Add .cumulative to cumsum & cumprod docstrings

[max-sixty]

As discussed in a couple of issues, we should be directing folks towards the .cumulative. (the only missing piece is skip_na...

I thought this was a reasonable way to have the generation script work for these; ofc open to feedback.

I also added the namedarray file to the instructions for generating

[dcherian]

Note that the methods on the cumulative method are more performant and better supported

I'm surprised they're more performant than using numpy's cumprod, cumsum directly. Is this because cumulative redirects to numbagg?

[max-sixty]

Yes! :)

I don't have benchmarks at https://github.com/numbagg/numbagg, since cumulative().sum() just calls move_sum with the appropriate window, and numpy doesn't have a comparable function.

But to confirm — 10x faster one over 10 columns, 2x faster over 1 column:

[nav] In [36]: A = np.random.rand(60000, 10)

[ins] In [40]: %timeit np.cumsum(A, axis=0)
2.44 ms ± 82.8 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

[ins] In [47]: %timeit numbagg.move_sum(A, window=60000, min_count=0, axis=0)
371 µs ± 16.5 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

A = np.random.rand(60000, 1)

[nav] In [50]: %timeit np.cumsum(A, axis=0)
211 µs ± 5.34 µs per loop (mean ± std. dev. of 7 runs, 1,000 loops each)

[ins] In [49]: %timeit numbagg.move_sum(A, window=60000, min_count=0, axis=0)
106 µs ± 1.62 µs per loop (mean ± std. dev. of 7 runs, 10,000 loops each)

[dcherian]

we should clarify that then!

for the equivalent non-numbagg benchmark you could use xarray with use_numbagg=False?

[max-sixty]

we should clarify that then!

In the .cumulative docstring?

[max-sixty]

Merged, also improved the docs & suggested commands in generate_aggregations.py