Make kwargs names in scale.py not include the axis direction. #14916

anntzer · 2019-07-29T21:29:19Z

Make names of kwargs to Scale subclasses independent of whether the
underlying axis is x or y (so that after the deprecation period, one
can just have a normal signature --
def __init__(self, *, base=10, ...).

We could possibly also change semilogx and semilogy to use the
unsuffixed names (with deprecation), leaving only loglog with the
suffixed names, or even also make loglog use unsuffixed names, in
which case it would become impossible to set separate x and y bases
directly in loglog -- one should then do gca().set_xscale("log", base=...) which doesn't seem too onerous. (loglog is the only case
where the suffixed names has some utility.)
EDIT: see Make kwargs names in scale.py not include the axis direction. #14916 (comment).

In general, note that having kwargs that depend on the axis direction
doesn't really scale -- for example, if we were to finally add
log-scale support to mplot3d, should scale.py know about it and add
basez? Should we have baser for log-scale polar plots? (at least
in the radial direction, they make sense)
Move argument validation up the the Transform subclasses.
Make SymmetricalLogScale.{base,linthresh,linscale} properties
mapping to the underlying transform so that they can't go out of sync.

As discussed with @tacaswell during the dev call.

PR Summary

PR Checklist

Has Pytest style unit tests
Code is Flake 8 compliant
New features are documented, with examples if plot related
Documentation is sphinx and numpydoc compliant
Added an entry to doc/users/next_whats_new/ if major new feature (follow instructions in README.rst there)
Documented in doc/api/api_changes.rst if API changed in a backward-incompatible way

timhoffm

I've always found the abbreviations quite confusing. Since we're renaming the parameters anyway, we could move to more telling names:

subs -> subticks
nonpos -> nonpositive
linthresh -> linthreshold

But to be fair, that'd be a bit more work; at least for subs and linthresh. They are used as public attributes in some places, which should then be deprecated/changed as well. nonpos is only a private attribute.

timhoffm · 2019-08-10T19:03:03Z

lib/matplotlib/scale.py

+        @rename_parameter("3.2", f"linthresh{axis.axis_name}", "linthresh")
+        @rename_parameter("3.2", f"subs{axis.axis_name}", "subs")
+        @rename_parameter("3.2", f"linscale{axis.axis_name}", "linscale")
+        def __init__(*, base=10, linthresh=2, subs=None, linscale=1, **kwargs):


any particular reason to call this __init__? I would just use init because it's just a regular local method. The dunder may be a bit misleading.

Yes, this is so that when the wrong arg names are passed, Python generates the error message with the right name (__init__() got an unexpected argument...).

Maybe a comment in the code for why you did this? I guess its just a place holder until a future version?

anntzer · 2019-08-10T21:43:12Z

I agree that if we are changing the names it may be better to directly pick better names rather than have two changes in a row... but subs and linthresh are also used as parameter/attribute names in ticker.py, which would also need to be changed for consistency, but that seems to not cross the threshold of "is it really worth the disruption".

timhoffm · 2019-08-11T09:20:22Z

We should at least change nonpos -> nonpositive. That one is particularly confusing because "pos" is often used as an abbreviation of "position". And there is no parameter/attribute involved that would need changing as well.

For reference:

Usages of subs:

As parameter in __init__ and set_params of LogLocator and SymmetricalLogLocator
As method in LogLocator (should rather be set_subs() anyway)
As attribute in LogScale (and its derivatives)

Usage of linthresh:

As parameter in LogFormatter, SymmetricalLogFormatter, SymmetricalLogTransform, InvertredSymmetricalLogTransform, SymLogNorm
As attribute in SymmetricalLogTransform, InvertredSymmetricalLogTransform, SymmetricalLogScale

I agree that linthresh is not worth the disruption. I support changing subs but won't insist on it.

anntzer · 2019-08-11T09:34:01Z

nonpos also maps to minpos, which appears in various places.

timhoffm · 2019-08-11T10:24:34Z

nonpos also maps to minpos, which appears in various places.

