API: expose to_numeric/to_datetime/to_timedelta methods on Series

[jreback]

if we have a string Series, e.g. s = Series(['2013-01-02', '2013-1-03'])

we normally just run this thru pd.to_datetime(s)

it might be convenient to do these operations (.to_numeric,.to_datetime,.to_timedelta)
directly as methods.

• s.str.to_numeric(....) , IOW in the .str namespace
• s.to_numeric(...), in main Series namespace
• s.astype(.., dtype=...) but actually pass thru to using the .to_numeric/.to_datetime functions directly, so you could have something like
  s.astype('datetime', format='%Y-%m-%d') and it would do the right thing. This overloads things a bit though

right now we have a bit of confusion because we support .astype which is pretty much a pass thru to numpy, and the more dtype-aware (and functional) .to_* ones.

Furthermore I think we could update the .astype docs (and the .to_*) ones to cross reference.

[chris-b1]

I was thinking this would be nice the other day. Don't want to get too carried away with the accesssors, but what about a .type (may be a better name) one to stuff these into?

s.type.astype(...) # numpy style
s.type.to_datetime(...)`
# maybe add?
s.type.to_category(...)

In general the current astype(...) / to_datetime(...) difference is a bit confusing.

[kernc]

+1 for simply overloading .astype(..., **kwargs) with kwargs passed to expecting functions. 👍

[topper-123]

I like @chris-b1 's idea, but I suggest the accessor be called .cast instead, because
IMO the name type can be confused with .dtype. So:

• s.cast.to_type,
• s.cast.to_datetime,
• etc.

I'm +1 on overloading .astype, i.e. making astype delegate to various subfunction depending on input. That would of course require a upgrade to the docs to avoid confusion.

I'm -1 on adding the functionality to dataframe/series namespace (e.g. s.to_datetime) as I think the API is already very large and we need to group similar functionality together to lessen the cognitive load of knowing the library.

[tdpetrou]

Bump!
The **kwargs are available in astype but apparently, they don't work for certain dtypes? The docstrings even mention the kwargs get passed to the constructor but this isn't happening for datetime64. And what is the constructor for datetime64? Is it to_datetime or Timestamp?

[mroeschke]

Looks like there's not much appetite to putting these methods on Series so closing