PERF: custom ops for RangeIndex.[all|any|contains] #26617

topper-123 · 2019-06-02T09:14:11Z

xref PERF: don't call RangeIndex._data unnecessarily #26565
tests added / passed
passes git diff upstream/master -u -- "*.py" | flake8 --diff
whatsnew entry

Follow-up to #26565. Make RangeIndex._data be created in fewer cases.

Performance examples (but the larger gain is from memory savings by not creating the _data array):

>>>  rng = pd.RangeIndex(-5_000_000, 0, 6)
>>> %timeit rng.all()
789 µs ± 5.24 µs per loop  # master
216 ns ± 0.957 ns per loop  # this PR
>>> %timeit rng.any()
763 µs ± 6.08 µs per loop  # master
124 ns ± 9.15 ns per loop  # this PR
>>> %timeit (-2 in rng)
380 ns ± 1.85 ns per loop  # master
689 ns ± 44.7 ns per loop  # this PR, slightly slower

topper-123 · 2019-06-02T09:22:22Z

pandas/core/indexes/range.py

+    def __contains__(self, key):
+        hash(key)
+        try:
+            key = com.ensure_python_int(key)


A Python int is needed, else Python iterates over the range, making the operation very slow.

codecov · 2019-06-02T09:50:37Z

Codecov Report

Merging #26617 into master will decrease coverage by <.01%.
The diff coverage is 92%.

@@            Coverage Diff             @@
##           master   #26617      +/-   ##
==========================================
- Coverage   91.85%   91.85%   -0.01%     
==========================================
  Files         174      174              
  Lines       50722    50741      +19     
==========================================
+ Hits        46593    46606      +13     
- Misses       4129     4135       +6

Flag	Coverage Δ
#multiple	`90.39% <92%> (-0.01%)`	⬇️
#single	`41.79% <72%> (-0.08%)`	⬇️

Impacted Files	Coverage Δ
pandas/core/common.py	`98.48% <100%> (+0.06%)`	⬆️
pandas/core/indexes/base.py	`96.71% <100%> (-0.01%)`	⬇️
pandas/core/indexes/range.py	`97.6% <87.5%> (-0.46%)`	⬇️
pandas/io/gbq.py	`78.94% <0%> (-10.53%)`	⬇️
pandas/core/frame.py	`97% <0%> (-0.12%)`	⬇️

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 437efa6...cfd68e5. Read the comment docs.

codecov · 2019-06-02T09:50:38Z

Codecov Report

Merging #26617 into master will decrease coverage by 1.46%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master   #26617      +/-   ##
==========================================
- Coverage   91.87%   90.41%   -1.47%     
==========================================
  Files         174      174              
  Lines       50661    50668       +7     
==========================================
- Hits        46547    45809     -738     
- Misses       4114     4859     +745

Flag	Coverage Δ
#multiple	`90.41% <100%> (ø)`	⬆️
#single	`?`

Impacted Files	Coverage Δ
pandas/core/indexes/base.py	`96.71% <100%> (-0.01%)`	⬇️
pandas/core/indexes/range.py	`98.52% <100%> (+0.04%)`	⬆️
pandas/io/feather_format.py	`21.05% <0%> (-68.43%)`	⬇️
pandas/core/computation/pytables.py	`62.5% <0%> (-27.75%)`	⬇️
pandas/io/pytables.py	`63.82% <0%> (-26.48%)`	⬇️
pandas/io/gbq.py	`78.94% <0%> (-10.53%)`	⬇️
pandas/core/computation/common.py	`84.21% <0%> (-5.27%)`	⬇️
pandas/core/computation/expr.py	`94.5% <0%> (-3.03%)`	⬇️
pandas/io/clipboard/clipboards.py	`31.88% <0%> (-2.9%)`	⬇️
pandas/io/formats/printing.py	`84.49% <0%> (-1.07%)`	⬇️
... and 4 more

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 891a419...4aea46d. Read the comment docs.

jreback · 2019-06-02T11:18:39Z

pandas/core/common.py

    return f
+
+
+def ensure_python_int(value: Union[int, Any]) -> int:


put in dtypes.cast but i don’t think u need this as we already have ensure_platform_int

ensure_platform_int returns a numpy int(32|64), so won't work here:

>>> rng = range(1, 1_000_000) >>> %timeit 999_999 in rng 133 ns ± 2.46 ns per loop >>> %timeit np.int64(999_999) in rng 306 ms ± 66.3 ms per loop

ok in any event move it to collocate with the ensure _ routines

pandas/core/indexes/range.py

jreback · 2019-06-02T23:40:53Z

pandas/core/dtypes/cast.py

        raise ValueError("Trying to coerce float values to integers")
+
+
+def ensure_python_int(value: Union[int, Any]) -> int:


I gave the wrong direction, can you put in core/dtypes/common (right below the other ensure_*)

and can you add a doc-string (I know its trivial)

the typing seems odd here, shouldn't this be Union[int, np.integer] ?

