HNSW Benchmarking script

## Comprehensive HNSW Benchmark Script
## Compares Original vs Optimized implementation
## Saves results to CSV and generates plots

addpath ("inst");
addpath ("inst/Clustering");

printf ("\n=== HNSW Benchmark: Original vs Optimized ===\n\n");

## Test configurations
N_values = [100, 250, 500, 1000];
D_values = [1, 5, 10, 50, 100];
K = 3;
num_queries = 10;

## Generate all combinations
num_N = length (N_values);
num_D = length (D_values);
total_tests = num_N * num_D;

## Storage for results
results = struct ();
results.N = zeros (total_tests, 1);
results.D = zeros (total_tests, 1);
results.orig_build = zeros (total_tests, 1);
results.orig_query = zeros (total_tests, 1);
results.opt_build = zeros (total_tests, 1);
results.opt_query = zeros (total_tests, 1);

printf ("Running %d configurations (K=%d, queries=%d)\n", total_tests, K, num_queries);
printf ("This will take a few minutes...\n\n");

test_idx = 0;
for n_idx = 1:num_N
  N = N_values(n_idx);
  for d_idx = 1:num_D
    D = D_values(d_idx);
    test_idx = test_idx + 1;

    printf ("[%2d/%d] N=%4d, D=%3d ... ", test_idx, total_tests, N, D);
    fflush (stdout);

    ## Generate reproducible data
    rand ("seed", 42);
    X = rand (N, D);
    Q = rand (num_queries, D);

    ## Original implementation
    tic; obj1 = hnswSearcher_original (X); b1 = toc;
    tic; [~, ~] = knnsearch (obj1, Q, "K", K); q1 = toc;

    ## Optimized implementation
    tic; obj2 = hnswSearcher (X); b2 = toc;
    tic; [~, ~] = knnsearch (obj2, Q, "K", K); q2 = toc;

    results.N(test_idx) = N;
    results.D(test_idx) = D;
    results.orig_build(test_idx) = b1;
    results.orig_query(test_idx) = q1;
    results.opt_build(test_idx) = b2;
    results.opt_query(test_idx) = q2;

    printf ("Build: %.2fs -> %.2fs  |  Query: %.3fs -> %.3fs\n", b1, b2, q1, q2);
    fflush (stdout);
  endfor
endfor

## Calculate speedups
build_speedup = results.orig_build ./ results.opt_build;
query_speedup = results.orig_query ./ results.opt_query;

## Save to CSV
csv_file = "benchmark_results.csv";
fid = fopen (csv_file, "w");
fprintf (fid, "N,D,Original_Build_sec,Original_Query_sec,Optimized_Build_sec,Optimized_Query_sec,Build_Speedup,Query_Speedup\n");
for i = 1:total_tests
  fprintf (fid, "%d,%d,%.4f,%.4f,%.4f,%.4f,%.2f,%.2f\n", ...
           results.N(i), results.D(i), ...
           results.orig_build(i), results.orig_query(i), ...
           results.opt_build(i), results.opt_query(i), ...
           build_speedup(i), query_speedup(i));
endfor
fclose (fid);
printf ("\n✓ Results saved to %s\n", csv_file);

## Print results table
printf ("\n--- Timing Results (seconds) ---\n");
printf ("+-------+-----+--------------+--------------+--------------+--------------+\n");
printf ("|   N   |  D  | Orig Build   | Orig Query   |  Opt Build   |  Opt Query   |\n");
printf ("+-------+-----+--------------+--------------+--------------+--------------+\n");
for i = 1:total_tests
  printf ("| %5d | %3d | %12.3f | %12.3f | %12.3f | %12.3f |\n", ...
          results.N(i), results.D(i), ...
          results.orig_build(i), results.orig_query(i), ...
          results.opt_build(i), results.opt_query(i));
endfor
printf ("+-------+-----+--------------+--------------+--------------+--------------+\n");

## Print speedup table
printf ("\n--- Speedup (Original / Optimized) ---\n");
printf ("+-------+-----+----------------+----------------+\n");
printf ("|   N   |  D  | Build Speedup  | Query Speedup  |\n");
printf ("+-------+-----+----------------+----------------+\n");
for i = 1:total_tests
  printf ("| %5d | %3d | %12.2fx   | %12.2fx   |\n", ...
          results.N(i), results.D(i), build_speedup(i), query_speedup(i));
endfor
printf ("+-------+-----+----------------+----------------+\n");

