HOWTO.rst

Astropy coordinates benchmark HOWTO

Setting up an environment to run the benchmarks

If you want to set up a virtual environment to run the benchmarks, you can do:

virtualenv env
source env/bin/activate
pip install "numpy<1.18" Cython jinja2
pip install -r requirements.txt

Running

Go to the repository top-level folder (where the ./make.py file is). The ./make.py command line tool gives you access to all functionality.

To print its help message which lists all sub-commands run:

./make.py

To run the benchmarks (output goes in output/tools):

./make.py benchmark-celestial

To generate a summary webpage (output goes in output):

./make.py summary-celestial

To generate a ton of plots for the webpage (optional) (output goes in output/plots):

./make.py plots

Checking

The benchmark results are stored in this repository and have been obtained on one machine, but it's easy for anyone to re-run (and should be re-run from time to time, especially after changes in the astropy.coordinates code).

To check if benchmark or summary output matches results previously obtained, use git diff output/tools and git diff output/summary.txt.

TODO: describe how many decimal digits we store and why that a) is enough precision (< milli-arcsec) b) should be the same on any 64-bit computer

Deploying

To deploy the latest results to Github pages:

./make.py deploy

and if there are no errors:

git push upstream gh-pages