indicate_inset_zoom sometimes draws incorrect connector lines

Bug report

Bug summary

Depending on location of inset, the connector lines are not drawn to the nearest two corners.

Code for reproduction

import matplotlib.pyplot as plt
import numpy as np
line=np.linspace(10,0, 101)
fig, ax=plt.subplots()
ax.plot(line, line)
axins=ax.inset_axes([0, .5, .4, .4])
axins.plot(line, line)
axins.set_xlim(7,8)
axins.set_ylim(7,8)
ax.indicate_inset_zoom(axins)

Actual outcome

Expected outcome

import matplotlib.pyplot as plt
import numpy as np
line=np.linspace(10,0, 101)
fig, ax=plt.subplots()
ax.plot(line, line)
axins=ax.inset_axes([0, .5, .4, .4])
axins.plot(line, line)
axins.set_xlim(7,8)
axins.set_ylim(7,8)
rectpatch, connects=ax.indicate_inset_zoom(axins)
connects[0].set_visible(False)
connects[1].set_visible(False)
connects[2].set_visible(True)
connects[3].set_visible(True)

Matplotlib version

Operating system: Windows 10 1803
Matplotlib version: 3.0.0
Matplotlib backend (print(matplotlib.get_backend())): module://ipykernel.pylab.backend_inline
Python version: Python 3.6.5 (Anaconda Python)
Jupyter version (if applicable): 6.4.0
Other libraries:

Installed matplotlib 3.0.0 from conda-forge

Actually, I think my "expected outcome" is not the proper solution, but looks the best without modifying the source. My ideal output would have the connecting lines drawn from the two corners of the rectangle closest to the inset to the closest two inset corners.

The algorithm tries to draw the best opposing corners. Your inset axes is basically centered on the indicator on the main axes so it chooses something that looks a little funny. Thats why we put the manual visibility toggles in there, for cases where the naive algorithm returns something that doesn't look nice.

Actually, I think my "expected outcome" is not the proper solution, but looks the best without modifying the source. My ideal output would have the connecting lines drawn from the two corners of the rectangle closest to the inset to the closest two inset corners.

That would be easy, but unexpected. The lines connect corners in the indicator with corners on the inset axes. I'd be against a change that didn't conserve this mapping.

There is mpl_toolkits.axes_grid1.inset_locator.inset_axes and mark_inset. You may get the "desired outcome" with those as

import numpy as np
import matplotlib.pyplot as plt
from mpl_toolkits.axes_grid1.inset_locator import inset_axes, mark_inset

line=np.linspace(10,0, 101)

fig, ax=plt.subplots()
ax.plot(line, line)

axins = inset_axes(ax, "100%", "100%", bbox_to_anchor=[0, .5, .4, .4],
                   bbox_transform=ax.transAxes, borderpad=0)

axins.plot(line, line)
axins.set(xlim=(7,8), ylim=(7,8))

mark_inset(ax, axins, loc1=1, loc2=4, fc="none", ec="0.5")
plt.show()

You may get the "ideal" outcome with a bit of customization.

import numpy as np
import matplotlib.pyplot as plt
from mpl_toolkits.axes_grid1.inset_locator import (inset_axes, TransformedBbox,
                                                   BboxPatch, BboxConnector)

line=np.linspace(10,0, 101)

fig, ax=plt.subplots()
ax.plot(line, line)

axins = inset_axes(ax, "100%", "100%", bbox_to_anchor=[0, .5, .4, .4],
                   bbox_transform=ax.transAxes, borderpad=0)

axins.plot(line, line)
axins.set(xlim=(7,8), ylim=(7,8))

def my_mark_inset(parent_axes, inset_axes, loc1a=1, loc1b=1, loc2a=2, loc2b=2, **kwargs):
    rect = TransformedBbox(inset_axes.viewLim, parent_axes.transData)
    pp = BboxPatch(rect, fill=False, **kwargs)
    parent_axes.add_patch(pp)
    p1 = BboxConnector(inset_axes.bbox, rect, loc1=loc1a, loc2=loc1b, **kwargs)
    inset_axes.add_patch(p1)
    p1.set_clip_on(False)
    p2 = BboxConnector(inset_axes.bbox, rect, loc1=loc2a, loc2=loc2b, **kwargs)
    inset_axes.add_patch(p2)
    p2.set_clip_on(False)

    return pp, p1, p2

my_mark_inset(ax, axins, loc1a=1, loc1b=2, loc2a=4, loc2b=3, fc="none", ec="0.5")

plt.show()

