Axes.axline: implement support transform argument (for points but not slope) #18647

johan12345 · 2020-10-03T15:09:46Z

PR Summary

for plotting a line through a point in axes coordinates and with a slope in data coordinates
see #18625 for details and example.

PR Checklist

Has pytest style unit tests (and pytest passes).
Is Flake 8 compliant (run flake8 on changed files to check).
New features are documented, with examples if plot related.
Documentation is sphinx and numpydoc compliant (the docs should build without error).
Conforms to Matplotlib style conventions (install flake8-docstrings and pydocstyle<4 and run flake8 --docstring-convention=all).
New features have an entry in doc/users/next_whats_new/ (follow instructions in README.rst there).
API changes documented in doc/api/next_api_changes/ (follow instructions in README.rst there).

for plotting a line through a point in axes coordinates and with a slope in data coordinates fixes matplotlib#18625

jklymak · 2020-10-03T15:58:49Z

When we originally put in axline, we had a big back and forth about how to specify the line, and it was decided that two points were just as good as point/slope. So I'm not sure this feature is particularly needed, though its actually how I would have preferred specifying the line.

johan12345 · 2020-10-03T16:10:41Z

Point and slope is already available in axline. This is specifically about specifying the point in axes coordinates instead of data coordinates, so that the line stays at the same position when moving the viewport.

jklymak · 2020-10-03T16:34:44Z

oops, sorry, I see...

lib/matplotlib/axes/_axes.py

jklymak · 2020-10-03T16:44:01Z

lib/matplotlib/tests/test_axes.py

@@ -4072,6 +4072,19 @@ def test_axline(fig_test, fig_ref):
    ax.axvline(-0.5, color='C5')


+@check_figures_equal()
+def test_axline_transform(fig_test, fig_ref):


I think it would be useful to add a test where you call the method, and then subsequently resize the figure (simulating a manual resize) and show this still yields the expected line. Also that its robust to panning and zooming.

Sure. I added some an additional test in 45e5bc7 which uses set_xlim/set_ylim and set_size_inches to simulate resize and pan/zoom, is that what you meant?

jklymak · 2020-10-03T16:46:42Z

lib/matplotlib/lines.py

+    def get_transform(self):
+        ax = self.axes
+        (vxlo, vylo), (vxhi, vyhi) = ax.transScale.transform(ax.viewLim)
+        x, y = (ax.transAxes + ax.transData.inverted()).transform(self._xy1)


Wouldn't it be simpler to calculate the slope into axes coordinates and simply draw the line in axes co-ordinates?

Good idea. I tried to stay close to the implementation of axline (as the original idea was to add this behavior into that function if a transform is passed, see #18625), but this may be an easier option.

Hm, but it won't really get much simpler because we still need to somehow convert to data coordinates to calculate at which edges of the viewport (top/bottom or left/right) the ends of the line would be. That will of course depend on the current x/y limits.

Wont the line clip if you just make it go from x_axes = -1 to 2?

Ah, right, it does. I implemented a simpler solution in da040d0, however it breaks the tests even though the plots visually seem to look the same. The difference seems to be very small, not sure why this is happening 😕

You could probably safely up the tolerance if this is just roundoff from passing the positions through the transform stack...

@check_figures_equal(extensions=['png'], tol=0.19)

jklymak · 2020-10-03T16:48:19Z

Overall this seems good, and I guess I agree that we can't just make this an example. I'm not at all sure about the name chosen?

Be sure to ping us for reviews if this starts slipping down the PR list. Things get forgotten....

johan12345 · 2020-10-03T17:25:06Z

I'm not at all sure about the name chosen?

Yeah, I'm not sure either. Some other ideas: axline_fixed (i.e. fixed to the viewport), axline_axcoord (i.e. axes coordinates)

anntzer · 2020-10-03T20:44:51Z

From the peanut gallery, I would rather add a special case in the codepath of axline (even something as direct as if transform == self.transAxes and <a slope was passed>: return ... (and document it) rather than another toplevel method, this will avoid the API growing further and further with more variants (axline_transfigure, etc.).

timhoffm · 2020-10-03T21:22:41Z

I'd like to see use-cases to guide us in the naming and exact functionality.

Proposed so far:

A regular skewed grid filling the viewport.

