Skip to content

Commit

Permalink
Fix calculation of best ops/sec in benchmarks (#6973)
Browse files Browse the repository at this point in the history
  • Loading branch information
SiarheiFedartsou authored Jun 29, 2024
1 parent 0e17869 commit cacb1b2
Show file tree
Hide file tree
Showing 2 changed files with 9 additions and 6 deletions.
6 changes: 3 additions & 3 deletions scripts/ci/e2e_benchmark.py
Original file line number Diff line number Diff line change
Expand Up @@ -86,9 +86,9 @@ def bootstrap_confidence_interval(data, num_samples=1000, confidence_level=0.95)
mean = np.mean(means)
return mean, lower_bound, upper_bound

def calculate_confidence_interval(data):
def calculate_confidence_interval(data, min_is_best=True):
mean, lower, upper = bootstrap_confidence_interval(data)
min_value = np.min(data)
min_value = np.min(data) if min_is_best else np.max(data)
return mean, (upper - lower) / 2, min_value


Expand Down Expand Up @@ -117,7 +117,7 @@ def main():


total_time, total_ci, total_best = calculate_confidence_interval(np.sum(all_times, axis=1))
ops_per_sec, ops_per_sec_ci, ops_per_sec_best = calculate_confidence_interval(float(all_times.shape[1]) / np.sum(all_times / 1000, axis=1))
ops_per_sec, ops_per_sec_ci, ops_per_sec_best = calculate_confidence_interval(float(all_times.shape[1]) / np.sum(all_times / 1000, axis=1), min_is_best=False)
min_time, min_ci, _ = calculate_confidence_interval(np.min(all_times, axis=1))
mean_time, mean_ci, _ = calculate_confidence_interval(np.mean(all_times, axis=1))
median_time, median_ci, _ = calculate_confidence_interval(np.median(all_times, axis=1))
Expand Down
9 changes: 6 additions & 3 deletions src/benchmarks/bench.cpp
Original file line number Diff line number Diff line change
Expand Up @@ -134,6 +134,7 @@ struct ConfidenceInterval
double mean;
double confidence;
double min;
double max;
};

// Helper function to calculate the bootstrap confidence interval
Expand Down Expand Up @@ -161,8 +162,10 @@ ConfidenceInterval confidenceInterval(const std::vector<double> &data,
double upper_bound = means[(int)((1 + confidence_level) / 2 * num_samples)];
double mean = std::accumulate(means.begin(), means.end(), 0.0) / means.size();

ConfidenceInterval ci = {
mean, (upper_bound - lower_bound) / 2, *std::min_element(data.begin(), data.end())};
ConfidenceInterval ci = {mean,
(upper_bound - lower_bound) / 2,
*std::min_element(data.begin(), data.end()),
*std::max_element(data.begin(), data.end())};
return ci;
}

Expand Down Expand Up @@ -260,7 +263,7 @@ std::ostream &operator<<(std::ostream &os, Statistics &statistics)
ConfidenceInterval ops_ci = statistics.ops_per_sec();

os << "ops: " << ops_ci.mean << " ± " << ops_ci.confidence << " ops/s. "
<< "best: " << ops_ci.min << "ops/s." << std::endl;
<< "best: " << ops_ci.max << "ops/s." << std::endl;
os << "total: " << total_ci.mean << " ± " << total_ci.confidence << "ms. "
<< "best: " << total_ci.min << "ms." << std::endl;
os << "avg: " << mean_ci.mean << " ± " << mean_ci.confidence << "ms" << std::endl;
Expand Down

0 comments on commit cacb1b2

Please sign in to comment.