Hi @Pepe78 that sounds annoying indeed. However, I think that this can likely be resolved with the existing tools in matplotlib. Can you please post about this (with a picture of the output and a description of what you'd expect) on https://discourse.matplotlib.org/ which is the best place for getting help with matplotlib.

Done (reported at https://discourse.matplotlib.org/). As far as "resolved with the existing tools in matplotlib", it would be nice to have by default less amount of blank space as it is done when not doing 3d projections (2d graphs). Thanks!

I'm not sure why there is such a large margin around axes3d plots. I have a sneaking suspicion that the margin was developed outside the subplot paradigm, which also have margins, and now we have two sets of margins. Definitely when you add an axes directly via Axes3d(fig) (which we want to deprecate), the margins are smaller.

This was great advice I got on discourse:

fig = plt.figure(figsize=(20,15), constrained_layout=True)
plt.savefig('1.png', bbox_inches='tight')

Left and right margins were minimal when saved like that (and top and bottom seem also smaller).

I'll actually re-open, because to me, at least, there is still a mystery about why the margins are so much larger than other axes.

That makes sense - as I said before, it would be nice to have less margins on sides by default. When doing 2d plots, this actually really works nicely by default.

Yes, the original Axes3D didn't work within the subplot() framework. And subplots leave extra room for the static axes labels, meanwhile, Axes3D needed to account for extra space for its axes labels within the axes plotting area. Originally, this was fine because you didn't make Axes3D objects using subplots, so you didn't get allocated extra margins for axes labels. When it became possible to specify projection='3d' to a subplot call, this had the inadvertent effect of getting extra margin space that wasn't really needed.

I never did figure out a good solution for this that wouldn't act counter-intuitively. Perhaps Axes3D needs to be reworked such that it can interpret and express the margin parameters itself rather than letting it be handled by 2D-based assumptions?

I stumbled over this issue as well while trying to plot a representation of the Riemann zeta function.

Zooming in just cuts off left and right of the plot. :-/

Just a note that you can control the margin sizes to some extent using gridspec_kws.

fig, ax = plt.subplots(subplot_kw={'projection': '3d'},
                       gridspec_kw=dict(top=1.07, bottom=0.02, left=0, right=1))
fig.savefig('axes3d.png', dpi=600, bbox_inches='tight')

Is there a current workaround for this ? I am having some problems with a figure similar to this situation:

import matplotlib.pyplot as plt
import numpy as np

X = np.arange(-5, 5, 0.25)
Y = np.arange(-5, 5, 0.25)
X, Y = np.meshgrid(X, Y)
R = np.sqrt(X**2 + Y**2)
Z = np.sin(R)


fig, ax = plt.subplots(1, 1, figsize=(10, 5), subplot_kw=dict(projection="3d"))
plt.subplots_adjust(0, 0, 1, 1, 0, 0)


surf = ax.plot_surface(X, Y, Z, linewidth=0, antialiased=False)

ax.view_init(elev=20, azim=-45, roll=0)

ax.set_box_aspect([5, 1, 1])  # Aspect ratio
ax.set_aspect("auto")

plt.show()

Not that I do not want to use the bbox_inches='tight' when saving the plot, as the figure is already prepared at the right size to be inserted in a paper.

Yes, the workaround is stated above - add Axes3D manually.

Could you give a brief example of this? Simply doing

fig = plt.figure(figsize=(10, 5))
ax = Axes3D(fig)

does not show any ax on the figure.

OK, my mistake. I guess I don't know how you can fix this. Maybe @scottshambaugh has some insight. The overall issue is that you set the box_aspect. I guess the algorithm is conservatice about guessing which direction to shrink based on the current view.

Is there any progress regarding topic? I'm having the same problem trying to fit the axis to the full width of the window - left and right sides just get cut off. I've tried many things, like the tight layout, adjusting the margins and other things, but none of them worked.

@AndrejHafner

My current workaround is to play with the ax position ax.set_position([-0.04, -0.32, 1.05, 1.1]) inside the figure.

If for some reason you are using this in a figure, this ax will overlap the other ones, so you might need to hide the background: ax.set_facecolor("none")

if in this figure you are using layout='constrained', you can do:

fig.canvas.draw()
fig.set_layout_engine("none")
axarr[1].set_position([-0.04, -0.32, 1.05, 1.1])
axarr[1].set_facecolor("none")

to fix axis positions from constrained layout, and then disable it to only change the ax position of you 3d figure.

@AndrejHafner

My current workaround is to play with the ax position ax.set_position([-0.04, -0.32, 1.05, 1.1]) inside the figure.

If for some reason you are using this in a figure, this ax will overlap the other ones, so you might need to hide the background: ax.set_facecolor("none")

if in this figure you are using layout='constrained', you can do:

fig.canvas.draw()
fig.set_layout_engine("none")
axarr[1].set_position([-0.04, -0.32, 1.05, 1.1])
axarr[1].set_facecolor("none")

to fix axis positions from constrained layout, and then disable it to only change the ax position of you 3d figure.

@cgadal is the axarr in your code the Axes3D? I'm not sure which axis you are actually setting the position for. I tried this approach, but I couldn't get the plot to become wider.

@AndrejHafner

Yes, it is the Axes3D.

import matplotlib.pyplot as plt
import numpy as np

X = np.arange(-5, 5, 0.25)
Y = np.arange(-5, 5, 0.25)
X, Y = np.meshgrid(X, Y)
R = np.sqrt(X**2 + Y**2)
Z = np.sin(R)

fig = plt.figure(figsize=(10, 5), layout="constrained")
ax = fig.add_subplot(111, projection="3d")

surf = ax.plot_surface(X, Y, Z, linewidth=0, antialiased=False)

ax.view_init(elev=20, azim=-45, roll=0)
ax.set_box_aspect([5, 1, 1])  # Aspect ratio


# fig.canvas.draw()
# fig.set_layout_engine("none")
# ax.set_position([-0.75, -0.75, 2.5, 2.5])
# ax.set_facecolor("none")

plt.savefig("Fig_2.png")
plt.show()

gives:

Uncommenting the commented lines gives:

The two first numbers in ax.set_position will control the position, and the two last ones the scaling factor.

@cgadal

Thank you for the detailed response, it works for my use case as well. I was hoping that I would be able to only change the width and keep the height as it is, but I guess it kind of works as a zoom workaround on the figure level.

There are two issues being talked about here:

1. The default 3D margins are really large (still an open question, and should be the focus of this issue).
2. 3D data is being clipped too aggressively. I have a separate issue for that, and discussion on that should move over here: [Bug]: 3D artist clipbox cuts off data sooner than axis panes/spines/etc #23450.

I took a stab at fixing (2) today but my approach wasn't the right one - I don't think that messing with the box_aspect calcs makes sense with the way 3D axes are set up. I believe the fix will be to expand the clip boxes for each of the 3D artists to the whole figure, since there's no real reason to truncate them. That's available for anyone to pick up and work on.