Relationship between extract_constants, read_constants and interpolate_constants

[robbibt]

Hi @tsutterley, I'm finally looking into refactoring some of our pyTMD tide modelling code to try and minimise IO by loading consistuents one time at the beginning of an analysis, interpolating values from those loaded consistuents in parallel, then model tides from those interpolated constituents as a final step.

It looks like model.read_constants and model.interpolate_constants are what I want, but I just wanted to check a few assumptions first:

1. Are read_constants and interpolate_constants in combination identical to extract_constants? e.g. are the following two equivelent? (in the code I can see that extract_constants doesn't call read_constants and interpolate_constants internally):

#1
amplitude, phase = extract_constants()

#2
c = read_constants()
amplitude, phase = interpolate_constants(c)

2. Do read_constants and interpolate_constants support the same overall params/functionality as extract_constants? I'm particularly thinking for things like crop, bounds etc - ideally I'd love to use read_constants to read in cropped constituents for my study area, then use that cropped data for the interpolation via interpolate_constants to keep run times quick.

[robbibt]

Actually, while looking into this I may have found a bug in the new model.extract_constants method; raised here: #356

[tsutterley]

Hey @robbibt

1. they should give identical answers and there are tests (e.g. here) to verify that they are consistent. The major difference between the workflows is that using read_constants and interpolate_constants will keep the (possibly cropped) constituents in memory. This can reduce computational time in the case where the same grids will be reused (no need to reread the data). extract_constants does not cache any of the constituent grids and so can be faster in the case where the data is only used once.
2. They should support the same overall functionality (e.g. crop and bounds). The read and reapply workflow is the target use case for read_constants and interpolate_constants.

[robbibt]

Fantastic, thanks @tsutterley - I have been meaning to refactor our workflows to use these sub-functions for a while but just wanted to check if my assumptions were all correct before starting on it. This sounds like it should do exactly what I am hoping for!