Benchmarking NumPy with Airspeed Velocity.
Airspeed Velocity manages building and Python virtualenvs by itself, unless told otherwise. Some of the benchmarking features in
runtests.py also tell ASV to use the NumPy compiled by
runtests.py. To run the benchmarks, you do not need to install a development version of NumPy to your current Python environment.
Run a benchmark against currently checked out NumPy version (don’t record the result):
python runtests.py --bench bench_core
Compare change in benchmark results to another version:
python runtests.py --bench-compare v1.6.2 bench_core
Run ASV commands (record results and generate HTML):
cd benchmarks asv run --skip-existing-commits --steps 10 ALL asv publish asv preview
More on how to use
asv can be found in ASV documentation Command-line help is available as usual via
asv --help and
asv run --help.
See ASV documentation for basics on how to write benchmarks.
Some things to consider:
- The benchmark suite should be importable with any NumPy version.
- The benchmark parameters etc. should not depend on which NumPy version is installed.
- Try to keep the runtime of the benchmark reasonable.
- Prefer ASV’s
time_methods for benchmarking times rather than cooking up time measurements via
time.clock, even if it requires some juggling when writing the benchmark.
- Preparing arrays etc. should generally be put in the
setupmethod rather than the
time_methods, to avoid counting preparation time together with the time of the benchmarked operation.
- Be mindful that large arrays created with
np.zerosmight not be allocated in physical memory until the memory is accessed. If this is desired behaviour, make sure to comment it in your setup function. If you are benchmarking an algorithm, it is unlikely that a user will be executing said algorithm on a newly created empty/zero array. One can force pagefaults to occur in the setup phase either by calling
arr.fill(value)after creating the array,