DEPR: Deprecate the convert parameter completely #17831

gfyoung · 2017-10-10T05:11:05Z

Previously, we weren't issuing a warning if the user happened to pass in the original default of True,
which would cause downstream code to break.

Closes #17828

jorisvandenbossche

Thanks! Left a few comments, in general IMO it is not needed to explain this for the user, for them the default is still True

jorisvandenbossche · 2017-10-10T06:34:44Z

pandas/core/generic.py

@@ -2135,13 +2135,18 @@ def _take(self, indices, axis=0, convert=True, is_copy=True):
        axis : int, default 0
            The axis on which to select elements. "0" means that we are
            selecting rows, "1" means that we are selecting columns, etc.
-        convert : bool, default True


You can leave this as True

Ok, will fix.

jorisvandenbossche · 2017-10-10T06:39:14Z

pandas/core/generic.py


            Whether to convert negative indices into positive ones.
            For example, ``-1`` would map to the ``len(axis) - 1``.
            The conversions are similar to the behavior of indexing a
            regular Python list.
+
+            If ``None`` is passed in, negative indices will be converted
+            to positive indices.


This is not needed IMO

Ok, will fix.

jorisvandenbossche · 2017-10-10T06:40:42Z

pandas/tests/frame/test_axis_select_reindex.py

@@ -856,6 +856,10 @@ def test_take(self):
                result = df.take(order, convert=False, axis=0)
                assert_frame_equal(result, expected)

+            with tm.assert_produces_warning(FutureWarning):
+                result = df.take(order, convert=False, axis=0)


Should this be True?

Yep, it should. Oops 😄

jorisvandenbossche · 2017-10-10T06:42:20Z

pandas/tests/sparse/test_series.py

            sp.take([1, 5], convert=False)

    def test_numpy_take(self):
        sp = SparseSeries([1.0, 2.0, 3.0])
        indices = [1, 2]

-        # gh-17352: older versions of numpy don't properly
-        # pass in arguments to downstream .take() implementations.
-        warning = FutureWarning if _np_version_under1p12 else None


Is there a reason this is not the case anymore?

Actually, I need to put that back. However, I different default will be needed, as convert=None breaks downstream compatibility with numpy:

https://travis-ci.org/pandas-dev/pandas/jobs/285909832

What is the problem with None for numpy? You linked to a sparse failure, but in the sparse code you left the default as None.

If it is needed, another solution might be to push it into the kwargs? (so the -1 is not visible)

BTW, can you push new commits instead of squashing into the previous one? That makes it easier to say what you changed.

Oh, sorry, missed your message before I pushed.

The problem is that numpy's take implementation for slightly older versions passes in None for one of its parameters by position. This parameter corresponds to the convert parameter for our version of take. It's messy compatibility stuff that I had to wade through at some point that I patched on both sides.

gfyoung · 2017-10-10T16:07:25Z

The new kwarg gets past the original failure but need a more robust comparator.

jreback · 2017-10-10T17:41:56Z

pandas/core/generic.py

-        nv.validate_take(tuple(), kwargs)
-
-        if not convert:
+    def take(self, indices, axis=0, convert=-1, is_copy=True, **kwargs):


why can’t u just use None

@jreback : breaks compatibility with older numpy versions

not sure how, you are stripping the kwarg before passing it on

jreback · 2017-10-10T17:42:35Z

pandas/core/sparse/series.py

-        convert = nv.validate_take_with_convert(convert, args, kwargs)
-
-        if not convert:
+    def take(self, indices, axis=0, convert=-1, *args, **kwargs):


same this is very non intuitive to use a placeholder value that is not valid

Well, none is not a valid parameter for a boolean input either, and the compatibility issue still stands

sure it is, you need to set the default before you check the inputs. That's the point.

if convert is not None: warnings.warn(.....) validate(convert)

@jreback : I do set the default beforehand, and everything is green now.

codecov · 2017-10-11T01:43:36Z

Codecov Report

Merging #17831 into master will increase coverage by <.01%.
The diff coverage is 88.88%.

@@            Coverage Diff             @@
##           master   #17831      +/-   ##
==========================================
+ Coverage   91.22%   91.22%   +<.01%     
==========================================
  Files         163      163              
  Lines       50014    50016       +2     
==========================================
+ Hits        45627    45629       +2     
  Misses       4387     4387

Flag	Coverage Δ
#multiple	`89.03% <88.88%> (+0.01%)`	⬆️
#single	`40.24% <55.55%> (-0.07%)`	⬇️

Impacted Files	Coverage Δ
pandas/core/generic.py	`92.09% <100%> (ø)`	⬆️
pandas/core/sparse/series.py	`95.28% <80%> (+0.01%)`	⬆️
pandas/io/gbq.py	`25% <0%> (-58.34%)`	⬇️
pandas/core/indexing.py	`92.82% <0%> (-0.19%)`	⬇️
pandas/core/frame.py	`97.77% <0%> (-0.1%)`	⬇️
pandas/core/indexes/datetimes.py	`95.48% <0%> (-0.1%)`	⬇️
pandas/plotting/_converter.py	`65.2% <0%> (+1.81%)`	⬆️

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 727ea20...762fd6e. Read the comment docs.

codecov · 2017-10-11T01:43:43Z

Codecov Report

Merging #17831 into master will decrease coverage by 0.04%.
The diff coverage is 80%.

