wifi-densepose/examples/neural-trader/core/hnsw-vector-search.js

/**
 * HNSW Vector Search Integration
 *
 * Demonstrates using Neural Trader's native HNSW implementation
 * (150x faster than pure JS) with RuVector's vector database
 *
 * Features:
 * - Native Rust HNSW indexing via NAPI
 * - SIMD-accelerated distance calculations
 * - Approximate nearest neighbor search
 * - Pattern matching for trading signals
 */

import NeuralTrader from 'neural-trader';

// HNSW configuration optimized for trading patterns
const hnswConfig = {
  // Index construction parameters
  m: 32,                    // Max connections per node (higher = better recall, more memory)
  efConstruction: 200,      // Build-time search depth (higher = better index, slower build)

  // Search parameters
  efSearch: 100,            // Query-time search depth (higher = better recall, slower search)

  // Distance metric
  distanceMetric: 'cosine', // cosine, euclidean, dotProduct, manhattan

  // Performance optimizations
  simd: true,               // Use SIMD for distance calculations
  quantization: {
    enabled: false,         // Enable for 4x memory reduction
    bits: 8                 // Quantization precision
  }
};

// Vector dimension for trading features
const VECTOR_DIM = 256;
const PATTERN_LOOKBACK = 50;  // Days to analyze for patterns

async function main() {
  console.log('='.repeat(60));
  console.log('HNSW Vector Search - Neural Trader Integration');
  console.log('='.repeat(60));
  console.log();

  // 1. Initialize HNSW Index
  console.log('1. Initializing HNSW Index...');
  console.log(`   Dimensions: ${VECTOR_DIM}`);
  console.log(`   M (connections): ${hnswConfig.m}`);
  console.log(`   ef_construction: ${hnswConfig.efConstruction}`);
  console.log(`   ef_search: ${hnswConfig.efSearch}`);
  console.log(`   SIMD acceleration: ${hnswConfig.simd ? 'Enabled' : 'Disabled'}`);
  console.log();

  // 2. Generate historical trading patterns
  console.log('2. Generating historical trading patterns...');
  const patterns = generateHistoricalPatterns(10000);
  console.log(`   Generated ${patterns.length} historical patterns`);
  console.log();

  // 3. Build HNSW index
  console.log('3. Building HNSW index...');
  const buildStart = performance.now();

  // Simulate native HNSW index building
  const index = await buildHNSWIndex(patterns, hnswConfig);

  const buildTime = performance.now() - buildStart;
  console.log(`   Index built in ${buildTime.toFixed(2)}ms`);
  console.log(`   Throughput: ${(patterns.length / (buildTime / 1000)).toFixed(0)} vectors/sec`);
  console.log();

  // 4. Real-time pattern matching
  console.log('4. Real-time pattern matching...');
  const currentPattern = generateCurrentPattern();

  const searchStart = performance.now();
  const matches = await searchHNSW(index, currentPattern.vector, 10);
  const searchTime = performance.now() - searchStart;

  console.log(`   Query time: ${searchTime.toFixed(3)}ms`);
  console.log(`   Found ${matches.length} similar patterns:`);
  console.log();

  // Display matches
  console.log('   Rank | Similarity | Symbol | Date       | Next Day Return');
  console.log('   ' + '-'.repeat(55));

  matches.forEach((match, i) => {
    const date = new Date(match.metadata.timestamp).toISOString().split('T')[0];
    const returnStr = (match.metadata.nextDayReturn * 100).toFixed(2) + '%';
    console.log(`   ${(i + 1).toString().padStart(4)} | ${match.similarity.toFixed(4).padStart(10)} | ${match.metadata.symbol.padEnd(6)} | ${date} | ${returnStr.padStart(15)}`);
  });
  console.log();

  // 5. Generate trading signal based on historical patterns
  console.log('5. Trading Signal Analysis...');
  const signal = analyzePatterns(matches);

  console.log(`   Expected return: ${(signal.expectedReturn * 100).toFixed(2)}%`);
  console.log(`   Win rate: ${(signal.winRate * 100).toFixed(1)}%`);
  console.log(`   Confidence: ${(signal.confidence * 100).toFixed(1)}%`);
  console.log(`   Signal: ${signal.action.toUpperCase()}`);
  console.log();

  // 6. Benchmark comparison
  console.log('6. Performance Benchmark...');
  await runBenchmark(patterns);
  console.log();

  console.log('='.repeat(60));
  console.log('HNSW Vector Search completed!');
  console.log('='.repeat(60));
}

// Generate historical trading patterns with labels
function generateHistoricalPatterns(count) {
  const symbols = ['AAPL', 'GOOGL', 'MSFT', 'AMZN', 'NVDA', 'META', 'TSLA', 'AMD'];
  const patterns = [];

  for (let i = 0; i < count; i++) {
    const symbol = symbols[i % symbols.length];
    const vector = generatePatternVector();
    const nextDayReturn = (Math.random() - 0.48) * 0.1; // Slight positive bias

    patterns.push({
      id: `pattern_${i}`,
      vector,
      metadata: {
        symbol,
        timestamp: Date.now() - (count - i) * 86400000,
        nextDayReturn,
        volatility: Math.random() * 0.05,
        volume: Math.floor(Math.random() * 10000000)
      }
    });
  }

  return patterns;
}

// Generate a pattern vector with technical features
function generatePatternVector() {
  const vector = new Float32Array(VECTOR_DIM);

  // Price returns (0-49)
  for (let i = 0; i < 50; i++) {
    vector[i] = (Math.random() - 0.5) * 0.1;
  }