## Summary statistics
printf ("\n--- Summary Statistics ---\n");
printf ("Build Speedup:  Min = %.2fx  |  Max = %.2fx  |  Mean = %.2fx\n", ...
        min(build_speedup), max(build_speedup), mean(build_speedup));
printf ("Query Speedup:  Min = %.2fx  |  Max = %.2fx  |  Mean = %.2fx\n", ...
        min(query_speedup), max(query_speedup), mean(query_speedup));

## Generate plots
printf ("\nGenerating plots...\n");

colors = {"r", "g", "b", "m", "c"};
markers = {"o", "s", "d", "^", "v"};

## Plot 1: Speedup comparison
figure (1, "visible", "off");
clf;
set (gcf, "paperposition", [0 0 12 5]);

subplot (1, 2, 1);
hold on;
for d_idx = 1:num_D
  D = D_values(d_idx);
  mask = results.D == D;
  plot (results.N(mask), build_speedup(mask), [colors{d_idx} "-" markers{d_idx}], ...
        "linewidth", 2, "markersize", 8, "markerfacecolor", colors{d_idx});
endfor
xlabel ("Dataset Size (N)", "fontsize", 11);
ylabel ("Speedup (x)", "fontsize", 11);
title ("Build Time Speedup", "fontsize", 12, "fontweight", "bold");
legend (arrayfun (@(d) sprintf("D=%d", d), D_values, "UniformOutput", false), ...
        "location", "northwest", "fontsize", 9);
grid on;
set (gca, "fontsize", 10);
ylim ([0.9 max(build_speedup)*1.15]);

subplot (1, 2, 2);
hold on;
for d_idx = 1:num_D
  D = D_values(d_idx);
  mask = results.D == D;
  plot (results.N(mask), query_speedup(mask), [colors{d_idx} "-" markers{d_idx}], ...
        "linewidth", 2, "markersize", 8, "markerfacecolor", colors{d_idx});
endfor
xlabel ("Dataset Size (N)", "fontsize", 11);
ylabel ("Speedup (x)", "fontsize", 11);
title ("Query Time Speedup", "fontsize", 12, "fontweight", "bold");
legend (arrayfun (@(d) sprintf("D=%d", d), D_values, "UniformOutput", false), ...
        "location", "northwest", "fontsize", 9);
grid on;
set (gca, "fontsize", 10);
ylim ([0.9 max(query_speedup)*1.15]);

print (1, "benchmark_speedup.png", "-dpng", "-r200");
printf ("✓ Speedup plot saved to benchmark_speedup.png\n");

## Plot 2: Timing comparison
figure (2, "visible", "off");
clf;
set (gcf, "paperposition", [0 0 12 5]);

subplot (1, 2, 1);
bar_data = [];
for n_idx = 1:num_N
  N = N_values(n_idx);
  mask = results.N == N;
  bar_data = [bar_data; mean(results.orig_build(mask)), mean(results.opt_build(mask))];
endfor
h = bar (bar_data, 0.8);
set (h(1), "facecolor", [0.2 0.4 0.8]);
set (h(2), "facecolor", [0.9 0.4 0.2]);
set (gca, "xticklabel", N_values, "fontsize", 10);
xlabel ("Dataset Size (N)", "fontsize", 11);
ylabel ("Build Time (seconds)", "fontsize", 11);
title ("Mean Build Time by N", "fontsize", 12, "fontweight", "bold");
legend ({"Original", "Optimized"}, "location", "northwest", "fontsize", 9);
grid on;

subplot (1, 2, 2);
bar_data = [];
for n_idx = 1:num_N
  N = N_values(n_idx);
  mask = results.N == N;
  bar_data = [bar_data; mean(results.orig_query(mask)), mean(results.opt_query(mask))];
endfor
h = bar (bar_data, 0.8);
set (h(1), "facecolor", [0.2 0.4 0.8]);
set (h(2), "facecolor", [0.9 0.4 0.2]);
set (gca, "xticklabel", N_values, "fontsize", 10);
xlabel ("Dataset Size (N)", "fontsize", 11);
ylabel ("Query Time (seconds)", "fontsize", 11);
title ("Mean Query Time by N", "fontsize", 12, "fontweight", "bold");
legend ({"Original", "Optimized"}, "location", "northwest", "fontsize", 9);
grid on;

print (2, "benchmark_timing.png", "-dpng", "-r200");
printf ("✓ Timing plot saved to benchmark_timing.png\n");

printf ("\nBenchmark complete!\n\n");