/**
 * @fileoverview PatternStore helper classes and utility functions
 * @module @skillsmith/core/learning/PatternStore.helpers
 *
 * Contains FisherInformationMatrix implementation, SQL schema, and utility functions.
 */
import type { PatternRecommendationContext, StoredPattern, PatternOutcome, PatternRow, EWCConfig, ConsolidationState } from './PatternStore.types.js';
import type { Database } from '../db/database-interface.js';
/**
 * SQLite schema for pattern storage
 */
export declare const PATTERN_STORE_SCHEMA = "\n-- Patterns table: stores recommendation patterns with outcomes\nCREATE TABLE IF NOT EXISTS patterns (\n  pattern_id TEXT PRIMARY KEY,\n  context_embedding BLOB NOT NULL,\n  skill_id TEXT NOT NULL,\n  skill_features TEXT NOT NULL,\n  context_data TEXT NOT NULL,\n  outcome_type TEXT NOT NULL,\n  outcome_reward REAL NOT NULL,\n  importance REAL NOT NULL DEFAULT 0.1,\n  original_score REAL NOT NULL,\n  source TEXT NOT NULL,\n  access_count INTEGER NOT NULL DEFAULT 0,\n  created_at INTEGER NOT NULL DEFAULT (unixepoch()),\n  last_accessed_at INTEGER NOT NULL DEFAULT (unixepoch())\n);\n\n-- Indexes for efficient queries\nCREATE INDEX IF NOT EXISTS idx_patterns_skill_id ON patterns(skill_id);\nCREATE INDEX IF NOT EXISTS idx_patterns_outcome_type ON patterns(outcome_type);\nCREATE INDEX IF NOT EXISTS idx_patterns_importance ON patterns(importance DESC);\nCREATE INDEX IF NOT EXISTS idx_patterns_created_at ON patterns(created_at DESC);\n\n-- Fisher Information matrix state\nCREATE TABLE IF NOT EXISTS fisher_info (\n  id INTEGER PRIMARY KEY DEFAULT 1,\n  matrix_data BLOB NOT NULL,\n  update_count INTEGER NOT NULL DEFAULT 0,\n  last_decay_at INTEGER NOT NULL DEFAULT (unixepoch()),\n  updated_at INTEGER NOT NULL DEFAULT (unixepoch())\n);\n\n-- Consolidation history for monitoring\nCREATE TABLE IF NOT EXISTS consolidation_history (\n  id INTEGER PRIMARY KEY AUTOINCREMENT,\n  timestamp INTEGER NOT NULL DEFAULT (unixepoch()),\n  patterns_processed INTEGER NOT NULL,\n  patterns_preserved INTEGER NOT NULL,\n  patterns_pruned INTEGER NOT NULL,\n  preservation_rate REAL NOT NULL,\n  duration_ms INTEGER NOT NULL,\n  average_importance REAL NOT NULL\n);\n";
/**
 * Fisher Information Matrix interface
 */
export interface IFisherInformationMatrix {
    getImportance(dimensionIndex: number): number;
    update(gradient: Float32Array): void;
    decay(decayFactor: number): void;
    getImportanceVector(): Float32Array;
    getAverageImportance(): number;
    serialize(): Buffer;
    deserialize(buffer: Buffer): void;
    reset(): void;
    getUpdateCount(): number;
}
/**
 * Fisher Information Matrix implementation for EWC++
 *
 * Stores diagonal approximation of Fisher Information,
 * indicating which "weights" (pattern dimensions) are important.
 *
 * In the context of pattern storage:
 * - Each dimension of the context embedding has an importance value
 * - High importance = changing this dimension would harm prediction
 * - Low importance = safe to overwrite with new patterns
 */
export declare class FisherInformationMatrix implements IFisherInformationMatrix {
    private dimensions;
    /** Diagonal of Fisher Information (importance per dimension) */
    private importance;
    /** Running sum for online updates */
    private runningSum;
    /** Number of updates performed */
    private updateCount;
    constructor(dimensions: number);
    getImportance(dimensionIndex: number): number;
    update(gradient: Float32Array): void;
    decay(decayFactor: number): void;
    getImportanceVector(): Float32Array;
    getAverageImportance(): number;
    serialize(): Buffer;
    deserialize(buffer: Buffer): void;
    reset(): void;
    getUpdateCount(): number;
}
/**
 * Convert pattern context to text for embedding
 */
export declare function contextToText(context: PatternRecommendationContext): string;
/**
 * Compute gradient between two embeddings
 */
export declare function computeGradient(a: Float32Array, b: Float32Array): Float32Array;
/**
 * Deserialize embedding from buffer
 */
export declare function deserializeEmbedding(buffer: Buffer, dimensions: number): Float32Array;
/**
 * Calculate cosine similarity between two embeddings
 */
export declare function cosineSimilarity(a: Float32Array, b: Float32Array): number;
/**
 * Calculate importance-weighted similarity
 */
export declare function importanceWeightedSimilarity(a: Float32Array, b: Float32Array, importance: Float32Array): number;
/**
 * Calculate pattern importance based on outcome and access patterns
 */
export declare function calculatePatternImportance(pattern: StoredPattern, outcome: PatternOutcome): number;
/**
 * Calculate dimension-based importance using Fisher Information
 */
export declare function calculateDimensionImportance(pattern: StoredPattern, importanceVector: Float32Array, dimensions: number, lambda: number): number;
/**
 * Convert database row to StoredPattern
 */
export declare function rowToStoredPattern(row: PatternRow, dimensions: number): StoredPattern;
/**
 * Compute the average embedding across stored patterns
 *
 * @param db - Database instance
 * @param limit - Max patterns to sample
 * @param dimensions - Embedding dimensions
 * @returns Average embedding vector
 */
export declare function computeAverageEmbedding(db: Database, limit: number, dimensions: number): Promise<Float32Array>;
/**
 * Determine whether the store should run consolidation
 *
 * @param state - Current consolidation state
 * @param ewcConfig - EWC configuration
 * @returns True if consolidation should run
 */
export declare function shouldConsolidate(state: ConsolidationState, ewcConfig: EWCConfig): boolean;
//# sourceMappingURL=PatternStore.helpers.d.ts.map