC,QAAQ,GAAG,IAAI,CAAC;QAC1B,CAAC;aAAM,CAAC;YACN,kEAAkE;YAClE,IAAI,aAAa,GAAG,CAAC,CAAC;YACtB,KAAK,MAAM,OAAO,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;gBACvC,MAAM,cAAc,GAAG,gBAAgB,CAAC,GAAG,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;gBAC1D,aAAa,IAAI,cAAc,CAAC;YAClC,CAAC;YAED,IAAI,OAAO,CAAC,SAAS,IAAI,aAAa,EAAE,CAAC;gBACvC,6DAA6D;gBAC7D,gBAAgB,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,EAAE,OAAO,CAAC,SAAS,CAAC,CAAC;gBACpD,OAAO,CAAC,QAAQ,GAAG,IAAI,CAAC;gBAExB,KAAK,MAAM,OAAO,IAAI,OAAO,CAAC,QAAQ,EAAE,CAAC;oBACvC,MAAM,KAAK,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC;oBACtC,IAAI,KAAK;wBAAE,KAAK,CAAC,QAAQ,GAAG,KAAK,CAAC;gBACpC,CAAC;YACH,CAAC;iBAAM,CAAC;gBACN,gDAAgD;gBAChD,gBAAgB,CAAC,GAAG,CAAC,OAAO,CAAC,EAAE,EAAE,aAAa,CAAC,CAAC;gBAChD,OAAO,CAAC,QAAQ,GAAG,KAAK,CAAC;YAC3B,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;AAC7D,CAAC;AAED;;;GAGG;AACH,SAAS,mBAAmB,CAAC,QAA6B;IACxD,MAAM,MAAM,GAAa,EAAE,CAAC;IAE5B,KAAK,MAAM,OAAO,IAAI,QAAQ,EAAE,CAAC;QAC/B,IAAI,OAAO,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;YAClC,MAAM,CAAC,IAAI,CAAC,OAAO,CAAC,EAAE,CAAC,CAAC;YACxB,OAAO,CAAC,QAAQ,GAAG,IAAI,CAAC;QAC1B,CAAC;aAAM,CAAC;YACN,OAAO,CAAC,QAAQ,GAAG,KAAK,CAAC;QAC3B,CAAC;IACH,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAED;;;;;;GAMG;AACH,MAAM,UAAU,YAAY,CAC1B,IAAmB,EACnB,gBAA0B;IAE1B,MAAM,MAAM,GAAG,IAAI,KAAK,CAAS,IAAI,CAAC,SAAS,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,CAAC;IAE1D,IAAI,gBAAgB,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAClC,OAAO,MAAM,CAAC;IAChB,CAAC;IAED,+DAA+D;IAC/D,KAAK,IAAI,YAAY,GAAG,CAAC,EAAE,YAAY,GAAG,gBAAgB,CAAC,MAAM,EAAE,YAAY,EAAE,EAAE,CAAC;QAClF,MAAM,SAAS,GAAG,gBAAgB,CAAC,YAAY,CAAC,CAAC;QACjD,MAAM,WAAW,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;QAC9C,IAAI,CAAC,WAAW;YAAE,SAAS;QAE3B,iEAAiE;QACjE,MAAM,MAAM,GAAG,mBAAmB,CAAC,IAAI,EAAE,SAAS,CAAC,CAAC;QAEpD,KAAK,MAAM,OAAO,IAAI,MAAM,EAAE,CAAC;YAC7B,IAAI,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,EAAE,CAAC;gBAC3B,MAAM,CAAC,OAAO,CAAC,GAAG,YAAY,CAAC;YACjC,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAED;;GAEG;AACH,SAAS,mBAAmB,CAAC,IAAmB,EAAE,SAAiB;IACjE,MAAM,MAAM,GAAa,EAAE,CAAC;IAC5B,MAAM,WAAW,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,SAAS,CAAC,CAAC;IAC9C,IAAI,CAAC,WAAW,IAAI,CAAC,WAAW,CAAC,SAAS,EAAE,CAAC;QAC3C,OAAO,MAAM,CAAC;IAChB,CAAC;IAED,yEAAyE;IACzE,MAAM,aAAa,GAAG,WAAW,CAAC,WAAW,CAAC;IAE9C,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,IAAI,CAAC,SAAS,EAAE,CAAC,EAAE,EAAE,CAAC;QACxC,MAAM,SAAS,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC;QACpC,IAAI,CAAC,SAAS;YAAE,SAAS;QAEzB,oCAAoC;QACpC,kFAAkF;QAClF,+EAA+E;QAC/E,MAAM,YAAY,GAAG,WAAW,CAAC,WAAW,CAAC;QAC7C,MAAM,WAAW,GAAG,SAAS,CAAC,WAAW,CAAC;QAE1C,IAAI,WAAW,IAAI,YAAY,IAAI,WAAW,IAAI,aAAa,GAAG,KAAK,EAAE,CAAC;YACxE,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACjB,CAAC;IACH,CAAC;IAED,6BAA6B;IAC7B,KAAK,MAAM,OAAO,IAAI,WAAW,CAAC,QAAQ,EAAE,CAAC;QAC3C,MAAM,WAAW,GAAG,mBAAmB,CAAC,IAAI,EAAE,OAAO,CAAC,CAAC;QACvD,KAAK,MAAM,CAAC,IAAI,WAAW,EAAE,CAAC;YAC5B,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,CAAC;gBACxB,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;YACjB,CAAC;QACH,CAAC;IACH,CAAC;IAED,OAAO,MAAM,CAAC;AAChB,CAAC;AAED;;GAEG;AACH,MAAM,UAAU,qBAAqB,CACnC,IAAmB;IAEnB,MAAM,WAAW,GAAG,IAAI,GAAG,EAAkB,CAAC;IAE9C,KAAK,MAAM,CAAC,EAAE,EAAE,IAAI,CAAC,IAAI,IAAI,CAAC,KAAK,EAAE,CAAC;QACpC,IAAI,IAAI,CAAC,SAAS,IAAI,EAAE,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;YAC3C,WAAW,CAAC,GAAG,CAAC,EAAE,EAAE,IAAI,CAAC,SAAS,CAAC,CAAC;QACtC,CAAC;IACH,CAAC;IAED,OAAO,WAAW,CAAC;AACrB,CAAC"}

