Matplotlib animations are clipped

[jenssss]

ALL software version info

holoviews 1.14.4 pyhd3eb1b0_1
matplotlib 3.3.4 py39h06a4308_0
python 3.9.5 h12debd9_4
Ubuntu 20.04.2 LTS

Description of expected behavior and the observed behavior

When plotting an animation with the matplotlib backend, the edges of the figure are clipped. This happens for both gif and webm animations. When saving the same figure to a static figure such as png, the edges are not clipped.

Complete, minimal, self-contained example code that reproduces the issue

import holoviews as hv
import numpy as np

hv.extension("matplotlib")

frequencies = np.linspace(0.5, 0.6, 5)

def sine_curve(phase, freq):
    xvals = [0.1* i for i in range(100)]
    return hv.Curve((xvals, [np.sin(phase+freq*x) for x in xvals]))

curve_dict = {f:sine_curve(0,f) for f in frequencies}
hmap = hv.HoloMap(curve_dict, kdims='frequency')

hv.save(hmap, 'sines.webm', fmt='webm')
hv.save(hmap, 'sines.gif', fmt='gif', fps=5)
hv.save(hmap, 'sines.png', fmt='png')

layout = hmap + hmap
hv.save(layout, 'sines_layout.webm', fmt='webm')
hv.save(layout, 'sines_layout.gif', fmt='gif', fps=5)
hv.save(layout, 'sines_layout.png', fmt='png')

Screenshots or screencasts of the bug in action

sines.png file showing no clipping

sines.gif files showing clipping of the y-label, and a wide right margin

Same as above for a layout. Here the clipping is even more clear, affecting the title in addition to the left and lower margins

[jenssss]

I did a bit of digging, and saving the first frame of the animation

anim = hv.render(layout)
fig = anim._fig
anim._fig.savefig("sines0.png")

with layout from the above code also gives a clipped image

So it does not appear that the problem is in the animation stage, but rather in each frame figure. Maybe the axes are being put in different positions in an animation compared to a static image export?

[jenssss]

I found a workaround for this issue, by manually tuning the fig_bounds option:

layout.opts(fig_bounds=(0.1, 0.1, 0.95, 0.9))

anim = hv.render(layout)
anim._fig.savefig("sines_fig_bounds.png")

with layout from the first post in this thread. This gives

In this plot the title is still cropped, but this can be worked around with a hook

def move_title_hook(plot, elem):
    fig = plot.state
    text = fig._suptitle
    if text:
        text.set_position((0.5, 0.97))

layout.opts(hv.opts.Curve(hooks=[move_title_hook]))

anim = hv.render(layout)
anim._fig.savefig("sines_title_moved.png")

giving

The underlying cause for this problem seems to be related to what's going on in the _figure_data method of the holoviews.plotting.mpl.renderer.MPLRenderer class

holoviews/holoviews/plotting/mpl/renderer.py

Line 135 in c425b55

def _figure_data(self, plot, fmt, bbox_inches='tight', as_script=False, **kwargs):

For static image export a bounding box containing all the artists is calculated using the _compute_bbox method, and this box ensures that no part of the figure is clipped. This bounding box is not calculated for an animation.

I tried manually calculating a bounding box, and then pass it into the savefig_kwargs in anim_kwargs in the anim.save call in the _anim_data method of the MPLRenderer, but then matplotlib simply ignores this bounding box, with the following warning

Warning: discarding the 'bbox_inches' argument in 'savefig_kwargs' as it may cause frame size to vary, which is inappropriate for animation.

This leads to me think that maybe the best solution is maybe just to add the workaround given above to the documentation, perhaps in the Plotting_with_Matplotlib notebook.

[MastaMinds]

I am here to confirm that this issue still exists to this date.

[ChrisZonghaoLi]

Same issue here...