A few performance improvements when running CoCiP with a Fleet - Githubissues

contrailcirrus / pycontrails

Python library for modeling contrails and other aviation climate impacts

https://py.contrails.org/

Apache License 2.0

59 stars 17 forks source link

A few performance improvements when running CoCiP with a Fleet #221

Closed zebengberg closed 3 months ago

zebengberg commented 4 months ago

Fixes

Fix the step_threshold in Flight._altitude_interpolation. This correction is only relevant when passing in a non-default value for freq in Flight.resample_and_fill.
Fix the VectorDataset.__eq__ method to check for the same keys. Previously, if the other dataset had a superset of the instance keys, the method may still return True.

Internals

Improve the runtime performance of Fleet.to_flight_list. For a large Fleet, this method is now 5x faster. The plot below shows my benchmarking.
Improve the runtime performance and memory footprint of Cocip._bundle_results. When running Cocip with a large Fleet source, Cocip.eval is now slightly faster and uses much less memory.

Tests

[ ] QC test passes locally (make test)
[ ] CI tests pass

zebengberg commented 3 months ago

One question: does our definition of a "breaking change" just include changes to intended behavior, or also bug fixes that change behavior for cases affected by the bug? (The step_threshold changes are in the second category, and I'm not sure if we should note them in the breaking changes section of the changelog.)

You're absolutely right -- we often blur this line. In this case, I think it's enough to note the change just in the fix category ... but it's definitely a matter of opinion.