5 minor divisions when major ticks are 2.5 units apart #13069

hershen · 2018-12-31T05:03:04Z

PR Summary

Currently when the major ticks are spaced 2.5 units apart there are 3 minor ticks spaced 0.625 units apart.
This PR increases the number of minor ticks to 4, spaced 0.5 units apart. To me, this is a much more natural spacing.

Before:

After:

from matplotlib import pyplot as plt
import matplotlib as mpl

fix, ax = plt.subplots()

ax.set_xlim(0, 5)
ax.set_xticks([0, 2.5, 5])
ax.minorticks_on()

ax.tick_params(which='major', length=7)

ax.yaxis.set_major_locator(mpl.ticker.NullLocator())

PR Checklist

[x ] Has Pytest style unit tests
[x ] Code is Flake 8 compliant

hershen · 2018-12-31T19:36:00Z

As a side note, I personally feel that major ticks spaced 1 unit apart are more naturally divided into 4 minor divisions (spaced 0.25 units apart) instead of the current 5 minor divisions (spaced 0.2 units apart). If others agree, I can adjust that in this PR.

timhoffm · 2019-01-01T12:07:45Z

I‘m also positive on on changing the subdivision of 1 unit from 5 to 4.

hershen · 2019-01-01T18:39:23Z

Thanks @timhoffm. I'll wait a bit to see if there's anyone opposed.

jklymak · 2019-01-01T20:50:08Z

I'm not necessarily "opposed", but this seems to be special-casing a major tick interval you had to specify by hand. I don't know that its reasonable to special case all the possible manual tick intervals folks might give us in the minor tick code. If you specify the major ticks by hand, I think its reasonable to specify the minor ticks by hand. But maybe I'm missing a subtlety whereby this is more useful in general.

hershen · 2019-01-01T22:04:01Z

Ah, sorry, I see that example "masks" the fact that 2.5 is one of the default major tick intervals.
The issue can be seen without setting the major ticks:

from matplotlib import pyplot as plt

fig, ax = plt.subplots()

plt.plot([0, 10, 20], [0, 1, 2])
ax.minorticks_on()

plt.show()

It's just more "crowded" so I tried to make it clearer, sacrificing the point that 2.5 major tick interval is one of the defaults.

jklymak · 2019-01-01T22:18:17Z

Ah, OK, thats different. Thanks!

jklymak · 2019-01-01T22:23:57Z

lib/matplotlib/ticker.py

+            majorlocs_no_exponent = 10 ** (np.log10(majorstep) % 1)
+
+            # round to closest 0.5
+            majorlocs_no_exponent = round(majorlocs_no_exponent * 2) / 2


I'm not sure if we want to round 2.4 (for instance) to 2.5 do we?

majorlocs_no_exponent is only used to decide if there should be 4 or 5 minor intervals. The rounding is there to take care of floating point inaccuracies so the major tick step interval can be searched for in the following line (number 2627).
There should be no effect on the major ticks (if that's your concern).

P.S. I just realized that the name of that variable is wrong - it should be majorstep_no_exponent. I'll fix that.

No it’s that you are rounding 2.4 to 2.5 so if that was the major tick the minor ticks will now be different. I think you want to use np.isclose or some such to check and allow some float inaccuracies.

Ah, ok. Thanks. Replaced the round with np.isclose.

hershen · 2019-01-02T03:36:00Z

lib/matplotlib/tests/test_ticker.py

+
+    # make sure we're checking all major steps in ticker.AutoLocator
+    majorsteps = [x[0] for x in majorstep_minordivisions]
+    assert np.allclose(majorsteps, mticker.AutoLocator()._steps)


I wanted to make sure that the major steps in majorstep_minordivisions are the ones in AutoLocator()._steps with the goal that test_number_of_minor_ticks will fail if that's not true. So I added this assert outside any test function.
But if this assert fails pytest emits an error in the collection phase and doesn't run any tests.
Any suggestions how to improve this?

Just move it to its own test with the comment that it's really intended to test that the next test is correctly parametrized.

Done - thanks.
I created a staticmethod, because it's the only way I found for the parameters to be accessible both in a test and by the @pytest.mark.parametrize decorator.

