Features which Interval / IntervalIndex should probably have

[alexlenail]

Here's a list of possible features which should be added to the Interval and IntervalIndex types. Some of them might already exist, in which case, please excuse my mistake. Also note, I'm not asking for these, just listing some ideas I had which might be useful for others.

closest

>>> Interval(3, 4).closest(IntervalIndex.from_tuples([(1, 2), (5, 9)]), how="min|mean|max")
>>> IntervalIndex.from_tuples([(1, 2), (5, 9)]).closest(IntervalIndex.from_tuples([(1, 4), (2, 5)]))

Big question: what if there are two or more identically distant intervals?

complement

>>> IntervalIndex.from_tuples([(1, 2), (5, 9)]).complement(lower_bound=np.neginf, upper_bound=np.inf)
IntervalIndex.from_tuples([(np.neginf, 1), (2, 5), (9, np.inf)])

intersection

>>> IntervalIndex.from_tuples([(1, 2), (5, 9)]).intersect(IntervalIndex.from_tuples([(3, 4), (6, 7)]))
IntervalIndex.from_tuples([(6, 7)])
>>> IntervalIndex.from_tuples([(1, 2), (5, 9)]).intersect(Interval(6, 7))
???

union

>>> IntervalIndex.from_tuples([(1, 2), (5, 9)]).intersect(IntervalIndex.from_tuples([(3, 4), (6, 7)]))
IntervalIndex.from_tuples([(1, 2), (3, 4), (5, 9)])

subtract|difference

>>> IntervalIndex.from_tuples([(1, 2), (5, 9)]) - IntervalIndex.from_tuples([(3, 4), (6, 7)])
IntervalIndex.from_tuples([(1, 2), (5, 6), (7, 9)])
>>> IntervalIndex.from_tuples([(1, 2), (5, 9)]) - Interval(6, 7)
IntervalIndex.from_tuples([(1, 2), (5, 6), (7, 9)])
>>> Interval(6, 7) - IntervalIndex.from_tuples([(1, 2), (5, 9)])
???

sort

(I'm guessing this is probably already implicitly implemented. Does it work for multi-index in which one of the levels is Intervalindex?)

shift

>>> IntervalIndex.from_tuples([(1, 2), (5, 9)]).shift(2)
IntervalIndex.from_tuples([(3, 4), (7, 11)])

(Could be really useful for datetimes?)

slop|grow|window|better name

>>> IntervalIndex.from_tuples([(1, 2), (5, 9)]).slop(1)
IntervalIndex.from_tuples([(0, 3), (4, 10)])

[jschendel]

Note that union/intersection/difference/shift will likely need new names, as they already have more generic uses (e.g. IntervalIndex.union is already implemented with set theoretic behavior in terms of individual elements).

Some more ideas, all of which should return booleans, mostly taken from postgres range types:

• is_adjacent_to
• strictly_left_of
• strictly_right_of
• is_empty

(see the postgres docs for examples)

Another postgres function that could be nice is range_merge, which we'd almost certainly want to rename to something like combine. It gives the smallest interval containing two given intervals, e.g. Interval(0, 1).combine(Interval(3, 4)) would return Interval(0, 4).

[tsabsch]

I was currently working with an IntervalIndex and needed to get the interval ranges. Maybe this would be useful for others as well?

>>> IntervalIndex.from_tuples([(1, 2), (5, 9)]).ranges()
array([1, 4])

[jschendel]

@tsabsch : By ranges, you're referring to the length of each interval in the IntervalIndex, right?

If so, note that this was implemented as the length property in #18805, and will be present in 0.23.0:

In [2]: pd.__version__
Out[2]: '0.23.0.dev0+230.gf391cbf'

In [3]: pd.IntervalIndex.from_tuples([(1, 2), (5, 9)]).length
Out[3]: Int64Index([1, 4], dtype='int64')

[tsabsch]

@jschendel Yes, this is exactly what I meant. Thank you for pointing it out. I wasn't aware that it was already implemented for the next release. I'm on pandas 0.22.0.

[jschendel]

@tsabsch : No worries. Glad to hear length will be useful!

[mezzode]

It might also be nice to be able to check if an Interval is in another Interval.
Also, this issue could use the Interval label for discoverability.