Infer Decimal dtype

[dhirschfeld]

In [15]: from decimal import Decimal

In [16]: s = pd.Series([Decimal(3), Decimal(1), Decimal(4)])
    ...: s
Out[16]: 
0    3
1    1
2    4
dtype: object

In [17]: s.dtype
Out[17]: dtype('O')

In [18]: pd.lib.infer_dtype(s)
Out[18]: 'mixed'

In [19]: set(map(type, s))
Out[19]: {decimal.Decimal}

It would be nice if the infer_dtype function could determine if the column was all of decimal type which could be useful when working with databases which return data as decimal types

[jreback]

you can do this in master

In [2]: from pandas.types.common import is_decimal

n [4]: s.apply(is_decimal).all()
Out[4]: True

your suggestion is quite easy to do if you'd like to submit a PR! (FYI, pandas._libs/src/inference.pyx is the new home of infer_dtype)

[margaret]

I'm going to work on this (pycon sprints)

[jorisvandenbossche]

@jreback Repeating my comment from on the PR: What I am wondering is what are potential consequences of such a change. As there are quite some places in the codebase where we wheck if the inferred type is 'mixed', so changing this would mean potential changes in the taken code path for Decimal objects (with potential changes in behaviour).
But don't know if those cases in the code would be relevant for Decimal objects.

[jreback]

@jorisvandenbossche #16426 (comment)

after the PR is merged, should create an issue to use the new inference (though as I pointed, maybe not that useful ATM).