@efiring's review and suggestions for gamma_GP_from_SP_pt.py #7
Description
(Trivial first aesthetic tweak: use a list comprehension or generator in the first line. But maybe all this would be handled in a decorator eventually anyway.)
In this case, the simplest change I would make is to leave the coefficient tables as lists of lists, or lists of tuples. Converting to ndarrays is not helping anything, and is actually adding a tiny bit of overhead.
The loops could use the more idiomatic form:
gamma = np.zeros_like(SP) for i, j, c in Fit: gamma += c * (SP**i * pt**j)Going farther, I would take all the Fit tables out of the functions, make them named module-level variables, and probably collect them in a dictionary. (Same for the polygons, which I would probably save as 2-D ndarrays--I think shapely can handle them efficiently.) Then I would use a single function to handle the loop, with the Fit table as a third argument. This would allow for future optimization of that single critical function, if desired; it would clarify the common logic; it would minimize repetition; and it would be slightly faster because all the table initialization would be done only once, at import time.
Additionally, I would look at the main function to see whether it is really necessary to calculate gamma for all the oceans; I suspect one could calculate weights first, and omit the gamma calculation for any region with zero weight.
Is some normalization of the input lat/lon range needed to make everything work with the region polygons? Are there potential problems with crossing the dateline, or the prime meridian?