implement .dt accessor for Index classes

[jbrockmendel]

index.dt --> self for DatetimeIndex, TimedeltaIndex, PeriodIndex. Raises AttributeError for others (identical error to Series.dt).

@jorisvandenbossche in the discussion in #17134 you had a concern about tab completion/namespacing. Is that a sticking point for you?

• closes DISC: add accessor attributes to Index for consistency with Series #17134
• tests added / passed
• passes git diff upstream/master -u -- "*.py" | flake8 --diff
• whatsnew entry

[codecov]

Codecov Report

Merging #18429 into master will decrease coverage by 0.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master   #18429      +/-   ##
==========================================
- Coverage   91.35%   91.33%   -0.02%     
==========================================
  Files         163      163              
  Lines       49714    49719       +5     
==========================================
- Hits        45415    45410       -5     
- Misses       4299     4309      +10

Flag	Coverage Δ
#multiple	89.13% <100%> (-0.01%)	⬇️
#single	39.64% <60%> (-0.06%)	⬇️

Impacted Files	Coverage Δ
pandas/core/indexes/datetimelike.py	97.13% <100%> (+0.01%)	⬆️
pandas/core/indexes/base.py	96.43% <100%> (ø)	⬆️
pandas/io/gbq.py	25% <0%> (-58.34%)	⬇️
pandas/io/json/table_schema.py	94.44% <0%> (-1.39%)	⬇️
pandas/core/frame.py	97.8% <0%> (-0.1%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d421a09...bda766e. Read the comment docs.

[codecov]

Codecov Report

Merging #18429 into master will decrease coverage by 0.01%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master   #18429      +/-   ##
==========================================
- Coverage   91.35%   91.33%   -0.02%     
==========================================
  Files         163      163              
  Lines       49714    49719       +5     
==========================================
- Hits        45415    45410       -5     
- Misses       4299     4309      +10

Flag	Coverage Δ
#multiple	89.13% <100%> (-0.01%)	⬇️
#single	39.64% <60%> (-0.06%)	⬇️

Impacted Files	Coverage Δ
pandas/core/indexes/base.py	96.43% <100%> (ø)	⬆️
pandas/core/indexes/datetimelike.py	97.13% <100%> (+0.01%)	⬆️
pandas/io/gbq.py	25% <0%> (-58.34%)	⬇️
pandas/io/json/table_schema.py	94.44% <0%> (-1.39%)	⬇️
pandas/core/frame.py	97.8% <0%> (-0.1%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update d421a09...bda766e. Read the comment docs.

[jorisvandenbossche]

@jorisvandenbossche in the discussion in #17134 you had a concern about tab completion/namespacing. Is that a sticking point for you?

Yes, I think so (and not only for tab completion). Doing df.index.dt.mean() should be invalid IMO.

[jbrockmendel]

Yes, I think so (and not only for tab completion). Doing df.index.dt.mean() should be invalid IMO.

OK. Would returning self.to_series().dt solve your concerns? @jreback back in the issue you were pretty adamant that returning anything other than self would be inconsistent. Can you two reach a consensus on this?

[jorisvandenbossche]

Would returning self.to_series().dt solve your concerns?

No, because that would return a Series, while the Index.dt. attributes should return an Index IMO (see #17134 (comment), which @jreback agrees with).
So it's more an implementation question: it should be like returning self, but a restricted version.

Eg we could add a switch to _delegate_method to return either raw index result or wrapped in Series ?

[jbrockmendel]

No, because that would return a Series, while the Index.dt. attributes should return an Index IMO

I had hoped to avoid introducing a new class, but that should be doable. There is a lot of machinery in indexes.datetimelike, some of which can probably be made useful.

[jreback]

how is this any more complicated than for .str?
str = accessor.AccessorProperty(strings.StringMethods)

[jbrockmendel]

how is this any more complicated than for .str?
str = accessor.AccessorProperty(strings.StringMethods)

StringMethods has special handling for cases where it is derived from a Series vs Index. I can do something analogous in indexes.accessors, but last time I tried that you said it was unnecessarily complicated and should just return self (in fairness, that implementation was unnecessarily complicated and returned a Series instead of an Index...)

So let's nail down the specs before I implement anything else. It modifying indexes.accessors to behave analogously to StringMethods going to make everyone happy?

[jreback]

So let's nail down the specs before I implement anything else. It modifying indexes.accessors to behave analogously to StringMethods going to make everyone happy?

I think these should all be consistent, so yes, not sure there should actually be that much complexity.

[jreback]

closing as stale, if you want to work on this, pls ping.

I think the idea of this PR is fine, but needs resolution