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

Jump to bottom

PERF: improves performance in GroupBy.cumcount #11039

Closed
wants to merge 1 commit into from
Closed

PERF: improves performance in GroupBy.cumcount #11039

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
43 changes: 42 additions & 1 deletion doc/source/whatsnew/v0.18.1.txt
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -138,6 +138,47 @@ API changes
- Provide a proper ``__name__`` and ``__qualname__`` attributes for generic functions (:issue:`12021`)
- ``pd.concat(ignore_index=True)`` now uses ``RangeIndex`` as default (:issue:`12695`)

.. _whatsnew_0181.enhancements.groubynth:

Index in ``Groupby.nth`` output is now more consistent with ``as_index``
argument passed in (:issue:`11039`):

Previous Behavior:

.. code-block:: ipython

In [4]: df
Out[4]:
A B
0 a 1
1 b 2
2 a 3

In [5]: df.groupby('A', as_index=True)['B'].nth(0)
Out[5]:
0 1
1 2
Name: B, dtype: int64
Copy link
Contributor

Choose a reason for hiding this comment

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

show the same as_index=False here as well (as u r showing below)

Copy link
Contributor Author

Choose a reason for hiding this comment

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

the point here is that as_index=True is ignored in old behaviour; as_index=False is not relevant or informative (and has not changed).

this has been going back and forth for too many times. if you like to add/modify anything please do so on your end.

Copy link
Contributor

Choose a reason for hiding this comment

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

If you are showing it in the new, then pls show it in the original.

this has been going back and forth for too many times.

Well, that's just how it is. The docs have to be in the proper format and be consistent.

Copy link
Contributor Author

Choose a reason for hiding this comment

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

@jreback please go ahead and make any changes you find necessary

Copy link
Contributor

Choose a reason for hiding this comment

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

@behzadnouri of course, but that's not the point is it. ok thank you for the PR.



New Behavior:

.. code-block:: ipython
Copy link
Contributor

Choose a reason for hiding this comment

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

use an ipython block

Copy link
Contributor

Choose a reason for hiding this comment

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

use ipython blocks in the new

Copy link
Contributor Author

Choose a reason for hiding this comment

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

@jreback from what i see from similar cases, this is already in correct form; v0.18.1.txt#L159 as an example

Copy link
Contributor

Choose a reason for hiding this comment

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

yes for the Previous Behvaior, not the NEW. pls make the change


In [7]: df.groupby('A', as_index=True)['B'].nth(0)
Out[7]:
A
a 1
b 2
Name: B, dtype: int64

In [8]: df.groupby('A', as_index=False)['B'].nth(0)
Out[8]:
0 1
1 2
Name: B, dtype: int64


.. _whatsnew_0181.apply_resample:

Using ``.apply`` on groupby resampling
Expand Down Expand Up @@ -239,7 +280,7 @@ Deprecations