OK, but fair warning that the goal is to deprecate axes_grid1. If you want to make your own connection lines, you can use patches.ConnectionPatch:

xyA = (1, 0)
xyB = (bounds[0], bounds[0])
connect = mpatches.ConnectionPatch(xyA, xyB, 'axes fraction', 'data',
                            axesA=inset_ax, axesB=parent_ax, arrowstyle="-", edgecolor=edgecolor)
self.add_patch(connect)

will add a line from the lower right of the inset to the lower left of the inset indicator.

import matplotlib.pyplot as plt
import numpy as np
import matplotlib.patches as mpatches

line=np.linspace(10,0, 101)
fig, ax=plt.subplots()
ax.plot(line, line)
axins=ax.inset_axes([0, .5, .4, .4])
axins.plot(line, line)
xlims = (7, 8)
ylims = (7, 8)
axins.set_xlim(xlims)
axins.set_ylim(ylims)
rectpatch, connects=ax.indicate_inset_zoom(axins)
for con in connects:
    con.set_visible(False)

for j in range(2):
    # inset axes co-ordinates
    xyA = (1, j)
    # data co-ordinates
    xyB = (xlims[0], ylims[j])
    connect = mpatches.ConnectionPatch(xyA, xyB, 'axes fraction', 'data',
                                axesA=axins, axesB=ax, arrowstyle="-", edgecolor='0.6')
    ax.add_patch(connect)

plt.show()

Some people's goal it is to deprecate axes_grid1. I would be careful to make that a self-fulfilling prophecy. People's bandwidth is much better used with solving real-world problems than with deconstructing parts of the library.

I think indicate_inset_zoom (or indicate_inset) could well take 4 location parameters as input, similar to the hand-written function above, or as part of a single list of dict, to allow the user to easily modify the connection lines.

Maintaining and documenting axes_grid1 was also taking up a lot of bandwidth. 😉

In this case the new API has the same coverage as the old.

We could discuss a more flexible connector line API, but again, there is a cost/benefit to doing so.

I think getting the "expected result" automatically is not too difficult, but the so-called "ideal result" is perhaps not so easy and open to debate as to whether it's wanted. I know how to write the former at least and can open a PR for it.

I don't feel like what's called "ideal" here should be the default. The "expected" is probably what people would expect indeed.
But giving users the option to easily obtain their custom desired look is very desireable in my eyes.

Changing the defaults is fine w me. It’s just an aesthetic decision and for most placements I thought what was implemented was the best without doing anything too complicated. But if something can be done that’s reasonable that’s be great.

Expanding the API could also be nice, but a bit of thought should go into the best form.

Thanks for the interest in improving this!

The immediate issue ("expected outcome") is solved by #12334. Should this be reopened to discuss adding customization ("ideal outcome")?

You should leave it closed. I think the "expected outcome" is correct and my "ideal outcome" was not, after reading the comments.

@dstraus

Thanks for the report @dstraus - that lead to a nice improvement by @QuLogic !

Navigation Menu

Repository navigation

indicate_inset_zoom sometimes draws incorrect connector lines #12323

Bug report

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

indicate_inset_zoom sometimes draws incorrect connector lines #12323

Description

Bug report

Activity

dstraus commented on Sep 28, 2018

jklymak commented on Sep 28, 2018

ImportanceOfBeingErnest commented on Sep 28, 2018

jklymak commented on Sep 28, 2018

jklymak commented on Sep 28, 2018

ImportanceOfBeingErnest commented on Sep 28, 2018

jklymak commented on Sep 28, 2018

QuLogic commented on Sep 29, 2018

ImportanceOfBeingErnest commented on Sep 29, 2018

jklymak commented on Sep 29, 2018

ImportanceOfBeingErnest commented on Sep 29, 2018

dstraus commented on Sep 30, 2018

jklymak commented on Sep 30, 2018

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions