Function run_matching_tests

tests/test-dataset-streaming.py:701–774

Run matching accuracy tests and measure query latencies

(test_vectors, collection, brute_force_results)

    return matches_test_vectors, rps_test_vectors

def run_matching_tests(test_vectors, collection, brute_force_results):
    """Run matching accuracy tests and measure query latencies"""
    print(f"\nStarting similarity search tests with {len(test_vectors)} queries...")

    total_recall = 0
    total_queries = 0
    latencies = []

    for i, query_vec in enumerate(test_vectors):
        try:
            test_vec = next(
                test_vec
                for test_vec in brute_force_results
                if str(query_vec["id"]) == str(test_vec["query_id"])  # Compare as strings
            )

            # Measure query latency
            query_start_time = time.time()
            results = collection.search.dense(
                query_vector=query_vec["dense_values"],
                top_k=5,
                return_raw_text=True
            )
            query_end_time = time.time()
            query_latency = (query_end_time - query_start_time) * 1000  # Convert to ms
            latencies.append(query_latency)

            if "results" in results:
                query_id = test_vec["query_id"]
                print(f"\nQuery {i + 1} (Vector ID: {query_id}):")
                print(f"Query latency: {query_latency:.2f} ms")

                server_top5 = [str(match["id"]) for match in results["results"][:5]]  # Convert to string
                print("Server top 5:", server_top5)

                brute_force_top5 = [str(test_vec[f"top{j}_id"]) for j in range(1, 6)]  # Convert to string
                print("Brute force top 5:", brute_force_top5)

                matches = sum(1 for id in server_top5 if id in brute_force_top5)
                recall = (matches / 5) * 100
                total_recall += recall
                total_queries += 1

                print(f"Recall@5 for this query: {recall}% ({matches}/5 matches)")

                time.sleep(0.1)

        except Exception as e:
            print(f"Error in query {i + 1}: {e}")

    # Calculate and display latency statistics
    if latencies:
        latencies.sort()
        avg_latency = sum(latencies) / len(latencies)
        p50_latency = latencies[int(len(latencies) * 0.5)]
        p90_latency = latencies[int(len(latencies) * 0.9)]
        p95_latency = latencies[int(len(latencies) * 0.95)]
        min_latency = min(latencies)