Just make it a class variable like params just above?

anntzer · 2019-01-19T10:34:22Z

lib/matplotlib/ticker.py

+
+            majorstep_no_exponent = 10 ** (np.log10(majorstep) % 1)
+
+            # The part after the 'or' is for #13069 to make minimal changes to


Please make the comment understandable without refering to github (perhaps just describe the desired behavior instead of refering to the behavior change).

Removed comment as it's no longer relevant.

anntzer · 2019-01-19T10:44:57Z

lib/matplotlib/ticker.py

+            # behaviour
+            if np.isclose(majorstep_no_exponent, 2.5) or (
+                    (not np.isclose(majorstep_no_exponent, [1, 10]).any() and
+                     int(np.round(majorstep_no_exponent)) in [1, 5, 10])):


The two conditions (not close to 1 or 10, but possibly in [1 or 10]) seem contradictory.
Given that you are already going to use round(), I would just use its ability to round to one decimal point:

majorstep_no_exponent = round(10**(np.log10(majorstep) % 1), 1) # (in fact the following looks more natural to me, but heh) majorstep_no_exponent = round(majorstep / 10**int(np.log(majorstep)), 1) if majorstep_no_exponent in [2.5, 5]: ... else: ...

Thanks @anntzer for the comments.

The conditions are not close to 1 or 10 and when rounded are in [1 or 10], so not always contradictory (for example 1.1 satisfies both).
I was trying to make as minimal a change to the number of minor ticks calculation.
Rounding to 1 decimal place:

if round(majorstep_no_exponent, 1) in [2.5, 5]: ndivs = 5 else: ndivs = 4

will change the number of minor divisions from 5 to 4 for major ticks spaced 1.1 units apart (for example), compared to the situation before this PR. @jklymak indicated this is undesirable. The code is a bit ugly because of this, though.

I think the calculation used % in order to guarantee 1 <= majorstep_no_exponent <= 10 because that's the range of the steps in AutoLocator. majorstep / 10**int(np.log10(majorstep)) can return values outside this range (for example if marjostep is 0.5).

Re 1: I personally think if we're already changing the minorstep mapping for the common case of 1/5/10-spaced ticks (which appear by default), it's really not a big leap to also change the case of 1.1-spaced ticks (which don't even appear by default). This just needs to be documented... (plus, I'm pretty sure there's a lot of other places in the code which probably don't work so well with "weird" tick spacings like 1.1)
Re 2: Sorry, that should have been m/10**np.floor(np.log10(m)) (int rounds towards zero, but I want to round down), which works for m<1 too. Anyways, the formulas are equivalent, no big deal.

Simplified condition according to @anntzer's suggestion.

hershen · 2019-01-22T23:59:29Z

I went ahead and adjusted the number of minor ticks from 5 to 4 when the major ticks are spaced 1 or 10 units apart.

anntzer · 2019-01-23T10:15:50Z

Needs a changelog entry too.

hershen · 2019-01-25T23:34:42Z

I added an entry in doc/users/next_whats_new/. I hope that's the right place for it.

I think that addresses all current comments.
Additional comments welcome :)

anntzer

Leaving this open for comments for a bit longer as this is a behavior change, but looks reasonable to me.

efiring · 2019-01-28T20:26:25Z

lib/matplotlib/ticker.py

+
+            majorstep_no_exponent = 10 ** (np.log10(majorstep) % 1)
+
+            if np.isclose(majorstep_no_exponent, [2.5, 5.0]).any():


I do not think that removing 1, 10 from this list is a good idea.

Changed to the original 5 divisions for 1, 10.

efiring

Please put 1, 10 back in the list for 5 minor ticks.

anntzer · 2019-01-28T20:28:13Z

Agreed during the call that this could be merged after restoring the old behavior for 1 & 10.

hershen · 2019-01-28T20:54:30Z

My preference for 4 divisions for 1 and 10 is not strong (and might be just because that's what I'm used to). But I'd be interested to know why the consensus is that 5 divisions is better for 1 and 10.

efiring · 2019-01-28T21:16:28Z

One advantage of 5 minors for intervals of 1 or 10 is that it minimizes the numbers of digits of precision when labeling. It's avoiding using fractions when not needed. It's also a bit like the difference between a metric ruler and an English ruler: the former uses millimeters between centimeters, while the latter uses eighths or sixteenths. Which is more useful, and which is expected, depends on context.
The other reason for not wanting to make this change is that we try to be conservative about style changes, making them only when the rationale is clear and minimally controversial.

hershen · 2019-01-28T21:32:12Z

Thanks for the explanation @efiring.
I see the advantages.

