Add a new DataFrame.dti accessor

[ghost]

There's the dt accessor on Series which operates on the Series values:

pd.Series([1,2,3]).dt.day 
#exists. But checks that series *values*  datetime64[ns]

pd.Series([1,2,3]).day
# doesn't exist

There's a top-level method df.first:

Convenience method for subsetting initial periods of time series data
based on a date offset.

which looks like it should live behind dti accessor on DataFrame.

pd.DataFrame().first(0) 
# method is applicable only for datetime, but is in the top DataFrame namespace

There's currently no datetime accessor on dataframe. Would it make sense to move it below a dti accessor on DataFrame?

There's also an unfortunate unfortunate name collision betweenGroupby().first

Compute first of group values.

and DataFrame.first:

Convenience method for subsetting initial periods of time series data
based on a date offset.

Adding a dti accessor to DataFrame would clean that up as well.

[WillAyd]

Not sure about the idea of a data frame accessor. In the Series world when thinking about an accessor it operates on the values; this would essentially operate on the index of the object so I think not super clean from an API perspective.

Agreed though the naming of first isn't intuitive and does clash with other items in the API. Have you thought of anything outside the accessor? Maybe we just need to rename the function (the more verbose first_offset?)

[ghost]

That's a good point. I was thinking of accessors as a way to offer methods conditionally, which fits here.

this would essentially operate on the index of the object

it would slice the DataFrame object based on the index, yes. But does that count or not?

i = pd.date_range('2018-04-09', periods=4, freq='2D')
ts = pd.DataFrame({'A': [1,2,3,4]}, index=i)
# Get the rows for the first 3 days:

ts.first('3D')
            A
2018-04-09  1
2018-04-11  2

For one method, doing this wouldn't make sense. I'll see if there are others, and if not perhaps a name change really is the simplest solution. Thanks.

[WillAyd]

Yea my wording was slightly off but I think we are on the same page. If you find others lmk otherwise maybe we just repurpose this issue to focus solely on the first naming inconsistency

[ghost]

ok, sounds good.

[ghost]

first       
asfreq
asof 
at_time
between_time
tshift 
tz_convert
tz_localize 

all the above are DataFrame methods which are candidates for moving beneath a dti accessor.

We might want to rename first even if it were moved.

[mroeschke]

Thanks for the suggestion, but sounds like the community and other core devs have not expressed further interest in a dti accessor. Closing but happy to reopen if theres further interest