Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

BUG: x in MultiIndex.drop(x) #19054

Merged
merged 10 commits into from
Jan 10, 2018
Merged

BUG: x in MultiIndex.drop(x) #19054

Show file tree
Hide file tree
Changes from 4 commits
Commits
Show all changes
10 commits
Select commit Hold shift + click to select a range
129b286
BUG: x in MultiIndex.drop(x)
Jan 3, 2018
9aec9de
fixing stack and typeerror checks
Jan 3, 2018
35ee8f0
added test
Jan 3, 2018
698853f
made fixes, added additional test
Jan 5, 2018
5da9830
updated failing test based on new behavior
Jan 5, 2018
a0495a4
Merge updated 'master' into x-in-MultiIndex-drop-x
Jan 5, 2018
552392f
add/fix comments and fix merge conflict
Jan 5, 2018
22d3f69
Merge branch 'master' into PR_TOOL_MERGE_PR_19054
jreback Jan 6, 2018
9b79e96
remove redundant dropped columns conditional
Jan 7, 2018
a516fcc
Merge origin branch.
Jan 7, 2018
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion doc/source/whatsnew/v0.23.0.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -321,7 +321,7 @@ Indexing
- Bug in indexing non-scalar value from ``Series`` having non-unique ``Index`` will return value flattened (:issue:`17610`)
- Bug in :func:`DatetimeIndex.insert` where inserting ``NaT`` into a timezone-aware index incorrectly raised (:issue:`16357`)
- Bug in ``__setitem__`` when indexing a :class:`DataFrame` with a 2-d boolean ndarray (:issue:`18582`)

- Bug in :func:`MultiIndex.__contains__` where non-tuple keys would return ``True`` even if they had been dropped (:issue:`19027`)

I/O
^^^
Expand Down
4 changes: 4 additions & 0 deletions pandas/core/indexes/multi.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2122,6 +2122,10 @@ def _maybe_to_slice(loc):

if not isinstance(key, tuple):
loc = self._get_level_indexer(key, level=0)
# _get_level_indexer returns an empty slice if the key has
# been dropped from the MultiIndex
if isinstance(loc, slice) and loc.start == loc.stop:
raise KeyError(key)
return _maybe_to_slice(loc)

keylen = len(key)
Expand Down
6 changes: 5 additions & 1 deletion pandas/core/reshape/reshape.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -628,7 +628,11 @@ def _convert_level_number(level_num, columns):
levsize = len(level_labels)
drop_cols = []
for key in unique_groups:
loc = this.columns.get_loc(key)
try:
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

what test exercises this?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

pandas/pandas/tests/test_multilevel.py

Line 1195 in 35b2aba

def test_stack_order_with_unsorted_levels(self):

pandas/pandas/tests/frame/test_reshape.py

Line 136 in 35b2aba

def test_stack_mixed_level(self):

pandas/pandas/tests/frame/test_reshape.py

Line 730 in 35b2aba

def test_stack_partial_multiIndex(self):

These were the 3. And it looks like Travis doesn't have permalinks to specific lines like GitHub?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Also just took a look at the test that caused most of the builds to fail:

pandas/pandas/tests/frame/test_mutate_columns.py

Lines 186 to 198 in 35b2aba

del df['A']
assert len(df.columns) == 2
# A still in the levels, BUT get a KeyError if trying
# to delete
assert ('A', ) not in df.columns
with pytest.raises(KeyError):
del df[('A',)]
# xref: https://github.com/pandas-dev/pandas/issues/2770
# the 'A' is STILL in the columns!
assert 'A' in df.columns

and we just changed this so I'll go ahead and negate the assert.

loc = this.columns.get_loc(key)
except KeyError:
drop_cols.append(key)
continue

# can make more efficient?
# we almost always return a slice
Expand Down
18 changes: 18 additions & 0 deletions pandas/tests/indexing/test_multiindex.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -705,6 +705,24 @@ def test_multiindex_symmetric_difference(self):
result = idx ^ idx2
assert result.names == [None, None]

def test_multiindex_contains_dropped(self):
# GH 19027
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

can you add a comment

idx = MultiIndex.from_product([[1, 2], [3, 4]])
assert 2 in idx
idx = idx.drop(2)