Split and expand related test.

hershen · 2019-02-11T03:54:20Z

Anything else I can do to make this PR more mergeable?

efiring · 2019-02-11T06:01:08Z

The test failure was unrelated, so I went ahead with the merge. Thank you for the contribution.

timhoffm approved these changes Jan 1, 2019

View reviewed changes

jklymak added the topic: ticks axis labels label Jan 1, 2019

jklymak reviewed Jan 1, 2019

View reviewed changes

hershen commented Jan 2, 2019

View reviewed changes

hershen force-pushed the five_minor_ticks_2.5_major_ticks branch from 55135d6 to d079d67 Compare January 19, 2019 02:15

anntzer reviewed Jan 19, 2019

View reviewed changes

hershen changed the title ~~5 minor divisions when major ticks are 2.5 units apart~~ 5 minor divisions when major ticks are 2.5 units apart and 4 minor divisions when major ticks are 1 or 10 units apart Jan 22, 2019

hershen force-pushed the five_minor_ticks_2.5_major_ticks branch from a769bef to cfa5393 Compare January 25, 2019 23:06

anntzer approved these changes Jan 25, 2019

View reviewed changes

tacaswell added this to the v3.1 milestone Jan 28, 2019

efiring reviewed Jan 28, 2019

View reviewed changes

efiring requested changes Jan 28, 2019

View reviewed changes

hershen force-pushed the five_minor_ticks_2.5_major_ticks branch from cfa5393 to 67b22e4 Compare January 28, 2019 21:25

hershen changed the title ~~5 minor divisions when major ticks are 2.5 units apart and 4 minor divisions when major ticks are 1 or 10 units apart~~ 5 minor divisions when major ticks are 2.5 units apart Jan 28, 2019

Adjust minor tick spacing for major ticks spaced 2.5 units apart.

4e227bd

Split and expand related test.

hershen force-pushed the five_minor_ticks_2.5_major_ticks branch from 67b22e4 to 4e227bd Compare February 4, 2019 21:33

efiring approved these changes Feb 11, 2019

View reviewed changes

efiring merged commit d01b3d1 into matplotlib:master Feb 11, 2019


		majorstep_no_exponent = 10 ** (np.log10(majorstep) % 1)

		# The part after the 'or' is for #13069 to make minimal changes to


		majorstep_no_exponent = 10 ** (np.log10(majorstep) % 1)

		if np.isclose(majorstep_no_exponent, [2.5, 5.0]).any():

Uh oh!

5 minor divisions when major ticks are 2.5 units apart #13069

5 minor divisions when major ticks are 2.5 units apart #13069

Uh oh!

Conversation

hershen commented Dec 31, 2018

PR Summary

PR Checklist

Uh oh!

hershen commented Dec 31, 2018

Uh oh!

timhoffm commented Jan 1, 2019

Uh oh!

hershen commented Jan 1, 2019

Uh oh!

jklymak commented Jan 1, 2019

Uh oh!

hershen commented Jan 1, 2019

Uh oh!

jklymak commented Jan 1, 2019

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!

hershen Jan 23, 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!

Choose a reason for hiding this comment

Uh oh!

hershen commented Jan 22, 2019

Uh oh!

anntzer commented Jan 23, 2019

Uh oh!

hershen commented Jan 25, 2019

Uh oh!

anntzer 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!

efiring left a comment

Choose a reason for hiding this comment

Uh oh!

anntzer commented Jan 28, 2019

Uh oh!

hershen commented Jan 28, 2019

Uh oh!

efiring commented Jan 28, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

hershen commented Jan 28, 2019

Uh oh!

hershen commented Feb 11, 2019

Uh oh!

hershen Jan 23, 2019 •

edited

Loading

efiring commented Jan 28, 2019 •

edited

Loading