unumpy.sin() and similar functions should behave with Pandas object like NumPy functions do

[lebigot]

More specifically, numpy.sin() applied to a Pandas Series returns a Pandas Series, whereas unumpy.sin() returns a NumPy array.

See #133 (comment).

[MichaelTiemannOSC]

I'm looking to do two things:

1. Use uncertainties as a Pandas ExtensionDtype (see https://stackoverflow.com/questions/68893521/simple-example-of-pandas-extensionarray)
2. Use the new Pandas uncertainties ExtensionDtype in PintPandas do to unit-aware calculations with uncertainties (see Pint+Pandas+Uncertainties...how would you ... hgrecco/pint#1605)

Anybody else interested/able to help and/or guide? This is mostly new territory for me (but I have done big lifts in the past).

[MichaelTiemannOSC]

This is a nuisance that will have to be fixed: https://stackoverflow.com/questions/70908349/pandas-rolling-operation-on-categorical-column

[lebigot]

I'd be happy to punctually give a hand and brainstorm some; precisely, if you keep posting in the issues, I'll do my best to help.

I don't know how related this can be, but uncertainties could use a simpler interface with NumPy, and maybe this would help with Pandas as well? Some initial brainstorming can be found in #47.

[MichaelTiemannOSC]

Wonderful! I just read through that history. Thus far, I've made very good progress by simply deleting from PintType (in Pint-Pandas) the code that forces conversion to float when casting via astype (see https://github.com/hgrecco/pint-pandas/blob/master/pint_pandas/pint_array.py#:~:text=if%20not%20np,copy%20%3D%20False).

I'm now confronting how code I wrote messes with Pint's assumption that it can read and understand what it writes. I think it's relatively common idiom (unless I'm an idiot) to create a quantity somewhere and then read that quantity back via the unit registry like so:

ureg(str(Q_(1, 'kt CO2e')))
# prints <Quantity(1, 'CO2e * kt')>

When the unit registry sees an uncertainty, it gets confused:

ureg(str(Q_(ufloat(1.0, 0.1), 'kt CO2e')))
*** pint.errors.DefinitionSyntaxError: missing unary operator "/"

But so much else works, except for needing to change my nan handling (unp.isnan). I'll report more as I get my code passing its first full run-trough of a test case.

[MichaelTiemannOSC]

Made some good progress, updated here: os-climate/ITR#159

Forks to Pint and Pandas mentioned in PR above and described at bottom of this thread: hgrecco/pint#1605