@@            Coverage Diff             @@
##           master   #17831      +/-   ##
==========================================
- Coverage   91.25%    91.2%   -0.05%     
==========================================
  Files         163      163              
  Lines       50038    50040       +2     
==========================================
- Hits        45661    45638      -23     
- Misses       4377     4402      +25

Flag	Coverage Δ
#multiple	`89.01% <80%> (-0.04%)`	⬇️
#single	`40.26% <20%> (-0.07%)`	⬇️

Impacted Files	Coverage Δ
pandas/core/generic.py	`92.2% <ø> (ø)`	⬆️
pandas/core/sparse/series.py	`95.28% <80%> (+0.01%)`	⬆️
pandas/io/gbq.py	`25% <0%> (-58.34%)`	⬇️
pandas/plotting/_converter.py	`63.38% <0%> (-1.82%)`	⬇️
pandas/compat/numpy/function.py	`92.12% <0%> (-1.22%)`	⬇️
pandas/core/indexing.py	`92.82% <0%> (-0.19%)`	⬇️
pandas/core/frame.py	`97.75% <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 7e159ae...8f946b7. Read the comment docs.

gfyoung · 2017-10-11T03:33:10Z

Everything is passing. PTAL.

bkmrkr · 2017-10-11T08:14:09Z

Can I please unsubscribe

On Tue, Oct 10, 2017 at 7:11 AM gfyoung ***@***.***> wrote: Previously, we weren't issuing a warning if the user happened to pass in the original default of True, which would cause downstream code to break. Closes #17828 <#17828> ------------------------------ You can view, comment on, or merge this pull request online at: #17831 Commit Summary - DEPR: Deprecate the convert parameter completely File Changes - *M* pandas/compat/numpy/function.py <https://github.com/pandas-dev/pandas/pull/17831/files#diff-0> (6) - *M* pandas/core/generic.py <https://github.com/pandas-dev/pandas/pull/17831/files#diff-1> (13) - *M* pandas/core/sparse/series.py <https://github.com/pandas-dev/pandas/pull/17831/files#diff-2> (8) - *M* pandas/tests/frame/test_axis_select_reindex.py <https://github.com/pandas-dev/pandas/pull/17831/files#diff-3> (4) - *M* pandas/tests/sparse/test_series.py <https://github.com/pandas-dev/pandas/pull/17831/files#diff-4> (14) Patch Links: - https://github.com/pandas-dev/pandas/pull/17831.patch - https://github.com/pandas-dev/pandas/pull/17831.diff — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#17831>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAJfa3Nd7QUQ4eegHECsoBwQ54cWa-D2ks5sqvxrgaJpZM4PzYie> .

-- Greg Gurevich greggurevich@gmail.com cell: 1-917-504-7105

jorisvandenbossche · 2017-10-11T09:51:57Z

Can I please unsubscribe

@bkmrkr If you scroll above, there is a button 'Unwatch' (or watch), if you click on the dropdown, you can choose which issue to follow.

TomAugspurger · 2017-10-11T15:28:14Z

Thoughts @jreback? This is blocking the RC now.

@gfyoung could you explain in a bit more detail why making convert=None the default breaks compat w/ older NumPy? I would think that something like

def take(self, indices, axis=0, convert=None, is_copy=True, **kwargs):
    if convert is None:
        convert = True
    else:
        warnings.warn(...)

would work fine. Not a huge deal, just a little strange to use -1 instead of None I think.

gfyoung · 2017-10-11T15:36:01Z

@TomAugspurger : I've tried various logic (including that one), and older versions of numpy don't react well when you try to call np.take. The reason is that take=None is getting confused with older numpy's positional argument function calls internally. Trying to accommodate numpy results in me breaking compatibility with perfectly reasonable function calls to .take() in pandas.

If we were to drop support for numpy < 1.12, this wouldn't be an issue 😄

jreback · 2017-10-11T18:20:50Z

@gfyoung I think let's keep the signature with convert=None. If we break np.take(...) of a pandas expression, too bad. just fix the test (and you can put a note). putting in a -1 is really odd here.

jreback · 2017-10-11T18:21:30Z

If we were to drop support for numpy < 1.12, this wouldn't be an issue

not goning to happen :> (actually you can leave the test for >= 1.12 for this as it will work with convert=None

gfyoung · 2017-10-11T19:22:17Z

I think let's keep the signature with convert=None. If we break np.take(...) of a pandas expression, too bad.

