Benchmarks

bootstrapx focuses on CPU-portable performance that works on most machines.

What is fast

• Batched iid resampling for scalar statistics.
• Vectorized statistics where the user provides a batch-aware callable.
• Time-series sieve implementation with efficient filtering-based generation.

Benchmark philosophy

Compare methods under the same:

• sample size
• number of resamples
• statistic
• confidence method
• random seed when possible

Reproducible benchmark command

python benchmarks/bench_speed.py

What to report

When publishing benchmark numbers, include:

• CPU model
• Python version
• NumPy / SciPy version
• method name
• n_resamples
• sample size n

This keeps performance claims credible and repeatable.