I currently don't see any others: The slope is in data space, but the positions are in Axes coordinates, so the positions are data-independent; i.e. the lines only describe "a data direction" (maybe that's already a hint for a name). If it's just a direction without position, would we not anyway want a grid to indicate the direction homogeneously across all positions? If so, we might want to instead make a function that directly creates the grid (e.g. directional_grid()).

jklymak · 2020-10-03T23:36:14Z

I agree with the above, though I somewhat cringe at the idea of a grid API versus just a single line. Specifying a sloping grid in axes co-ordinates in a general way is not trivial.

I also agree that maybe this should be handled by axline. I think the original discussion made it clear that directly using transform was problematic, but maybe renaming as point_transform would alleviate that concern?

jklymak · 2020-10-04T20:51:31Z

I'll add this one to the agenda of the weekly call since the API choices seem complicated.

… axline

jklymak · 2020-10-20T18:15:13Z

@johan12345 We discussed this on the weekly developer call.

The consensus was that this can and should be handled with the transform argument to axline. If the user provides a point and slope, the point would be transformed by the transform argument. If the user provides two points, the line would be in the transformed space. The default transform would still be ax.transData, but ax.transAxes should give the behaviour you are after. Its not clear to me if this should be limited so that strange transforms are prevented, but, I've not thought about it personally.

Hopefully that is not too annoying, as you have gone a ways down this path, but the consensus was that adding a transform was the easiest API...

jklymak · 2020-10-20T19:34:45Z

Ooops sorry I see Tim commented on the original issue. Sorry for the noise.

johan12345 · 2020-10-21T06:23:47Z

@jklymak Thanks! Yes, as written in the original issue #18625 I have now tried to adapt axline to do this (9292f18). With that, the alternative solution for the calculation of axline_transaxes implemented da040d0 was discarded, to have a common solution based on the previous implementation of axline.

One thing that is not implemented for the case where a transformation is passed is the adjustment of data limits:

What I'm not sure about is the adjustment of data limits. For the moment, I set it to simply not adjust the data limits if a transform is passed (because at least for this use case I think it doesn't really make sense to adjust them).

What do you think?

jklymak · 2020-10-21T13:53:37Z

I mean if the line is in Axes space (but with a data slope) I don't think the intention is for it to be treated as data so the limits should not change? But I don't 100% understand the use of the feature....

johan12345 · 2020-10-27T21:57:45Z

Sure, next_whats_new was added in db44748 and an example in 690177e.

examples/pyplots/axline.py

doc/users/next_whats_new/axline_transform.rst

Co-authored-by: Tim Hoffmann <[email protected]>

johan12345 · 2020-10-30T07:26:32Z

Thanks @timhoffm!

doc/users/next_whats_new/axline_transform.rst

lib/matplotlib/lines.py

(already done by super.__init__())

QuLogic · 2020-12-04T05:31:14Z

So code coverage is complaining, and though it's pointing to a weird place, it is correctly finding a bug: the ._transform is never None. This is because axline calls self.set_artist_props(line) which will set the transform to transData if not set. (Note that it checks a separate property ._transformSet so setting ._transform only would still have done this.)

However, this turns out to work, since transform - transData == IdentityTransform, and axline doesn't support alternate scales, so transScale is also an identity.

So perhaps that check is unnecessary and handling None transform is not needed.

johan12345 · 2020-12-04T07:26:33Z

Well, that is probably what changed when I removed the explicit setting of self._transform... Okay, I can remove that as well.

QuLogic · 2020-12-04T08:12:49Z

No, as I said, it checks ._transformSet, which would not have been changed by setting ._transform in the old code.

johan12345 · 2020-12-06T21:16:55Z

Ah, okay. I removed the unused code in a2584af, and also improved the test coverage a bit more in a5aa349.

QuLogic · 2020-12-15T07:16:35Z

Thanks @johan12345! Congratulations on your first PR to Matplotlib 🎉 We hope to hear from you again.

implement a function axline_transaxes

cf835e3

for plotting a line through a point in axes coordinates and with a slope in data coordinates fixes matplotlib#18625

johan12345 mentioned this pull request Oct 3, 2020

Axes.axline: allow to specify position in axes coordinates instead of data coordinates #18625