I'm not quite sure. In principle a float or a np.floating and possibly others could for some subvalues pass the test (e.g. if value is 1.0). But ok, passing a float would be a code smell, so just as well make the programmer see passing those as typing errors.

jreback · 2019-06-02T23:41:19Z

pandas/core/indexes/range.py

+    def __contains__(self, key):
+        hash(key)
+        try:
+            key = cast.ensure_python_int(key)


import directly

jreback · 2019-06-02T23:41:41Z

pandas/core/indexes/range.py

+    def all(self):
+        return 0 not in self._range
+
+    def any(self):


can you type

I will need to wait for this until #26581 is merged, as mypy doesn't understand/see attributes defined in __new__ or _simple_new . If that one is merged, I can go through this again wrt. typing.

jreback

small comments

topper-123 · 2019-06-04T07:37:26Z

Comments addressed. The typing issue must wait until #26581 is merged for mypy to see _range.

jreback · 2019-06-05T22:05:39Z

can you add the issue number onto the one you already have in Performance section; merge master as well. ping on green.

topper-123 added Index Related to the Index class or subclasses Performance Memory or execution speed performance labels Jun 2, 2019

topper-123 added this to the 0.25.0 milestone Jun 2, 2019

topper-123 commented Jun 2, 2019

View reviewed changes

topper-123 force-pushed the RangeIndex_all_any_contains branch from 8bca43c to cfd68e5 Compare June 2, 2019 09:27

jreback reviewed Jun 2, 2019

View reviewed changes

topper-123 force-pushed the RangeIndex_all_any_contains branch 3 times, most recently from 4826efc to 3ca546e Compare June 2, 2019 15:14

jschendel reviewed Jun 2, 2019

View reviewed changes

pandas/core/indexes/range.py Outdated Show resolved Hide resolved

jreback reviewed Jun 2, 2019

View reviewed changes

jreback requested changes Jun 2, 2019

View reviewed changes

topper-123 force-pushed the RangeIndex_all_any_contains branch 3 times, most recently from 34706ce to fa4c649 Compare June 3, 2019 05:34

topper-123 mentioned this pull request Jun 4, 2019

use range in RangeIndex instead of _start etc. #26581

Merged

4 tasks

topper-123 force-pushed the RangeIndex_all_any_contains branch 2 times, most recently from 3d52e46 to 949d452 Compare June 5, 2019 14:39

topper-123 added 4 commits June 6, 2019 01:30

PERF: custom ops for RangeIndex.[all|any|__contain__]

8628123

changes

d9b1238

Changes to .all()

9ef40b9

Rabased and added typing

1e92983

topper-123 force-pushed the RangeIndex_all_any_contains branch from 949d452 to 71ab9ed Compare June 5, 2019 23:34

add issue number to whatsnew

4aea46d

topper-123 force-pushed the RangeIndex_all_any_contains branch from 71ab9ed to 4aea46d Compare June 5, 2019 23:38

topper-123 merged commit 649ad5c into pandas-dev:master Jun 6, 2019

topper-123 deleted the RangeIndex_all_any_contains branch June 19, 2019 13:43

		return f


		def ensure_python_int(value: Union[int, Any]) -> int:

		raise ValueError("Trying to coerce float values to integers")


		def ensure_python_int(value: Union[int, Any]) -> int:

Uh oh!

PERF: custom ops for RangeIndex.[all|any|__contains__] #26617

PERF: custom ops for RangeIndex.[all|any|__contains__] #26617

Uh oh!

Conversation

topper-123 commented Jun 2, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

codecov bot commented Jun 2, 2019

Codecov Report

Uh oh!

codecov bot commented Jun 2, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

Uh oh!

Choose a reason for hiding this comment

Uh oh!

topper-123 Jun 2, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

jreback left a comment

Choose a reason for hiding this comment

Uh oh!

topper-123 commented Jun 4, 2019

Uh oh!

jreback commented Jun 5, 2019

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

PERF: custom ops for RangeIndex.[all|any|contains] #26617

PERF: custom ops for RangeIndex.[all|any|contains] #26617

topper-123 commented Jun 2, 2019 •

edited

Loading

codecov bot commented Jun 2, 2019 •

edited

Loading

topper-123 Jun 2, 2019 •

edited

Loading