Keep/keep-notes/lib/utils.ts

import { clsx, type ClassValue } from "clsx"
import { twMerge } from "tailwind-merge"
import { LABEL_COLORS, LabelColorName, QueryType } from "./types"

export function cn(...inputs: ClassValue[]) {
  return twMerge(clsx(inputs))
}

export function getHashColor(name: string): LabelColorName {
  let hash = 0;
  for (let i = 0; i < name.length; i++) {
    hash = name.charCodeAt(i) + ((hash << 5) - hash);
  }
  
  const colors = Object.keys(LABEL_COLORS) as LabelColorName[];
  // Skip 'gray' for colorful tags
  const colorfulColors = colors.filter(c => c !== 'gray');
  const colorIndex = Math.abs(hash) % colorfulColors.length;
  
  return colorfulColors[colorIndex];
}

export function cosineSimilarity(vecA: number[], vecB: number[]): number {
  if (vecA.length !== vecB.length) return 0;

  let dotProduct = 0;
  let mA = 0;
  let mB = 0;

  for (let i = 0; i < vecA.length; i++) {
    dotProduct += vecA[i] * vecB[i];
    mA += vecA[i] * vecA[i];
    mB += vecB[i] * vecB[i];
  }

  mA = Math.sqrt(mA);
  mB = Math.sqrt(mB);

  if (mA === 0 || mB === 0) return 0;

  return dotProduct / (mA * mB);
}

/**
 * Validate an embedding vector for quality issues
 */
export function validateEmbedding(embedding: number[]): { valid: boolean; issues: string[] } {
  const issues: string[] = [];

  // Check 1: Dimensionality > 0
  if (!embedding || embedding.length === 0) {
    issues.push('Embedding is empty or has zero dimensionality');
    return { valid: false, issues };
  }

  // Check 2: Valid numbers (no NaN or Infinity)
  let hasNaN = false;
  let hasInfinity = false;
  let hasZeroVector = true;

  for (let i = 0; i < embedding.length; i++) {
    const val = embedding[i];
    if (isNaN(val)) hasNaN = true;
    if (!isFinite(val)) hasInfinity = true;
    if (val !== 0) hasZeroVector = false;
  }

  if (hasNaN) {
    issues.push('Embedding contains NaN values');
  }
  if (hasInfinity) {
    issues.push('Embedding contains Infinity values');
  }
  if (hasZeroVector) {
    issues.push('Embedding is a zero vector (all values are 0)');
  }

  // Check 3: L2 norm is in reasonable range (0.7 to 1.2)
  const l2Norm = calculateL2Norm(embedding);
  if (l2Norm < 0.7 || l2Norm > 1.2) {
    issues.push(`L2 norm is ${l2Norm.toFixed(3)} (expected range: 0.7-1.2)`);
  }

  return {
    valid: issues.length === 0,
    issues
  };
}

/**
 * Calculate L2 norm of a vector
 */
export function calculateL2Norm(vector: number[]): number {
  let sum = 0;
  for (let i = 0; i < vector.length; i++) {
    sum += vector[i] * vector[i];
  }
  return Math.sqrt(sum);
}

/**
 * Normalize an embedding to unit L2 norm
 */
export function normalizeEmbedding(embedding: number[]): number[] {
  const norm = calculateL2Norm(embedding);
  if (norm === 0) return embedding; // Can't normalize zero vector

  return embedding.map(val => val / norm);
}

/**
 * Calculate the RRF (Reciprocal Rank Fusion) constant k
 *
 * RRF Formula: score = Σ 1 / (k + rank)
 *
 * The k constant controls how much we penalize lower rankings:
 * - Lower k (e.g., 20) penalizes low ranks more heavily
 * - Higher k (e.g., 60) is more lenient with low ranks
 *
 * Adaptive formula: k = max(20, totalNotes / 10)
 * - For small datasets (< 200 notes): k = 20 (strict)
 * - For larger datasets: k scales linearly
 *
 * Examples:
 * - 50 notes → k = 20
 * - 200 notes → k = 20
 * - 500 notes → k = 50
 * - 1000 notes → k = 100
 */
export function calculateRRFK(totalNotes: number): number {
  const BASE_K = 20;
  const adaptiveK = Math.floor(totalNotes / 10);

  return Math.max(BASE_K, adaptiveK);
}

/**
 * Detect the type of search query to adapt search weights
 *
 * Detection rules:
 * 1. EXACT: Query contains quotes (e.g., "Error 404")
 * 2. CONCEPTUAL: Query starts with question words or is a phrase like "how to X"
 * 3. MIXED: No specific pattern detected
 *
 * Examples:
 * - "exact phrase" → 'exact'
 * - "how to cook pasta" → 'conceptual'
 * - "what is python" → 'conceptual'
 * - "javascript tutorial" → 'mixed'
 */
export function detectQueryType(query: string): QueryType {
  const trimmed = query.trim().toLowerCase();

  // Rule 1: Check for quotes (exact match)
  if ((query.startsWith('"') && query.endsWith('"')) ||
      (query.startsWith("'") && query.endsWith("'"))) {
    return 'exact';
  }

  // Rule 2: Check for conceptual patterns
  const conceptualPatterns = [
    /^(how|what|when|where|why|who|which|whose|can|could|would|should|is|are|do|does|did)\b/,
    /^(how to|ways to|best way to|guide for|tips for|learn about|understand)/,
    /^(tutorial|guide|introduction|overview|explanation|examples)/,
  ];

  for (const pattern of conceptualPatterns) {
    if (pattern.test(trimmed)) {
      return 'conceptual';
    }
  }

  // Default: mixed search
  return 'mixed';
}

/**
 * Get search weight multipliers based on query type
 *
 * Returns keyword and semantic weight multipliers:
 * - EXACT: Boost keyword matches (2.0x), reduce semantic (0.7x)
 * - CONCEPTUAL: Reduce keyword (0.7x), boost semantic (1.5x)
 * - MIXED: Default weights (1.0x, 1.0x)
 */
export function getSearchWeights(queryType: QueryType): {
  keywordWeight: number;
  semanticWeight: number;
} {
  switch (queryType) {
    case 'exact':
      return {
        keywordWeight: 2.0,
        semanticWeight: 0.7
      };
    case 'conceptual':
      return {
        keywordWeight: 0.7,
        semanticWeight: 1.5
      };
    case 'mixed':
    default:
      return {
        keywordWeight: 1.0,
        semanticWeight: 1.0
      };
  }
}