Pandas ENH: Generalized head and tail method.

Is your feature request related to a problem?

I wondering if you maybe interested in include a new method context (I've chosen that name but can be other) which provides convenient slicing around an index (positional or not) for better data exploration. I see it as head or tail methods but generalized where the anchor can be other than the first or last index.

Describe the solution you'd like

I've implemented this https://github.com/mmngreco/pandas-context as a proof of concept which has the following code:

import pandas as pd


def context(
    pdobj, loc=None, iloc=None, around=2, pre=None, post=None, method="nearest"
):
    if loc is not None:
        idx = pdobj.index.get_loc(loc, method=method)
    elif iloc is not None:
        idx = iloc
    else:
        raise ValueError

    if pre is None:
        pre = around
    if post is None:
        post = around
    if pre > idx:
        pre = idx
    ini = idx - pre
    end = idx + post + 1
    return pdobj.iloc[ini:end]


pd.core.frame.NDFrame.context = context

Here're some examples of usage:

>>> import pandas as pd
... import numpy as np
... import pandas_context
... n = 1000   # long data
... data = np.random.randn(n)
... index = pd.date_range("19990101", periods=n)
... df = pd.DataFrame(data=data, index=index)
... date = index[50]

>>> date
Timestamp('1999-02-20 00:00:00', freq='D')

>>> df.context(date)
                   0
1999-02-18  1.086181
1999-02-19  0.174156
1999-02-20 -0.762305  # <-----
1999-02-21 -0.110677
1999-02-22 -0.766751

>>> df.context(date, around=1)
                   0
1999-02-19  0.174156
1999-02-20 -0.762305  # <-----
1999-02-21 -0.110677

>>> df.context(date, pre=3, post=1)
                   0
1999-02-17 -0.809791
1999-02-18  1.086181
1999-02-19  0.174156
1999-02-20 -0.762305  # <-----
1999-02-21 -0.110677

API breaking implications

[this should provide a description of how this feature will affect the API] None (I think)

Let me know if you are interested in this feature I will be glad to open a PR.

Comment From: jreback

i am almost always -1 on any new api

i do see the appeal here - but an insurmountable blocker is that this mixes positional and location based indexing and would be outright rejected on that alone

possible a .mid() that is only positional might gain some traction

Comment From: asdf8601

Ok, it makes sense. No problem on my side. So, the implementation would be something like this:

import pandas as pd


def mid(
    pdobj, loc=None, around=2, pre=None, post=None, method="nearest"
):

    idx = pdobj.index.get_loc(loc, method=method)

    if pre is None:
        pre = around
    if post is None:
        post = around
    if pre > idx:
        pre = idx
    ini = idx - pre
    end = idx + post + 1
    return pdobj.iloc[ini:end]


pd.core.frame.NDFrame.mid = mid

The implementation should extend the NDFrame class.

Comment From: jbrockmendel

-1 on new API here.