Revert column to be Quantity rather than MaskedQuantity for Kepler/TESS LCs, i.e., pre v2.1 / Astropy5 behavior.

[orionlee]

This PR is to make Quantity-like columns for TESS / Kepler lightcurves, e.g., flux, sap_flux, etc. to be Quantity, rather than MaskedQuantity.

It essentially reverts to the behavior of Lightkurve 2.0.x with Astropy 4.x.
The reason is to avoid various complication with MaskedQuantity, most recently errors involving astropy sigma_clip() with numpy 1.24 .

If this PR is merged, it does not negate the recent commits and PR #1280 on the workarounds in calling sigma_clip(). The PR would just make them less urgent, as MaskedQuantity will no longer be in common use cases. (Users can still explicitly use MaskedQuantity for whatever reasons.)

[christinahedges]

Hi @orionlee thanks for bringing this up! In this PR, if a user were to input mask data e.g. because of a flare or a cosmic ray, this data would be over-written with a np.nan value. Users may find that this isn't what they want, e.g. they may want a data value temporarily masked for some analysis.

What do you think?

[orionlee]

@christinahedges Could you elaborate on the temporary mask use case you mentioned?

In my mind, the existing boolean indexing works just fine without using MaskedQuantity:

mask = some_function_to_define_cosmic_ray_mask(lc)
some_analysis_function(lc[~mask])  # temporary mask out cadences with cosmic ray

Using MaskedQuantity is arguably more cumbersome, e.g.,

# temporarily mask out flux at index 123
lc_tmp = lc.copy()
lc_tmp.flux.mask[123] = True
# if sap_flux is also used by the analysis, need to mask it too
lc_tmp.sap_flux.mask[123] = True
some_analysis_function(lc_tmp)

The change in this PR is only concerned with reading TESS / Kepler FITS file.

• The current behavior is to replace nan values with mask.
• This PR exposes the underlying nan values (lightkurve 2.0.x behavior)

It has nothing direct bearing on user temporarily masking out specific cadence / values, unless users assume columns are MaskedQuantity and use Masked-specific API such as .mask .

---

Irrespective on the merits of using MaskedQuantity

1. This PR does not stop users from using MaskedQuantity, but it is no longer the de-facto column type for TESS / Kepler LCs. If users have specific needs to use MaskedQuantity, they can still explicitly replace a column (of type Quantity) with the MaskedQuantity counterpart.
2. If it's decided that we do want to retain MaskedQuantity as the column type for TESS / Kepler LCs, I think we should:
   a. document it somewhere
   b. ensure we do always return MaskedQuantity. The current behavior is dependent on the contents of the specific FITS file: if a a column contains nan, the column will be MaskedQuantity, otherwise, it'd be regular Quantity.
   In other words, lightkurve does not guarantee a column would be MaskedQuantity. Users cannot exactly assume it to be the case. MaskedQuantity becomes de-facto type for Kepler/TESS LCs because in most cases, columns have nan values.

[christinahedges]

Hi @orionlee thanks for the explanation, when you say

The current behavior is to replace nan values with mask.

what do you mean here?

There is also a comment in the astropy Kepler file reader that this is a workaround they want to get rid of when masked quantities are supported better. Given that, do you think this is still a good fix to implement in lightkurve?