  // Volume features (50-99)
  for (let i = 50; i < 100; i++) {
    vector[i] = Math.random() * 2 - 1;
  }

  // Moving averages (100-119)
  for (let i = 100; i < 120; i++) {
    vector[i] = (Math.random() - 0.5) * 0.2;
  }

  // RSI features (120-139)
  for (let i = 120; i < 140; i++) {
    vector[i] = Math.random() * 2 - 1; // Normalized RSI
  }

  // MACD features (140-159)
  for (let i = 140; i < 160; i++) {
    vector[i] = (Math.random() - 0.5) * 0.5;
  }

  // Bollinger band features (160-179)
  for (let i = 160; i < 180; i++) {
    vector[i] = (Math.random() - 0.5) * 2;
  }

  // Additional technical indicators (180-255)
  for (let i = 180; i < VECTOR_DIM; i++) {
    vector[i] = (Math.random() - 0.5) * 0.3;
  }

  // Normalize the vector
  const norm = Math.sqrt(vector.reduce((sum, v) => sum + v * v, 0));
  for (let i = 0; i < VECTOR_DIM; i++) {
    vector[i] /= norm;
  }

  return vector;
}

// Generate current market pattern
function generateCurrentPattern() {
  return {
    vector: generatePatternVector(),
    metadata: {
      symbol: 'CURRENT',
      timestamp: Date.now()
    }
  };
}

// Build HNSW index (simulates native binding)
async function buildHNSWIndex(patterns, config) {
  // In production with neural-trader native bindings:
  // const { HNSWIndex } = require('neural-trader/native');
  // const index = new HNSWIndex(VECTOR_DIM, config);
  // await index.addBatch(patterns);

  // Simulate index building
  const index = {
    size: patterns.length,
    patterns: patterns,
    config: config
  };

  // Simulate build time based on complexity
  const estimatedBuildTime = patterns.length * 0.05; // ~0.05ms per vector
  await new Promise(resolve => setTimeout(resolve, Math.min(estimatedBuildTime, 100)));

  return index;
}

// Search HNSW index
async function searchHNSW(index, queryVector, k) {
  // In production:
  // return await index.search(queryVector, k);

  // Simulate approximate nearest neighbor search
  const results = [];
  const queryNorm = Math.sqrt(queryVector.reduce((sum, v) => sum + v * v, 0));

  // Calculate cosine similarities (simulated - in production uses SIMD)
  const similarities = index.patterns.map((pattern, idx) => {
    let dotProduct = 0;
    for (let i = 0; i < VECTOR_DIM; i++) {
      dotProduct += queryVector[i] * pattern.vector[i];
    }
    return {
      index: idx,
      similarity: dotProduct // Already normalized
    };
  });

  // Sort by similarity (descending) and take top k
  similarities.sort((a, b) => b.similarity - a.similarity);

  for (let i = 0; i < k; i++) {
    const match = similarities[i];
    const pattern = index.patterns[match.index];
    results.push({
      id: pattern.id,
      similarity: match.similarity,
      metadata: pattern.metadata
    });
  }

  return results;
}

// Analyze matched patterns to generate trading signal
function analyzePatterns(matches) {
  // Calculate expected return from similar patterns
  const returns = matches.map(m => m.metadata.nextDayReturn);
  const weights = matches.map(m => m.similarity);
  const totalWeight = weights.reduce((sum, w) => sum + w, 0);

  const expectedReturn = returns.reduce((sum, r, i) => sum + r * weights[i], 0) / totalWeight;
  const winRate = returns.filter(r => r > 0).length / returns.length;

  // Confidence based on similarity and consistency
  const avgSimilarity = matches.reduce((sum, m) => sum + m.similarity, 0) / matches.length;
  const returnStd = Math.sqrt(
    returns.reduce((sum, r) => sum + Math.pow(r - expectedReturn, 2), 0) / returns.length
  );
  const confidence = avgSimilarity * (1 - returnStd * 5); // Penalize high variance

  // Determine action
  let action = 'hold';
  if (expectedReturn > 0.005 && confidence > 0.6) action = 'buy';
  else if (expectedReturn < -0.005 && confidence > 0.6) action = 'sell';

  return { expectedReturn, winRate, confidence: Math.max(0, confidence), action };
}

// Run performance benchmark
async function runBenchmark(patterns) {
  const testSizes = [100, 1000, 5000, 10000];
  const queryVector = generatePatternVector();

  console.log('   Dataset Size | Build Time | Query Time | Throughput');
  console.log('   ' + '-'.repeat(55));

  for (const size of testSizes) {
    if (size > patterns.length) continue;

    const subset = patterns.slice(0, size);

    // Build index
    const buildStart = performance.now();
    const index = await buildHNSWIndex(subset, hnswConfig);
    const buildTime = performance.now() - buildStart;

    // Query index
    const queryStart = performance.now();
    await searchHNSW(index, queryVector, 10);
    const queryTime = performance.now() - queryStart;

    const throughput = (size / (buildTime / 1000)).toFixed(0);
    console.log(`   ${size.toString().padStart(12)} | ${buildTime.toFixed(2).padStart(10)}ms | ${queryTime.toFixed(3).padStart(10)}ms | ${throughput.padStart(10)}/sec`);
  }

  console.log();
  console.log('   Note: Native Rust bindings provide 150x faster search');
  console.log('   with SIMD acceleration and optimized memory layout.');
}

// Run the example
main().catch(console.error);