package/dist/clusters/hdbscan/core-distance-worker.d.ts

@@ -0,0 +1,6 @@
+/**
+ * Worker thread for parallel core distance computation.
+ * This file is loaded by worker_threads.
+ */
+export {};
+//# sourceMappingURL=core-distance-worker.d.ts.map

package/dist/clusters/hdbscan/core-distance-worker.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"core-distance-worker.d.ts","sourceRoot":"","sources":["../../../src/clusters/hdbscan/core-distance-worker.ts"],"names":[],"mappings":"AAAA;;;GAGG"}

package/dist/clusters/hdbscan/core-distance-worker.js

@@ -0,0 +1,54 @@
+/**
+ * Worker thread for parallel core distance computation.
+ * This file is loaded by worker_threads.
+ */
+import { parentPort, workerData } from 'worker_threads';
+import { quickselect } from '../../utils/array-utils.js';
+/**
+ * Compute Euclidean distance between two points.
+ */
+function euclideanDistance(a, b) {
+    let sum = 0;
+    for (let i = 0; i < a.length; i++) {
+        const diff = a[i] - b[i];
+        sum += diff * diff;
+    }
+    return Math.sqrt(sum);
+}
+/**
+ * Compute angular distance between two normalized vectors.
+ */
+function angularDistance(a, b) {
+    let dot = 0;
+    for (let i = 0; i < a.length; i++) {
+        dot += a[i] * b[i];
+    }
+    return 1 - dot;
+}
+/**
+ * Compute core distances for assigned indices.
+ */
+function computeChunk(data) {
+    const { indices, allPoints, k, metric } = data;
+    const distFn = metric === 'euclidean' ? euclideanDistance : angularDistance;
+    const n = allPoints.length;
+    const effectiveK = Math.min(k, n - 1);
+    const coreDistances = [];
+    for (const i of indices) {
+        const distances = [];
+        for (let j = 0; j < n; j++) {
+            if (i !== j) {
+                distances.push(distFn(allPoints[i], allPoints[j]));
+            }
+        }
+        const coreDistance = effectiveK > 0 ? quickselect(distances, effectiveK - 1) : 0;
+        coreDistances.push({ index: i, coreDistance });
+    }
+    return { coreDistances };
+}
+// Run computation and send result back
+if (parentPort && workerData) {
+    const result = computeChunk(workerData);
+    parentPort.postMessage(result);
+}
+//# sourceMappingURL=core-distance-worker.js.map

