Python translation of NREL SPA code

[alorenzo175]

Since the NREL SPA C code isn't included in the repo anymore, I thought I would translate the algorithm to python. The translation produces the same results as the C code. When using numpy arrays and operations, this code is just as fast as the cythonized C code. When using numba, the import can take a while because it has to compile, but it is much faster for subsequent calls. Numba can also release the GIL and can use multiple threads.

I also made some changes to the spa C cython files and spa function to take pressure and temperature inputs (instead of hard coded values) and also output apparent zenith and elevation.

This isn't quite ready for a merge yet; I still want to get delta_t automatically from the USNO instead of having a set value or making the user set it.

In [21]:

print len(times)
%time npy = spa_python(times, loc, pressure=82000, temperature=11, delta_t=67.0, how='numpy')
DEBUG:pvlib:Calculating solar position with spa_python code
DEBUG:pvlib:tz_convert to UTC
DEBUG:pvlib:Reloading spa module without compiling
1441
CPU times: user 51 ms, sys: 3 ms, total: 54 ms
Wall time: 52.3 ms
In [22]:

print len(times)
%time nba = spa_python(times, loc, pressure=82000, temperature=11, delta_t=67.0, how='numba', numthreads=1)
DEBUG:pvlib:Calculating solar position with spa_python code
DEBUG:pvlib:tz_convert to UTC
DEBUG:pvlib:Reloading spa module, compiling with numba
DEBUG:pvlib:Only using one thread for calculation
1441
CPU times: user 3.72 s, sys: 16 ms, total: 3.74 s
Wall time: 3.74 s
In [23]:

print len(times)
%time nba = spa_python(times, loc, pressure=82000, temperature=11, delta_t=67.0, how='numba', numthreads=1)
DEBUG:pvlib:Calculating solar position with spa_python code
DEBUG:pvlib:tz_convert to UTC
DEBUG:pvlib:Only using one thread for calculation
1441
CPU times: user 34 ms, sys: 2 ms, total: 36 ms
Wall time: 35 ms
In [24]:

print len(times)
%time nba = spa_python(times, loc, pressure=82000, temperature=11, delta_t=67.0, how='numba', numthreads=4)
DEBUG:pvlib:Calculating solar position with spa_python code
DEBUG:pvlib:tz_convert to UTC
1441
CPU times: user 39 ms, sys: 1 ms, total: 40 ms
Wall time: 12.4 ms
In [25]:

print len(times)
%time eph = get_solarposition(times, loc, 'pyephem', pressure=82000, temperature=11)
DEBUG:pvlib:using PyEphem to calculate solar position
DEBUG:pvlib:tz_convert to UTC
1441
CPU times: user 155 ms, sys: 1 ms, total: 156 ms
Wall time: 154 ms
In [27]:

%time spac = get_solarposition(times, loc, 'spa_c', pressure=82000, temperature=11)
DEBUG:pvlib:using built-in spa code to calculate solar position
DEBUG:pvlib:tz_convert to UTC
CPU times: user 49 ms, sys: 3 ms, total: 52 ms
Wall time: 49.8 ms

[wholmgren]

This is very cool and very well done. To be clear, @alorenzo175 used NREL's paper, not their code, to make this.

[wholmgren]

Would have liked to comment on the diff, but my computer/browser chokes on long diffs...

On get_solarposition, I suggest that:

• 'spa' is removed, or it stays but as the default and performs a try: spa_numba; except: spa_python.
• If 'spa' is removed, 'spa_python' is the default.
• 'spa_c' becomes 'spa_nrel_c'.
• maybe add **kwargs to this and pass to worker functions?

I also suggest that the function spa becomes spa_nrel_c.

Would like to hear others input on the above (cc @Calama-Consulting @bmu).

In spa_python:

• First sentence in the docstring implies that this is modeled on the NREL C code instead of the NREL paper.
• atmos_refrac is not in the docstring.
• The how block needs a few comments for why it uses this logic and how it works with the other modules. I think I understand after studying the code, but import logic and decorators is definitely not beginner Python.
• how docstring is not consistent with the end of the how block.

[bmu]

This is great!! (I tried to do this myself a while ago, but I gave up after a while!)

One question:
A lot of the spa code is on time calculations. Do you think it is possible to separate the time calculations and the calculation of the sun position? I think (but I'm not sure), that true solar time is called apparent sideral time in the code (the code uses naming conventions of astronomers, and I am not an expert in this field ;-). see #47

[alorenzo175]

'spa' is removed, or it stays but as the default and performs a try: spa_numba; except: spa_python.
If 'spa' is removed, 'spa_python' is the default.

I would recommend that spa_numpy is the default. Numba can take a few seconds to compile and the numpy code is pretty fast. If someone is doing a lot of calculations and needs the speed, they can specify the use of numba and many threads.

'spa_c' becomes 'spa_nrel_c'.

I thought changing 'spa' to something else, but I didn't want to break the API. I think we should all agree on if/how to change it.

maybe add **kwargs to this and pass to worker functions?

Good idea.

[alorenzo175]

One question:
A lot of the spa code is on time calculations. Do you think it is possible to separate the time calculations and the calculation of the sun position? I think (but I'm not sure), that true solar time is called apparent sideral time in the code (the code uses naming conventions of astronomers, and I am not an expert in this field ;-). see #47

It is definitely possible to separate out the time calculations. I'll play around with it a little to see the best way to do it.

[alorenzo175]

I added a equation_of_time output from the solar position algorithm. I think any conversion to true solar time should be another function in solarposition.

I also renamed spa to spa_c, made the numpy calculation the default method to remove the pyephem dependency, and added a function to get sunrise, sunset, and sun transit time.

I think it is ready to merge as soon as it is reviewed.

[wholmgren]

This looks good to me. @bmu or @Calama-Consulting?

[bmu]

I don't see, were equation_of_time is returned and the code is to complex at a first look, so I cannot say I have understood everything. However I would merge it in its present form.

[wholmgren]

If anybody wants something changed further then they can make a new issue for it.

[uvchik]

There are no errors but unused imports (import warnings, import pytz) and unused variables e using the following expression: except ImportError as e but the e is not used anymore.

Furthermore there some PEP8 issues but I don't know how strict we wanna be.

[uvchik]

@alorenzo175
Do you want to add this or shall I create a new pull request about these issues.

[alorenzo175]

I can change it quickly.