Closed

jklymak reviewed Oct 3, 2020

View reviewed changes

lib/matplotlib/axes/_axes.py Outdated Show resolved Hide resolved

jklymak reviewed Oct 3, 2020

View reviewed changes

lib/matplotlib/axes/_axes.py Outdated Show resolved Hide resolved

jklymak reviewed Oct 3, 2020

View reviewed changes

jklymak added this to the v3.4.0 milestone Oct 3, 2020

jklymak added the New feature label Oct 3, 2020

johan12345 added 3 commits October 3, 2020 18:58

improve docs

527aa46

improve tests

45e5bc7

fix special case for slope=0

bd8e82b

johan12345 added 2 commits October 3, 2020 20:32

implement simpler solution for calculation of axline_transaxes

da040d0

fix flake8 errors

02b8c59

johan12345 force-pushed the master branch from c91537b to f39d35e Compare October 20, 2020 17:43

replace separate function axline_transaxes with new functionality for…

9292f18

… axline

johan12345 force-pushed the master branch from f39d35e to 9292f18 Compare October 20, 2020 17:44

remove unused imports

f6ee501

add next_whats_new entry

db44748

add demo for axline with transform parameter

690177e

timhoffm reviewed Oct 29, 2020

View reviewed changes

examples/pyplots/axline.py Outdated Show resolved Hide resolved

doc/users/next_whats_new/axline_transform.rst Outdated Show resolved Hide resolved

johan12345 and others added 2 commits October 30, 2020 08:25

Update examples/pyplots/axline.py

257be32

Co-authored-by: Tim Hoffmann <[email protected]>

Update doc/users/next_whats_new/axline_transform.rst

425ee2d

Co-authored-by: Tim Hoffmann <[email protected]>

fix docs warning

4649b9f

timhoffm reviewed Oct 30, 2020

View reviewed changes

doc/users/next_whats_new/axline_transform.rst Outdated Show resolved Hide resolved

what's new: re-wrap and fix duplicated *xy2*

82e63a9

timhoffm approved these changes Oct 31, 2020

View reviewed changes

QuLogic reviewed Dec 3, 2020

View reviewed changes

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

remove unnecessary set of self._transform

21deaf4

(already done by super.__init__())

QuLogic closed this Dec 3, 2020

QuLogic reopened this Dec 3, 2020

johan12345 added 3 commits December 6, 2020 22:00

remove unused code

a2584af

improve test coverage by testing for ValueError

a5aa349

fix exception description

91e851e

timhoffm added the status: needs review label Dec 12, 2020

QuLogic approved these changes Dec 15, 2020

View reviewed changes

QuLogic merged commit b556e62 into matplotlib:master Dec 15, 2020

QuLogic removed the status: needs review label Dec 15, 2020

anntzer mentioned this pull request Feb 3, 2021

axline transform support broke axline in loglog scale #19445

Closed

johan12345 mentioned this pull request Feb 9, 2021

fix regression of axline behavior with non-linear scales #19485

Merged

7 tasks

oscargus mentioned this pull request Aug 23, 2024

[Bug]: Minor issue - Drawing an axline sets slopes less than 1E-8 to 0 #28386

Closed

Uh oh!

Axes.axline: implement support transform argument (for points but not slope) #18647

Axes.axline: implement support transform argument (for points but not slope) #18647

Uh oh!

Conversation

johan12345 commented Oct 3, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

PR Summary

PR Checklist

Uh oh!

jklymak commented Oct 3, 2020

Uh oh!

johan12345 commented Oct 3, 2020

Uh oh!

jklymak commented Oct 3, 2020

Uh oh!

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!

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!

jklymak commented Oct 3, 2020

Uh oh!

johan12345 commented Oct 3, 2020

Uh oh!

anntzer commented Oct 3, 2020

Uh oh!

timhoffm commented Oct 3, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

jklymak commented Oct 3, 2020

Uh oh!

jklymak commented Oct 4, 2020

Uh oh!

jklymak commented Oct 20, 2020

Uh oh!

jklymak commented Oct 20, 2020

Uh oh!

johan12345 commented Oct 21, 2020

Uh oh!

jklymak commented Oct 21, 2020

Uh oh!

johan12345 commented Oct 27, 2020 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Uh oh!

Uh oh!

johan12345 commented Oct 30, 2020

Uh oh!

Uh oh!

Uh oh!

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

Uh oh!

Uh oh!

johan12345 commented Dec 4, 2020

Uh oh!

QuLogic commented Dec 4, 2020

Uh oh!

johan12345 commented Dec 6, 2020

Uh oh!

QuLogic commented Dec 15, 2020

Uh oh!

Uh oh!

johan12345 commented Oct 3, 2020 •

edited

Loading

timhoffm commented Oct 3, 2020 •

edited

Loading

johan12345 commented Oct 27, 2020 •

edited

Loading

QuLogic commented Dec 4, 2020 •

edited

Loading