package/dist/clusters/hdbscan/core-distance-worker.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"core-distance-worker.js","sourceRoot":"","sources":["../../../src/clusters/hdbscan/core-distance-worker.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAEH,OAAO,EAAE,UAAU,EAAE,UAAU,EAAE,MAAM,gBAAgB,CAAC;AAExD,OAAO,EAAE,WAAW,EAAE,MAAM,4BAA4B,CAAC;AAEzD;;GAEG;AACH,SAAS,iBAAiB,CAAC,CAAW,EAAE,CAAW;IACjD,IAAI,GAAG,GAAG,CAAC,CAAC;IACZ,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAClC,MAAM,IAAI,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;QACzB,GAAG,IAAI,IAAI,GAAG,IAAI,CAAC;IACrB,CAAC;IACD,OAAO,IAAI,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;AACxB,CAAC;AAED;;GAEG;AACH,SAAS,eAAe,CAAC,CAAW,EAAE,CAAW;IAC/C,IAAI,GAAG,GAAG,CAAC,CAAC;IACZ,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QAClC,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC;IACrB,CAAC;IACD,OAAO,CAAC,GAAG,GAAG,CAAC;AACjB,CAAC;AAED;;GAEG;AACH,SAAS,YAAY,CAAC,IAA4B;IAChD,MAAM,EAAE,OAAO,EAAE,SAAS,EAAE,CAAC,EAAE,MAAM,EAAE,GAAG,IAAI,CAAC;IAC/C,MAAM,MAAM,GAAG,MAAM,KAAK,WAAW,CAAC,CAAC,CAAC,iBAAiB,CAAC,CAAC,CAAC,eAAe,CAAC;IAC5E,MAAM,CAAC,GAAG,SAAS,CAAC,MAAM,CAAC;IAC3B,MAAM,UAAU,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,CAAC,GAAG,CAAC,CAAC,CAAC;IACtC,MAAM,aAAa,GAAmD,EAAE,CAAC;IAEzE,KAAK,MAAM,CAAC,IAAI,OAAO,EAAE,CAAC;QACxB,MAAM,SAAS,GAAa,EAAE,CAAC;QAC/B,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;YAC3B,IAAI,CAAC,KAAK,CAAC,EAAE,CAAC;gBACZ,SAAS,CAAC,IAAI,CAAC,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,SAAS,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;YACrD,CAAC;QACH,CAAC;QAED,MAAM,YAAY,GAAG,UAAU,GAAG,CAAC,CAAC,CAAC,CAAC,WAAW,CAAC,SAAS,EAAE,UAAU,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QACjF,aAAa,CAAC,IAAI,CAAC,EAAE,KAAK,EAAE,CAAC,EAAE,YAAY,EAAE,CAAC,CAAC;IACjD,CAAC;IAED,OAAO,EAAE,aAAa,EAAE,CAAC;AAC3B,CAAC;AAED,uCAAuC;AACvC,IAAI,UAAU,IAAI,UAAU,EAAE,CAAC;IAC7B,MAAM,MAAM,GAAG,YAAY,CAAC,UAAoC,CAAC,CAAC;IAClE,UAAU,CAAC,WAAW,CAAC,MAAM,CAAC,CAAC;AACjC,CAAC"}

package/dist/clusters/hdbscan/core-distance.d.ts

@@ -0,0 +1,26 @@
+/**
+ * Core distance computation for HDBSCAN.
+ * Core distance = distance to k-th nearest neighbor.
+ */
+/**
+ * Compute core distances for all points.
+ * Core distance is the distance to the k-th nearest neighbor.
+ *
+ * @param points All points (embeddings).
+ * @param k k value (typically minSamples).
+ * @param metric Distance metric.
+ * @param useKDTree Whether to use KD-tree for approximate k-NN.
+ */
+export declare function computeCoreDistances(points: number[][], k: number, metric?: 'euclidean' | 'angular', useKDTree?: boolean): number[];
+/**
+ * Compute core distances for a chunk of indices (for parallel processing).
+ * @param indices Indices to process.
+ * @param allPoints All points.
+ * @param k k value.
+ * @param metric Distance metric.
+ */
+export declare function computeCoreDistancesChunk(indices: number[], allPoints: number[][], k: number, metric?: 'euclidean' | 'angular'): Array<{
+    index: number;
+    coreDistance: number;
+}>;
+//# sourceMappingURL=core-distance.d.ts.map

