Infer ExtensionDtype based on object

[andrewgsavage]

Would it be possible to infer ExtensionDtypes based on the objects inside a list/array that are used to construct a series?

Eg with an array of pint's Quantity objects:

>>> array_of_qty
array([<Quantity(67.0, 'meter')>, <Quantity(61.0, 'meter')>,
       <Quantity(30.0, 'meter')>, ..., <Quantity(293.0, 'meter')>,
       <Quantity(886.0, 'meter')>, <Quantity(165.0, 'meter')>],
      dtype=object)

Could pandas infer the dtype as PintType in this command?:

df.elevation = array_of_qty

[jbrockmendel]

There has been some discussion about this, and so far the consensus is that this would be worthwhile if we can find a performant way to implement it. A PR with a proof of concept might be helpful.

[TomAugspurger]

This probably boils down to

pandas._libs.lib.infer_dtype(...)

being able to return an extension type. Then things should work pretty much everywhere in pandas.

Also, don't do df.elevation = array_of_qty (That should be triggering a warning) 😄

[andrewgsavage]

Hmm, that isn't triggering a warning - what are you expecting?