😭

In the trade-off I wouldn't let that prevent the nonpos change. They are only loosely related, and IMHO a clearer name is more important than the similarity here.

anntzer · 2019-11-29T19:56:23Z

I'm fine with renaming nonpos to nonpositive if there's a consensus in that direction (i.e. if another dev agrees).
Another advantage of making scales more independent of the axis is that this could later help in constructing some norms (which are independent of axis) as wrappers around scales -- thus allowing norms to take advantage of e.g. the scale -> locator/formatter mapping that already exists, instead of having to hard-code it into colorbar.py.

jklymak

This looks great to me, just a few minor questions. I think this is worth the API change - I don't see any reason a scale should care if its an x or y axis....

jklymak · 2020-01-31T17:30:17Z

lib/matplotlib/axes/_axes.py

@@ -1791,16 +1791,13 @@ def loglog(self, *args, **kwargs):
        **kwargs
            All parameters supported by `.plot`.
        """
-        dx = {k: kwargs.pop(k) for k in ['basex', 'subsx', 'nonposx']
+        dx = {k[:-1]: kwargs.pop(k) for k in ['basex', 'subsx', 'nonposx']


Sorry, why are you excluding the last element here?

Because the kwarg name for semilogx/y is still "basex"; this PR doesn't change that. Specifically, while I could change the kwarg name in semilogx to "base" (with the same deprecation), this doesn't work for loglog(), which needs a separate basex/basey (although it would also be reasonable to deprecate supporting directly setting two different bases for x and y in loglog() and require the user to call set_x/yscale("log", base=...) for that specific case. I think we just need to decide on the best API change here.

OK. Duh, I see now, but this might be a rare case where not using comprehension might be a lot clearer:

for k in ['basex', 'subex', 'nonposx']: dx{k[:-1]} = kwargs.pop(k)

I agree that the side-effect of popping kwargs in the comprehension may not be nicest, and can change that, but before doing so, do you have comments on the API changes suggested above?

I agree that most folks won't use different basex, basey. OTOH there is call to use different subs and nonpos isn't there?

People can still use set_xscale("log", subs=...) in that case. Note that in the (common?) case of using the same subs on x and y, the proposed new API is actually nicer (loglog(..., subs=...) instead of repeating yourself with subs(..., subsx=..., subsy=...)).
I have kind of changed my mind compared to when I first wrote the PR; I now think that having the same kwargs in loglog/semilogx/y as in set_scale("log", ...) would actually be quite nice (whereas I was previously more ok with keeping basex/basey on loglog/semilog).

I'd be onboard with this, but we should prob be prepared to not do this if the deprecation warning freaks too many people out.

Other comments? @timhoffm @ImportanceOfBeingErnest @efiring?

basex, basey are poor names anyway; I would be happy to see them go. This sort of regularization of the API will inevitably cause some pain, but I think we should do it.

jklymak · 2020-01-31T17:31:20Z

lib/matplotlib/scale.py

@@ -280,8 +280,11 @@ class LogTransform(Transform):

    def __init__(self, base, nonpos='clip'):
        Transform.__init__(self)
+        if base <= 0 or base == 1:
+            raise ValueError('The log base cannot be <= 0 or == 1')


What happened before? Did we just get a ValueError later?

The error checking was in LogScale (which calls into LogTransform), but that meant that you could construct invalid LogTransforms which only errored out at use time. With the change you can't construct invalid LogTransforms or LogScales anymore.

jklymak · 2020-01-31T17:32:12Z

lib/matplotlib/scale.py

+        if linthresh <= 0.0:
+            raise ValueError("'linthresh' must be positive")
+        if linscale <= 0.0:
+            raise ValueError("'linscale' must be positive")


Same question here

jklymak · 2020-01-31T17:33:34Z

lib/matplotlib/scale.py

+        @rename_parameter("3.2", f"linthresh{axis.axis_name}", "linthresh")
+        @rename_parameter("3.2", f"subs{axis.axis_name}", "subs")
+        @rename_parameter("3.2", f"linscale{axis.axis_name}", "linscale")
+        def __init__(*, base=10, linthresh=2, subs=None, linscale=1, **kwargs):