Performance Improvements
~~~~~~~~~~~~~~~~~~~~~~~~

- Performance improvements in ``GroupBy.cumcount`` (:issue:`11039`)


- Improved performance of ``DataFrame.to_sql`` when checking case sensitivity for tables. Now only checks if table has been created correctly when table name is not lower case. (:issue:`12876`)
Expand Down
105 changes: 36 additions & 69 deletions pandas/core/groupby.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -653,37 +653,37 @@ def _iterate_slices(self):
def transform(self, func, *args, **kwargs):
raise AbstractMethodError(self)

def _cumcount_array(self, arr=None, ascending=True):
def _cumcount_array(self, ascending=True):
"""
arr is where cumcount gets its values from
Parameters
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 Parameters section

----------
ascending : bool, default True
If False, number in reverse, from length of group - 1 to 0.

Note
----
this is currently implementing sort=False
(though the default is sort=True) for groupby in general
"""
if arr is None:
arr = np.arange(self.grouper._max_groupsize, dtype='int64')

len_index = len(self._selected_obj.index)
cumcounts = np.zeros(len_index, dtype=arr.dtype)
if not len_index:
return cumcounts
ids, _, ngroups = self.grouper.group_info
sorter = _get_group_index_sorter(ids, ngroups)
ids, count = ids[sorter], len(ids)

indices, values = [], []
for v in self.indices.values():
indices.append(v)
if count == 0:
return np.empty(0, dtype=np.int64)

if ascending:
values.append(arr[:len(v)])
else:
values.append(arr[len(v) - 1::-1])
run = np.r_[True, ids[:-1] != ids[1:]]
rep = np.diff(np.r_[np.nonzero(run)[0], count])
out = (~run).cumsum()

indices = np.concatenate(indices)
values = np.concatenate(values)
cumcounts[indices] = values
if ascending:
out -= np.repeat(out[run], rep)
else:
out = np.repeat(out[np.r_[run[1:], True]], rep) - out

return cumcounts
rev = np.empty(count, dtype=np.intp)
rev[sorter] = np.arange(count, dtype=np.intp)
return out[rev].astype(np.int64, copy=False)

def _index_with_as_index(self, b):
"""
Expand Down Expand Up @@ -1170,47 +1170,21 @@ def nth(self, n, dropna=None):
else:
raise TypeError("n needs to be an int or a list/set/tuple of ints")

m = self.grouper._max_groupsize
# filter out values that are outside [-m, m)
pos_nth_values = [i for i in nth_values if i >= 0 and i < m]
neg_nth_values = [i for i in nth_values if i < 0 and i >= -m]

nth_values = np.array(nth_values, dtype=np.intp)
self._set_selection_from_grouper()
if not dropna: # good choice
if not pos_nth_values and not neg_nth_values:
# no valid nth values
return self._selected_obj.loc[[]]

rng = np.zeros(m, dtype=bool)
for i in pos_nth_values:
rng[i] = True
is_nth = self._cumcount_array(rng)

if neg_nth_values:
rng = np.zeros(m, dtype=bool)
for i in neg_nth_values:
rng[- i - 1] = True
is_nth |= self._cumcount_array(rng, ascending=False)
if not dropna:
mask = np.in1d(self._cumcount_array(), nth_values) | \
np.in1d(self._cumcount_array(ascending=False) + 1, -nth_values)

result = self._selected_obj[is_nth]
out = self._selected_obj[mask]
if not self.as_index:
return out

# the result index
if self.as_index:
ax = self.obj._info_axis
names = self.grouper.names
if self.obj.ndim == 1:
# this is a pass-thru
pass
elif all([x in ax for x in names]):
indicies = [self.obj[name][is_nth] for name in names]
result.index = MultiIndex.from_arrays(
indicies).set_names(names)
elif self._group_selection is not None:
result.index = self.obj._get_axis(self.axis)[is_nth]

result = result.sort_index()
ids, _, _ = self.grouper.group_info
out.index = self.grouper.result_index[ids[mask]]

return result
return out.sort_index() if self.sort else out

if isinstance(self._selected_obj, DataFrame) and \
dropna not in ['any', 'all']:
Expand Down Expand Up @@ -1241,8 +1215,8 @@ def nth(self, n, dropna=None):
axis=self.axis, level=self.level,
sort=self.sort)

sizes = dropped.groupby(grouper).size()
