Description
Bug summary
When creating a nested grid of axes using GridSpecFromSubplotSpec (EG by calling axis.get_subplotspec().subgridspec(...)), and plotting a figure using layout="constrained", the nested axes are not displayed correctly. Specifically, the inner grids do not respect the spacing between the outer grids, as shown below.
Code for reproduction
import matplotlib.pyplot as plt
import matplotlib.axes
import matplotlib.gridspec
import matplotlib._layoutgrid
def make_demo(
plot_name,
outer_space=0.1,
inner_space=0.1,
layout=None,
):
file_name = "%s.png" % plot_name
print(file_name)
figure = plt.figure(figsize=[10, 6], layout=layout)
grid_spec = matplotlib.gridspec.GridSpec(
nrows=1,
ncols=2,
figure=figure,
wspace=outer_space,
hspace=outer_space,
width_ratios=[2, 1],
)
axis_array = grid_spec.subplots(squeeze=False)
a0, a1 = axis_array.flatten().tolist()
assert isinstance(a0, matplotlib.axes.Axes)
assert isinstance(a1, matplotlib.axes.Axes)
for axis in [a0, a1]:
axis.get_xaxis().set_visible(False)
axis.get_yaxis().set_visible(False)
for s in axis.spines.values():
s.set(color="r", lw=10)
subplot_spec = axis.get_subplotspec()
subgrid_spec = subplot_spec.subgridspec(
nrows=3,
ncols=3,
wspace=inner_space,
hspace=inner_space,
)
axis_array = subgrid_spec.subplots(squeeze=False)
figure.suptitle(plot_name, fontsize=25)
figure.savefig(file_name)
def fix_layout_grid():
defaults = matplotlib._layoutgrid.LayoutGrid.__init__.__defaults__
defaults = list(defaults)
defaults[2] = True
defaults = tuple(defaults)
matplotlib._layoutgrid.LayoutGrid.__init__.__defaults__ = defaults
if __name__ == "__main__":
make_demo("1_original")
make_demo("2_more_outer_space", outer_space=0.8)
make_demo("3_more_inner_space", outer_space=0.8, inner_space=0.8)
make_demo("4_constrained", layout="constrained")
make_demo("5_constrained_more_outer_space", layout="constrained", outer_space=0.3)
make_demo("6_constrained_more_inner_space", layout="constrained", outer_space=0.3, inner_space=0.3)
fix_layout_grid()
make_demo("7_fixed_constrained", layout="constrained")
make_demo("8_fixed_constrained_more_outer_space", layout="constrained", outer_space=0.3)
# filenames = [
# "1_original.png",
# "2_more_outer_space.png",
# "3_more_inner_space.png",
# "4_constrained.png",
# "5_constrained_more_outer_space.png",
# "6_constrained_more_inner_space.png",
# "7_fixed_constrained.png",
# "8_fixed_constrained_more_outer_space.png",
# ]
# from jutility import util, plotting
# mp = plotting.MultiPlot(
# *[plotting.ImShow(util.load_image(s)) for s in filenames],
# colour="grey",
# )
# mp.save()Actual outcome
In images 1-3 below (not constrained layout), the inner grids respond to the spacing between the outer grids. However, when using layout="constrained" in images 4-6 below (before my workaround is applied), the inner grids ignore the spacing between the outer grids.
Expected outcome
The inner grids should respond to the spacing between the outer grids (which is the case in images 7-8 after my workaround is applied).
Additional information
I have found a workaround, which doesn't involve modifying the matplotlib installation (included in the fix_layout_grid function above):
defaults = matplotlib._layoutgrid.LayoutGrid.__init__.__defaults__
defaults = list(defaults)
defaults[2] = True
defaults = tuple(defaults)
matplotlib._layoutgrid.LayoutGrid.__init__.__defaults__ = defaultsI have also found 2 possible fixes (either one works, the first is more conservative):
- In https://github.com/matplotlib/matplotlib/blob/v3.9.2/lib/matplotlib/_constrained_layout.py#L226 , in
layoutgrids[rep] = mlayoutgrid.LayoutGrid(...), includeparent_inner=True - In https://github.com/matplotlib/matplotlib/blob/v3.9.2/lib/matplotlib/_layoutgrid.py#L36 , in
LayoutGrid.__init__(...), changeparent_inner=Falsetoparent_inner=True
Operating system
Ubuntu
Matplotlib Version
3.9.2
Matplotlib Backend
qtagg
Python version
Python 3.10.12
Jupyter version
No response
Installation
pip