Enh optional tick deconflict #13826

tacaswell · 2019-04-01T04:09:17Z

The name is up for debate as is threading it through all of the __init__ methods.

Docs and tests will follow if we agree on this approach and settle on some names.

This now works as expected:

import numpy as np
import matplotlib.dates as mdates
import matplotlib.pyplot as plt

t = np.arange("2018-11-03", "2018-11-06", dtype="datetime64")
x = np.random.rand(len(t))

fig, ax = plt.subplots()
ax.plot(t,x)

ax.xaxis.set_major_locator(mdates.DayLocator())
ax.xaxis.set_major_formatter(mdates.DateFormatter('\n%a'))

ax.xaxis.set_minor_locator(mdates.HourLocator((0,6,12,18),
                                              enable_deconflict=False))
ax.xaxis.set_minor_formatter(mdates.DateFormatter('%H:%M'))
plt.show()

In matplotlib#13363 when `iter_ticks` was deprecated the in-lined logic did not account for the updates from matplotlib#13314.

anntzer

Basically good, although the name is pretty bad (yes, that's the name we were using in the discussion last week). Perhaps "allow_major_minor_overlap" or "allow_minor_overlap"?

timhoffm · 2019-04-01T06:26:47Z

I feel like the logic should be the other way round: "prevent_xxx_overlap=False" because this is the actual action. Alternative names: "remove_conflicting_ticks", "remove_duplicates", "clip_overlapping_ticks".

If you are worried on adding this to __init__, I think it's fine to just use the property/setter.

locator = mdates.HourLocator((0,6,12,18))
locator.clip_overlapping_ticks = False
ax.xaxis.set_minor_locator(locator)

It will be a rarely used property, so slightly longer code is ok.

ImportanceOfBeingErnest · 2019-04-01T14:18:53Z

This is a pretty clean and future-proof solution. I think to adress my concerns in #13618 it would equally be enough to allow people to manually set a (non-predefined) attribute on the locator (I don't know if that is bad style? But it would allow to not have to change all the tickers?)

Naming

(enable_deconflict = True) :=: (disable_conflict = True) :=: (conflict = False)
This is pretty hard to decipher (at least for non-native speakers?).

Maybe just allow_overlap? The other suggestions above are also fine. Though I'm not sure if "clip" is the right word?

pharshalp · 2019-04-01T19:01:04Z

I like the option allow_major_minor_overlap=False.

tacaswell · 2019-04-01T19:17:48Z

I think to adress my concerns in #13618 it would equally be enough to allow people to manually set a (non-predefined) attribute on the locator (I don't know if that is bad style? But it would allow to not have to change all the tickers?)

True, but threading it through and making it a property makes it easier to document and makes it more discover-able. I think we should at least fallback to putting a class-level placeholder in the Locator baseclass.

although the name is pretty bad

Indeed. I think it should be a 'default to True', as discussed on the call remove_overlapping is the current favorite.

This adds the logic to check if there is an attribute on the minor locator to control the deconfliction.

jklymak · 2019-04-02T14:09:16Z

remove_major_ticks=False? I think the kwarg should have something about “major” in them...

tacaswell · 2019-04-02T15:05:31Z

I think having 'major' in the name confuses things as it is a property of the minor locator and it affects what happens to the minor ticks. remove_major_ticks makes it sound like the minor locator is going to have priority over the major locator.

On the call we decided not put 'tick' in the name because this controls if we put in a Tick object, which provides the 'tick', the 'label', and the 'grid'.

jklymak · 2019-04-02T15:09:12Z

exclude_major_locations? I think it should have major in the name, or the user won't have any idea what's going on...

ImportanceOfBeingErnest · 2019-04-02T15:23:07Z

exclude_minor_locs_at_major_positions would be a very accurate description - though it's of course too long.

lib/matplotlib/axis.py

jklymak · 2019-04-02T15:50:55Z

lib/matplotlib/axis.py

@@ -1317,6 +1316,9 @@ def get_minorticklocs(self):
        # Remove minor ticks duplicating major ticks.
        major_locs = self.major.locator()
        minor_locs = self.minor.locator()
+        # we do check when we set the attribute that this is a Locator
+        # subclass, use getattr out of an over abundance of caution
+        remove_overlaps = getattr(self.minor.locator, 'remove_overlaps', True)


OK, so you went w/ the attribute path? So, how does this get set? I guess this PR needs tests and examples...

jklymak · 2019-04-02T15:59:20Z

lib/matplotlib/axis.py

@@ -1326,7 +1328,8 @@ def get_minorticklocs(self):
        tol = (hi - lo) * 1e-5


While we are here, I think the tolerance should be smaller here. .... 1.0001 == 1.000001 with this.... I would tend to go w/ something like 1e-10.

This is in pixel space though. If the goal is to prevent tick/grid/label collisions shouldn't we make this closer to 1?

I don't think this is a visual issue, but rather a math one. If the user wants a major tick at 1.0 and a minor tick at 1.00001 I don't think we should stop them (or make them use a kwarg) short of numerical roundoff.

All of the complaints about this have been due to visual overlap.

The problem we have is that we can not tell "the user intentionally put a major and minor ticks on top of each other so we should do what we are told" (ex, @ImportanceOfBeingErnest date example) vs "the user accidentally set up locators that collide and we should do the obviously correct thing and de-conflict them"

Previously the behavior was the first (which when we made the log minor locator smarter caused problems and @ImportanceOfBeingErnest was relying on) but now we do the second. If we accept that we want to de-conflict the ticks as a primarily a display issue then we should do the filtering based on how close it comes in pixel space.

Using this flag the user can pick between the two behavior (by telling us which of the two cases they want) and we chose to default to door number 2 ("do 'the right thing' visually) as I think that is going to be the more common intention.

Thats fine, I understand your point. However:

then we should do the filtering based on how close it comes in pixel space.

I don't think we should do any deciding of what to send to the renderer in pixel space because that means a 100-dpi figure is different than 200 dpi figure. You will end up with the inconsistent situation that a pdf that has 72 dpi virtual pixels will have fewer ticks than a png with 100 dpi.

If your argument were instead to do the deconflict as a fraction of tick width, I'd be fine w/ that. linewidths are in physical units.

Anyway, this comment wasn't meant at all to be a blocker on this PR. The overall idea seems fine to me.

The DPI dependence is a very good point. I would be comfortable tweaking the exact de-confliction criteria later.

tacaswell · 2019-04-08T14:16:23Z

The test coverage drop is due to travis not firing. The tests that do not have a png version or not run on windows and azure does not report coverage yet.

tacaswell · 2019-04-09T20:03:46Z

Closing in favor of #13908 which puts this functionality in the Axis object instead.

FIX: restore de-confliction logic for minor ticks

7680c19

In matplotlib#13363 when `iter_ticks` was deprecated the in-lined logic did not account for the updates from matplotlib#13314.

tacaswell added the Release critical For bugs that make the library unusable (segfaults, incorrect plots, etc) and major regressions. label Apr 1, 2019

tacaswell added this to the v3.1.0 milestone Apr 1, 2019

tacaswell requested review from anntzer and ImportanceOfBeingErnest April 1, 2019 04:09

anntzer reviewed Apr 1, 2019

View reviewed changes

tacaswell force-pushed the enh_optional_tick_deconflict branch from 98f9043 to 59b9308 Compare April 1, 2019 19:53

ENH: Add API from minor locators to opt-out of tick deconfliction

6642990

This adds the logic to check if there is an attribute on the minor locator to control the deconfliction.

tacaswell force-pushed the enh_optional_tick_deconflict branch from 59b9308 to 6642990 Compare April 1, 2019 19:56

jklymak reviewed Apr 2, 2019

View reviewed changes

lib/matplotlib/axis.py Outdated Show resolved Hide resolved

jklymak reviewed Apr 2, 2019

View reviewed changes

MNT: only do filtering if we need to

62838ad

jklymak reviewed Apr 2, 2019

View reviewed changes

tacaswell mentioned this pull request Apr 9, 2019

Enh control tick deconflict2 #13908

Merged

tacaswell closed this Apr 9, 2019

tacaswell removed this from the v3.1.0 milestone Apr 9, 2019

tacaswell removed the Release critical For bugs that make the library unusable (segfaults, incorrect plots, etc) and major regressions. label Apr 9, 2019

tacaswell added this to the unassigned milestone Apr 9, 2019

		@@ -1326,7 +1328,8 @@ def get_minorticklocs(self):
		tol = (hi - lo) * 1e-5

Uh oh!

Enh optional tick deconflict #13826

Enh optional tick deconflict #13826

Uh oh!

Conversation

tacaswell commented Apr 1, 2019

Uh oh!

anntzer left a comment

Choose a reason for hiding this comment

Uh oh!

timhoffm commented Apr 1, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ImportanceOfBeingErnest commented Apr 1, 2019

Uh oh!

pharshalp commented Apr 1, 2019

Uh oh!

tacaswell commented Apr 1, 2019

Uh oh!

jklymak commented Apr 2, 2019

Uh oh!

tacaswell commented Apr 2, 2019

Uh oh!

jklymak commented Apr 2, 2019

Uh oh!

ImportanceOfBeingErnest commented Apr 2, 2019

Uh oh!

Uh oh!

jklymak Apr 2, 2019

Choose a reason for hiding this comment

Uh oh!

jklymak Apr 2, 2019

Choose a reason for hiding this comment

Uh oh!

tacaswell Apr 2, 2019

Choose a reason for hiding this comment

Uh oh!

jklymak Apr 2, 2019

Choose a reason for hiding this comment

Uh oh!

tacaswell Apr 5, 2019

Choose a reason for hiding this comment

Uh oh!

jklymak Apr 5, 2019

Choose a reason for hiding this comment

Uh oh!

tacaswell Apr 8, 2019

Choose a reason for hiding this comment

Uh oh!

tacaswell commented Apr 8, 2019

Uh oh!

tacaswell commented Apr 9, 2019

Uh oh!

Uh oh!

timhoffm commented Apr 1, 2019 •

edited

Loading