result = dropped.groupby(grouper).nth(n)
grb = dropped.groupby(grouper, as_index=self.as_index, sort=self.sort)
sizes, result = grb.size(), grb.nth(n)
mask = (sizes < max_len).values

# set the results which don't meet the criteria
Expand Down Expand Up @@ -1380,11 +1354,8 @@ def head(self, n=5):
0 1 2
2 5 6
"""

obj = self._selected_obj
in_head = self._cumcount_array() < n
head = obj[in_head]
return head
mask = self._cumcount_array() < n
return self._selected_obj[mask]

@Substitution(name='groupby')
@Appender(_doc_template)
Expand All @@ -1409,12 +1380,8 @@ def tail(self, n=5):
0 a 1
2 b 1
"""

obj = self._selected_obj
rng = np.arange(0, -self.grouper._max_groupsize, -1, dtype='int64')
in_tail = self._cumcount_array(rng, ascending=False) > -n
tail = obj[in_tail]
return tail
mask = self._cumcount_array(ascending=False) < n
return self._selected_obj[mask]


@Appender(GroupBy.__doc__)
Expand Down
24 changes: 11 additions & 13 deletions pandas/tests/test_groupby.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -167,8 +167,7 @@ def test_first_last_nth(self):
self.df.loc[self.df['A'] == 'foo', 'B'] = np.nan
self.assertTrue(com.isnull(grouped['B'].first()['foo']))
self.assertTrue(com.isnull(grouped['B'].last()['foo']))
self.assertTrue(com.isnull(grouped['B'].nth(0)[0])
) # not sure what this is testing
self.assertTrue(com.isnull(grouped['B'].nth(0)['foo']))

# v0.14.0 whatsnew
df = DataFrame([[1, np.nan], [1, 4], [5, 6]], columns=['A', 'B'])
Expand Down Expand Up @@ -221,12 +220,12 @@ def test_nth(self):

assert_frame_equal(g.nth(0), df.iloc[[0, 2]].set_index('A'))
assert_frame_equal(g.nth(1), df.iloc[[1]].set_index('A'))
assert_frame_equal(g.nth(2), df.loc[[], ['B']])
assert_frame_equal(g.nth(2), df.loc[[]].set_index('A'))
assert_frame_equal(g.nth(-1), df.iloc[[1, 2]].set_index('A'))
assert_frame_equal(g.nth(-2), df.iloc[[0]].set_index('A'))
assert_frame_equal(g.nth(-3), df.loc[[], ['B']])
assert_series_equal(g.B.nth(0), df.B.iloc[[0, 2]])
assert_series_equal(g.B.nth(1), df.B.iloc[[1]])
assert_frame_equal(g.nth(-3), df.loc[[]].set_index('A'))
assert_series_equal(g.B.nth(0), df.set_index('A').B.iloc[[0, 2]])
assert_series_equal(g.B.nth(1), df.set_index('A').B.iloc[[1]])
assert_frame_equal(g[['B']].nth(0),
df.ix[[0, 2], ['A', 'B']].set_index('A'))

Expand Down Expand Up @@ -262,11 +261,11 @@ def test_nth(self):
4: 0.70422799999999997}}).set_index(['color',
'food'])

result = df.groupby(level=0).nth(2)
result = df.groupby(level=0, as_index=False).nth(2)
expected = df.iloc[[-1]]
assert_frame_equal(result, expected)

result = df.groupby(level=0).nth(3)
result = df.groupby(level=0, as_index=False).nth(3)
expected = df.loc[[]]
assert_frame_equal(result, expected)

Expand All @@ -290,8 +289,7 @@ def test_nth(self):
# as it keeps the order in the series (and not the group order)
# related GH 7287
expected = s.groupby(g, sort=False).first()
expected.index = pd.Index(range(1, 10), name=0)
result = s.groupby(g).nth(0, dropna='all')
result = s.groupby(g, sort=False).nth(0, dropna='all')
assert_series_equal(result, expected)

# doc example
Expand All @@ -316,14 +314,14 @@ def test_nth(self):
assert_frame_equal(
g.nth([0, 1, -1]), df.iloc[[0, 1, 2, 3, 4]].set_index('A'))
assert_frame_equal(g.nth([2]), df.iloc[[2]].set_index('A'))
assert_frame_equal(g.nth([3, 4]), df.loc[[], ['B']])
assert_frame_equal(g.nth([3, 4]), df.loc[[]].set_index('A'))

business_dates = pd.date_range(start='4/1/2014', end='6/30/2014',
freq='B')
df = DataFrame(1, index=business_dates, columns=['a', 'b'])
# get the first, fourth and last two business days for each month
result = df.groupby((df.index.year, df.index.month)).nth([0, 3, -2, -1
])
key = (df.index.year, df.index.month)
result = df.groupby(key, as_index=False).nth([0, 3, -2, -1])
expected_dates = pd.to_datetime(
['2014/4/1', '2014/4/4', '2014/4/29', '2014/4/30', '2014/5/1',
'2014/5/6', '2014/5/29', '2014/5/30', '2014/6/2', '2014/6/5',
Expand Down