Maybe a comment in the code for why you did this? I guess its just a place holder until a future version?

anntzer · 2020-02-03T14:42:19Z

re your last point: added a comment to clarify.

tacaswell · 2020-02-03T20:53:21Z

👍 in principle, but have not reviewed in detail.

anntzer · 2020-02-04T07:28:27Z

OK, added a second commit which also changes the kwarg names on the Axes methods.
#14916 (review) (rename "nonpos" to "nonpositive"?) still needs to be decided upon.

codecov · 2020-02-04T08:01:23Z

Codecov Report

Merging #14916 into master will decrease coverage by 0.38%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##           master   #14916      +/-   ##
==========================================
- Coverage   80.85%   80.46%   -0.39%     
==========================================
  Files         307      307              
  Lines       75746    74412    -1334     
  Branches     9691     9695       +4     
==========================================
- Hits        61244    59876    -1368     
- Misses      11962    11988      +26     
- Partials     2540     2548       +8

Impacted Files	Coverage Δ
lib/matplotlib/backends/backend_macosx.py	`4.54% <0.00%> (-36.12%)`	⬇️
lib/matplotlib/testing/jpl_units/StrConverter.py	`15.55% <0.00%> (-5.28%)`	⬇️
lib/matplotlib/testing/jpl_units/EpochConverter.py	`69.69% <0.00%> (-4.67%)`	⬇️
lib/matplotlib/docstring.py	`65.00% <0.00%> (-3.19%)`	⬇️
lib/matplotlib/backends/qt_compat.py	`48.42% <0.00%> (-3.16%)`	⬇️
...b/matplotlib/testing/jpl_units/UnitDblConverter.py	`65.85% <0.00%> (-2.33%)`	⬇️
lib/matplotlib/backend_managers.py	`20.95% <0.00%> (-2.30%)`	⬇️
lib/matplotlib/backends/backend_qt5.py	`55.94% <0.00%> (-1.44%)`	⬇️
lib/matplotlib/font_manager.py	`74.10% <0.00%> (-1.17%)`	⬇️
lib/matplotlib/pyplot.py	`71.72% <0.00%> (-1.14%)`	⬇️
... and 152 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 7b6eb77...dee07b2. Read the comment docs.

- Make names of kwargs to Scale subclasses independent of whether the underlying axis is x or y (so that after the deprecation period, one can just have a normal signature -- `def __init__(self, *, base=10, ...)`. We could possibly also change semilogx and semilogy to use the unsuffixed names (with deprecation), leaving only loglog with the suffixed names, or even also make loglog use unsuffixed names, in which case it would become impossible to set separate x and y bases directly in loglog -- one should then do `gca().set_xscale("log", base=...)` which doesn't seem too onerous. (loglog is the only case where the suffixed names has *some* utility.) In general, note that having kwargs that depend on the axis direction doesn't really scale -- for example, if we were to finally add log-scale support to mplot3d, should scale.py know about it and add `basez`? Should we have `baser` for log-scale polar plots? (at least in the radial direction, they make sense) - Move argument validation up the the Transform subclasses. - Make SymmetricalLogScale.{base,linthresh,linscale} properties mapping to the underlying transform so that they can't go out of sync.

anntzer · 2020-02-08T23:51:34Z

rebased

anntzer · 2020-02-08T23:52:18Z

before this gets merged please comment on @timhoffm's suggestion to switch nonpos to nonpositive as well.

efiring · 2020-02-09T01:51:50Z

It's a modest difference, but I agree that "nonpositive='clip'" reads better than "nonpos='clip'". I'm guessing the kwarg (the plain version, not "nonposx" etc.) is not widely used, so the additional irritation will be moderate.

efiring · 2020-02-09T02:00:54Z

@anntzer While you are in the neighborhood, see also #16451.

anntzer · 2020-02-09T10:32:20Z

@efiring see #16452.
I'll make the change to nonpositive here.

