JB2008 in Python, ~11x faster

JB2008 is the standard empirical model of thermospheric density, and a common baseline for ML density models. The Python port (PyAtmos) is convenient but slow for the hundreds of thousands of predictions that statistical or ML studies need.

I vectorized the input and data-preparation functions and used swifter (which falls back to Dask when it pays off) to parallelize the rest. On Google Colab, ~70,000 predictions went from nearly 8 minutes to about 40 seconds, and around 20 seconds on a faster machine. Details are in the write-up.

Write-up · GitHub