package/dist/clusters/hdbscan/core-distance.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"core-distance.d.ts","sourceRoot":"","sources":["../../../src/clusters/hdbscan/core-distance.ts"],"names":[],"mappings":"AAAA;;;GAGG;AAKH;;;;;;;;GAQG;AACH,wBAAgB,oBAAoB,CAClC,MAAM,EAAE,MAAM,EAAE,EAAE,EAClB,CAAC,EAAE,MAAM,EACT,MAAM,GAAE,WAAW,GAAG,SAAuB,EAC7C,SAAS,GAAE,OAAe,GACzB,MAAM,EAAE,CAoBV;AAkDD;;;;;;GAMG;AACH,wBAAgB,yBAAyB,CACvC,OAAO,EAAE,MAAM,EAAE,EACjB,SAAS,EAAE,MAAM,EAAE,EAAE,EACrB,CAAC,EAAE,MAAM,EACT,MAAM,GAAE,WAAW,GAAG,SAAuB,GAC5C,KAAK,CAAC;IAAE,KAAK,EAAE,MAAM,CAAC;IAAC,YAAY,EAAE,MAAM,CAAA;CAAE,CAAC,CAmBhD"}

package/dist/clusters/hdbscan/core-distance.js

@@ -0,0 +1,94 @@
+/**
+ * Core distance computation for HDBSCAN.
+ * Core distance = distance to k-th nearest neighbor.
+ */
+import { KDTree, euclideanDistance, angularDistance } from './kd-tree.js';
+import { quickselect } from '../../utils/array-utils.js';
+/**
+ * Compute core distances for all points.
+ * Core distance is the distance to the k-th nearest neighbor.
+ *
+ * @param points All points (embeddings).
+ * @param k k value (typically minSamples).
+ * @param metric Distance metric.
+ * @param useKDTree Whether to use KD-tree for approximate k-NN.
+ */
+export function computeCoreDistances(points, k, metric = 'euclidean', useKDTree = false) {
+    const n = points.length;
+    if (n === 0) {
+        return [];
+    }
+    // k must be at most n-1 (can't count self as neighbor)
+    const effectiveK = Math.min(k, n - 1);
+    if (effectiveK <= 0) {
+        // Only 1 point, core distance is 0
+        return new Array(n).fill(0);
+    }
+    if (useKDTree && metric === 'euclidean') {
+        return computeCoreDistancesKDTree(points, effectiveK);
+    }
+    return computeCoreDistancesBruteForce(points, effectiveK, metric);
+}
+/**
+ * Brute force core distance computation using quickselect.
+ */
+function computeCoreDistancesBruteForce(points, k, metric) {
+    const n = points.length;
+    const coreDistances = new Array(n);
+    const distFn = metric === 'euclidean' ? euclideanDistance : angularDistance;
+    for (let i = 0; i < n; i++) {
+        // Compute distances to all other points
+        const distances = [];
+        for (let j = 0; j < n; j++) {
+            if (i !== j) {
+                distances.push(distFn(points[i], points[j]));
+            }
+        }
+        // Use quickselect to find k-th smallest distance
+        coreDistances[i] = quickselect(distances, k - 1);
+    }
+    return coreDistances;
+}
+/**
+ * KD-tree based core distance computation.
+ */
+function computeCoreDistancesKDTree(points, k) {
+    const tree = new KDTree(points);
+    const coreDistances = new Array(points.length);
+    for (let i = 0; i < points.length; i++) {
+        const neighbors = tree.kNearest(points[i], k, i);
+        if (neighbors.length === k) {
+            coreDistances[i] = neighbors[k - 1].distance;
+        }
+        else {
+            // Fewer than k neighbors, use max distance
+            coreDistances[i] = neighbors.length > 0 ? neighbors[neighbors.length - 1].distance : 0;
+        }
+    }
+    return coreDistances;
+}
+/**
+ * Compute core distances for a chunk of indices (for parallel processing).
+ * @param indices Indices to process.
+ * @param allPoints All points.
+ * @param k k value.
+ * @param metric Distance metric.
+ */
+export function computeCoreDistancesChunk(indices, allPoints, k, metric = 'euclidean') {
+    const distFn = metric === 'euclidean' ? euclideanDistance : angularDistance;
+    const n = allPoints.length;
+    const effectiveK = Math.min(k, n - 1);
+    const results = [];
+    for (const i of indices) {
+        const distances = [];
+        for (let j = 0; j < n; j++) {
+            if (i !== j) {
+                distances.push(distFn(allPoints[i], allPoints[j]));
+            }
+        }
+        const coreDistance = effectiveK > 0 ? quickselect(distances, effectiveK - 1) : 0;
+        results.push({ index: i, coreDistance });
+    }
+    return results;
+}
+//# sourceMappingURL=core-distance.js.map