jklymak · 2020-02-09T22:32:42Z

I'm still happy with this. @efiring if you are OK can you merge?

anntzer force-pushed the morescalevalidation branch from 1e946e0 to acd0e2f Compare July 29, 2019 22:10

timhoffm reviewed Aug 10, 2019

View reviewed changes

timhoffm mentioned this pull request Aug 11, 2019

setting x_scale symlog distorts graph #14597

Closed

tacaswell added this to the v3.3.0 milestone Aug 12, 2019

anntzer force-pushed the morescalevalidation branch from acd0e2f to 6551cb1 Compare November 29, 2019 19:54

anntzer force-pushed the morescalevalidation branch from 6551cb1 to 965f599 Compare November 30, 2019 22:01

anntzer force-pushed the morescalevalidation branch 2 times, most recently from 92041d0 to 3e4af57 Compare December 9, 2019 06:39

anntzer mentioned this pull request Jan 30, 2020

Norms need scales for good colorbars... #16375

Open

jklymak reviewed Jan 31, 2020

View reviewed changes

anntzer force-pushed the morescalevalidation branch from 3e4af57 to bbb5d4f Compare February 3, 2020 14:41

anntzer force-pushed the morescalevalidation branch from bbb5d4f to eb37584 Compare February 3, 2020 16:05

anntzer force-pushed the morescalevalidation branch from eb37584 to dee07b2 Compare February 4, 2020 07:35

anntzer mentioned this pull request Feb 6, 2020

Re-write sym-log-norm #16391

Closed

jklymak approved these changes Feb 6, 2020

View reviewed changes

timhoffm approved these changes Feb 8, 2020

View reviewed changes

timhoffm added the status: needs rebase label Feb 8, 2020

anntzer force-pushed the morescalevalidation branch from dee07b2 to 82ee63e Compare February 8, 2020 23:51

anntzer removed the status: needs rebase label Feb 8, 2020

anntzer added the status: work in progress label Feb 9, 2020

anntzer added 2 commits February 9, 2020 11:44

Also change the kwarg names on semilogx/y and loglog.

a00c27c

Also rename nonpos->nonpositive.

ea57339

anntzer force-pushed the morescalevalidation branch from 82ee63e to ea57339 Compare February 9, 2020 10:48

anntzer removed the status: work in progress label Feb 9, 2020

timhoffm approved these changes Feb 9, 2020

View reviewed changes

efiring merged commit 93de426 into matplotlib:master Feb 9, 2020

anntzer deleted the morescalevalidation branch February 9, 2020 22:53

lostanlen mentioned this pull request Jan 28, 2021

MatplotlibDeprecationWarning with Matplotlib 3.3 librosa/librosa#1266

Closed

Uh oh!

Make kwargs names in scale.py not include the axis direction. #14916

Make kwargs names in scale.py not include the axis direction. #14916

Uh oh!

Conversation

anntzer commented Jul 29, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR Summary

PR Checklist

Uh oh!

timhoffm left a comment

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!

anntzer commented Aug 10, 2019

Uh oh!

timhoffm commented Aug 11, 2019

Uh oh!

anntzer commented Aug 11, 2019

Uh oh!

timhoffm commented Aug 11, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

anntzer commented Nov 29, 2019

Uh oh!

jklymak left a comment

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!

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!

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!

anntzer commented Feb 3, 2020

Uh oh!

tacaswell commented Feb 3, 2020

Uh oh!

anntzer commented Feb 4, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

codecov bot commented Feb 4, 2020

Codecov Report

Uh oh!

anntzer commented Feb 8, 2020

Uh oh!

anntzer commented Feb 8, 2020

Uh oh!

efiring commented Feb 9, 2020

Uh oh!

efiring commented Feb 9, 2020

Uh oh!

anntzer commented Feb 9, 2020

Uh oh!

jklymak commented Feb 9, 2020

anntzer commented Jul 29, 2019 •

edited

Loading

timhoffm commented Aug 11, 2019 •

edited

Loading

anntzer commented Feb 4, 2020 •

edited

Loading