# drop implementation keeps 2 in the levels
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

blank line before the comment

assert 2 in idx.levels[0]
# but it should no longer be in the index itself
Copy link
Contributor

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

can you test with a non-integer MI as well

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This comment made me realize that I had focused onto integers specifically when this should apply to all hashable keys (sorry). But, that would mean tuples would then be treated differently from how they currently are, e.g.:

In [88]: idx = pd.MultiIndex.from_product([[(1, 2), (2, 3)], [(3, 4), (4, 5)]])

In [89]: ((1, 2), (3, 4)) in idx
Out[89]: True

In [90]: (1, 2) in idx
Out[90]: False

In [91]: ((1, 2),) in idx
Out[91]: True

So, should it assume that a nested tuple like in In [89] should be treated like tuples are currently treated but a non-nested tuple like in In [90] should be treated like strings and ints?

Copy link
Contributor Author

@databasedav databasedav Jan 4, 2018 •
edited
Loading

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Off the same vein, idx.drop((1, 2)) raises a KeyError and might warrant a separate issue since one would expect similar behavior to MultiIndex.from_product([[1, 2], [3, 4]]).drop(2), right?

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This comment made me realize that I had focused onto integers specifically when this should apply to all hashable keys (sorry).

I think the idea should just be: any key k which is not already a tuple must become one ((k,)). The result will be, I think, consistent with current behavior (Out[89]:, Out[90]:, Out[91]:), and you don't even need to check if key can be hashed - if it can't, it will just fail later on.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Off the same vein, idx.drop((1, 2)) raises a KeyError and might warrant a separate issue

Sure, in principle we could try with ((1,2),) if (1,2) is not found. However, I personally don't think it's a good idea, as it introduces ambiguity (we should be clear to users that a tuple in a MultiIndex is first and foremost a key spanning across levels) and increases complexity.

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Got it, thanks!
Doing it by simply turning non-tuples into tuples like so:

if not isinstance(key, tuple):
    try:
        return self.get_loc((key,))
    except (LookupError, TypeError):
        raise KeyError(key)

produces a performance warning for slightly larger multi-indeces:

In [31]: idx = pd.MultiIndex.from_product([[1, 2], [3, 4]])

In [32]: 2 in idx
Out[32]: True

In [33]: idx = pd.MultiIndex.from_product([[2, 1, 2], [3, 4]])

In [34]: 2 in idx
/home/avi/anaconda3/envs/pandas-dev/lib/python3.6/site-packages/IPython/terminal/ipapp.py:356: PerformanceWarning: indexing past lexsort depth may impact performance.
  self.shell.mainloop()
Out[34]: True

Or we could use the fact that a key that has been dropped from a multi-index will return an empty slice when fed into the index's _get_level_indexer like so:

if not isinstance(key, tuple):
    loc = self._get_level_indexer(key, level=0)
    if isinstance(loc, slice) and loc.start == loc.stop:
        raise KeyError(key)
    return _maybe_to_slice(loc)

which does not produce the performance warnings and is ~3 times faster and simply adds to what it is right now:

pandas/pandas/core/indexes/multi.py

Lines 2123 to 2125 in 4a8496b

if not isinstance(key, tuple):
loc = self._get_level_indexer(key, level=0)
return _maybe_to_slice(loc)

but is potentially less clear (and more hacky?). Which one should I use?

Copy link
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

I ended up using the faster one and added a comment to explain what's going on.

Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Yes, I think it makes sense

assert 2 not in idx

# also applies to strings
idx = MultiIndex.from_product([['a', 'b'], ['c', 'd']])
assert 'a' in idx
idx = idx.drop('a')
assert 'a' in idx.levels[0]
assert 'a' not in idx


class TestMultiIndexSlicers(object):

Expand Down