diff --git a/main.cpp b/main.cpp
index 69e36cc..c5f111d 100644
--- a/main.cpp
+++ b/main.cpp
@@ -101,7 +101,7 @@ auto bench_pgm(const std::vector<uint64_t>& data, const std::vector<uint64_t>& q
     volatile uint64_t res = 0;
     
     std::cout << "Construct PGM index eps_l=" << Epsilon << " eps_i=" << EpsilonRecursive << std::endl;
-    pgm::PGMIndex<uint64_t, Epsilon, EpsilonRecursive, true, 16, float> index_branchless(data.begin(), data.end()-1);
+    pgm::PGMIndex<uint64_t, Epsilon, EpsilonRecursive, true, 0, float> index_branchless(data.begin(), data.end()-1);
     
     // branchless PGM without last-mile search
     auto start = std::chrono::high_resolution_clock::now();
@@ -131,8 +131,9 @@ auto bench_pgm(const std::vector<uint64_t>& data, const std::vector<uint64_t>& q
     // to avoid influence of hot cache
     std::vector<uint64_t> data_cpy(data);
     std::vector<uint64_t> queries_cpy(queries);
+    
     std::cout << "Construct PGM index eps_l=" << Epsilon << " eps_i=" << EpsilonRecursive << std::endl;
-    pgm::PGMIndex<uint64_t, Epsilon, EpsilonRecursive, false, 16, float> index(data_cpy.begin(), data_cpy.end()-1);
+    pgm::PGMIndex<uint64_t, Epsilon, EpsilonRecursive, false, 0, float> index(data_cpy.begin(), data_cpy.end()-1);
     // branchy PGM without last-mile search
     start = std::chrono::high_resolution_clock::now();
     for (auto q : queries_cpy) {
@@ -166,13 +167,18 @@ int main(int argc, const char * argv[]) {
 //    bench_search_repeat(20, 500, "/Users/liuqiyu/Desktop/bench_search_result_new.csv");
 //    exit(0);
     
-    const std::string fname = "/Users/liuqiyu/Desktop/SOSD_data/books_800M_uint64";
-    const size_t nq = 200;
+    const std::string fname = "/Users/liuqiyu/Desktop/SOSD_data/osm_cellids_800M_uint64";
+    const size_t nq = 500;
     const size_t repeat = 10;
     
     std::cout << "Load data from " << fname << std::endl;
-    auto data = benchmark::load_data<uint64_t>(fname);
-    std::sort(data.begin(), data.end()-1);
+    auto data = benchmark::load_data<uint64_t>(fname, true, 200000000);
+    std::sort(data.begin(), data.end());
+    
+    auto data_stats = benchmark::get_data_stats(data);
+    std::cout << "mean: " << data_stats.mean 
+              << " variance: " << data_stats.var
+              << " hardness ratio: " << data_stats.var/data_stats.mean << std::endl;
     
     std::vector<std::pair<size_t, stats>> bench_results;
     
@@ -243,7 +249,7 @@ int main(int argc, const char * argv[]) {
     }
     
     // start from 7 cold cache config
-    std::ofstream ofs("/Users/liuqiyu/Desktop/bench_pgm_result_books_repeat_10_0250.csv");
+    std::ofstream ofs("/Users/liuqiyu/Desktop/bench_pgm_result_osm_repeat_10_2103.csv");
     ofs << "round,eps_l,eps_i,levels,lls,ils,latency_branchy_i,latency_branchy_l,latency_branchless_i,latency_branchless_l" << std::endl;
     
     for (auto br : bench_results) {
diff --git a/utils.h b/utils.h
index 6c6c230..a8e85de 100644
--- a/utils.h
+++ b/utils.h
@@ -22,8 +22,7 @@ static uint64_t timing(std::function<void()> fn) {
 
 // Loads values from binary file into vector.
 template <typename T>
-static std::vector<T> load_data(const std::string& filename,
-                                bool print = true) {
+static std::vector<T> load_data(const std::string& filename, bool print = true, size_t sample_size = 0) {
     std::vector<T> data;
     const uint64_t ns = timing([&] {
         std::ifstream in(filename, std::ios::binary);
@@ -46,8 +45,16 @@ static std::vector<T> load_data(const std::string& filename,
               << ms << " ms (" << static_cast<double>(data.size()) / 1000 / ms
               << " M values/s)" << std::endl;
     }
-
-    return data;
+    
+    if (sample_size > 0) {
+        std::random_device rd;
+        std::mt19937 gen(rd());
+        std::vector<T> sample;
+        std::sample(data.begin(), data.end()-1, std::back_inserter(sample), sample_size, gen);
+        return sample;
+    } else {
+        return data;
+    }
 }
 
 template <typename K>
@@ -58,8 +65,10 @@ auto get_data_stats(const std::vector<K>& data) {
     }
     const auto n = gaps.size();
     double mean = std::accumulate(gaps.begin(), gaps.end(), 0.0)/n;
-    double sq_sum = std::inner_product(gaps.begin(), gaps.end(), gaps.begin(), 0.0);
-    double var = sq_sum/n - mean*mean;
+    double sq_sum = std::accumulate(gaps.begin(), gaps.end(), 0.0, [mean](double acc, double val) {
+        return acc + (val - mean) * (val - mean);
+    });
+    double var = sq_sum/n;
     struct data_stats {double mean; double var;};
     return data_stats {mean, var};
 }