Okay, will do.

not goning to happen :>

Oh, that's too bad. I thought there was a good chance we would do that 😄

Previously, we weren't issuing a warning if the user happened to pass in the original default of "True", which would cause downstream code to break. Closes pandas-devgh-17828.

gfyoung · 2017-10-12T09:42:54Z

Finally, all is green again. PTAL.

jreback · 2017-10-12T09:57:10Z

pandas/core/generic.py

@@ -2172,6 +2172,7 @@ def _take(self, indices, axis=0, convert=True, is_copy=True):
            selecting rows, "1" means that we are selecting columns, etc.
        convert : bool, default True
            .. deprecated:: 0.21.0
+               In the future, negative indices will always be converted.


leftover comment

@TomAugspurger prob just as easy to fix this on merge

I can fix that in #17858

Although, why do you say that's leftover? Wasn't that added deliberately @gfyoung?

actually nvm thought was a different comment

jreback · 2017-10-12T21:03:34Z

lgtm module a small doc comment..

Previously, we weren't issuing a warning if the user happened to pass in the original default of "True", which would cause downstream code to break. Closes pandas-devgh-17828.

* upstream/master: (76 commits) CategoricalDtype construction: actually use fastpath (pandas-dev#17891) DEPR: Deprecate tupleize_cols in to_csv (pandas-dev#17877) BUG: Fix wrong column selection in drop_duplicates when duplicate column names (pandas-dev#17879) DOC: Adding examples to update docstring (pandas-dev#16812) (pandas-dev#17859) TST: Skip if no openpyxl in test_excel (pandas-dev#17883) TST: Catch read_html slow test warning (pandas-dev#17874) flake8 cleanup (pandas-dev#17873) TST: remove moar warnings (pandas-dev#17872) ENH: tolerance now takes list-like argument for reindex and get_indexer. (pandas-dev#17367) ERR: Raise ValueError when week is passed in to_datetime format witho… (pandas-dev#17819) TST: remove some deprecation warnings (pandas-dev#17870) Refactor index-as-string groupby tests and fix spurious warning (Bug 17383) (pandas-dev#17843) BUG: merging with a boolean/int categorical column (pandas-dev#17841) DEPR: Deprecate read_csv arguments fully (pandas-dev#17865) BUG: to_json - prevent various segfault conditions (GH14256) (pandas-dev#17857) CLN: Use pandas.core.common for None checks (pandas-dev#17816) BUG: set tz on DTI from fixed format HDFStore (pandas-dev#17844) RLS: v0.21.0rc1 Whatsnew cleanup (pandas-dev#17858) DEPR: Deprecate the convert parameter completely (pandas-dev#17831) ...

Previously, we weren't issuing a warning if the user happened to pass in the original default of "True", which would cause downstream code to break. Closes pandas-devgh-17828.

gfyoung added the Deprecate Functionality to remove in pandas label Oct 10, 2017

gfyoung added this to the 0.21.0 milestone Oct 10, 2017

gfyoung force-pushed the fully-deprecate-convert branch from d5337d1 to 1080c19 Compare October 10, 2017 06:17

jorisvandenbossche reviewed Oct 10, 2017

View reviewed changes

gfyoung force-pushed the fully-deprecate-convert branch 2 times, most recently from 2f2db4f to c5d0f08 Compare October 10, 2017 08:07

TomAugspurger mentioned this pull request Oct 10, 2017

RLS: 0.21.0 #17285

Closed

64 tasks

jreback reviewed Oct 10, 2017

View reviewed changes

jreback requested changes Oct 10, 2017

View reviewed changes

gfyoung force-pushed the fully-deprecate-convert branch from c5d0f08 to 762fd6e Compare October 11, 2017 00:14

gfyoung force-pushed the fully-deprecate-convert branch 3 times, most recently from 5ac5e38 to 29b6b64 Compare October 12, 2017 07:32

DEPR: Deprecate the convert parameter completely

8f946b7

Previously, we weren't issuing a warning if the user happened to pass in the original default of "True", which would cause downstream code to break. Closes pandas-devgh-17828.

gfyoung force-pushed the fully-deprecate-convert branch from 29b6b64 to 8f946b7 Compare October 12, 2017 08:36

jorisvandenbossche approved these changes Oct 12, 2017

View reviewed changes

jreback requested changes Oct 12, 2017

View reviewed changes

jreback approved these changes Oct 12, 2017

View reviewed changes

TomAugspurger merged commit 92db5c9 into pandas-dev:master Oct 12, 2017

gfyoung deleted the fully-deprecate-convert branch October 13, 2017 01:29

DEPR: Deprecate the convert parameter completely #17831

DEPR: Deprecate the convert parameter completely #17831

Conversation

gfyoung commented Oct 10, 2017

jorisvandenbossche left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

gfyoung commented Oct 10, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jreback Oct 10, 2017 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

codecov bot commented Oct 11, 2017

Codecov Report

codecov bot commented Oct 11, 2017 • edited Loading

Codecov Report

gfyoung commented Oct 11, 2017

bkmrkr commented Oct 11, 2017 via email

jorisvandenbossche commented Oct 11, 2017

TomAugspurger commented Oct 11, 2017

gfyoung commented Oct 11, 2017

jreback commented Oct 11, 2017

jreback commented Oct 11, 2017

gfyoung commented Oct 11, 2017

gfyoung commented Oct 12, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jreback commented Oct 12, 2017

jreback Oct 10, 2017 •

edited

Loading

codecov bot commented Oct 11, 2017 •

edited

Loading