seaborn.objecs.Plot.save does not include legend

[evanhorn]

I am very excited about the new seaborn.objects interface. However, when using it and save a plot to a png file, the legend is omitted, but it is shown when plotting to the screen. (See below — first image is a screen capture, second image is the same image saved as a file.)

[evanhorn]

As a workaround, I found that the legend is saved with the plot if, rather than p.save("test.png"), I use

f = plt.figure()
p.on(f)
f.savefig("test.png")

where p is the Seaborn Plot object.

[mwaskom]

Yeah this is a rough edge. The easiest thing to do would be to delegate this to matplotlib, but their auto layout engines can't handle a figure legend. There's some code in FacetGrid to expand the figure size and then shrink the subplots to fit the legend, but it's pretty brittle and I wanted to avoid bringing it over to the new interface unless necessary. So this was left without a clear solution to avoid getting stuck committing to a suboptimal approach.

What's happening when you show the plot in a notebook is that it's saving the figure with bbox_inches="tight", which expands the figure so that all elements fit. The downside to this is that the figure size ends up different (larger or smaller) than what you might specify in theme or layout. But if that's ok, it will work in Plot.save too:

(
    Plot(...)
    .save(fname, bbox_inches="tight")
)

Alternatively, the legend is not placed outside of the plot by default when you use Plot.on because it expects that there might be other subplots / subfigures in that space, and so that it will work well in a GUI context. So (as you discovered) that's another temporary workaround.

[evanhorn]

Thanks for the insight. For what I need, I can get by with the workaround. Now, if I could move the legend, e.g., “upper right”… but I did read about the difficulties dealing with legends.

Regardless, thanks for all your work on this project, it is truly remarkable how great it is to using this package to make great-looking images for presenting my research! I look forward to using the new objects interface… it feels much more like “language of graphics” approach of GGPlot, which, IMO, is a positive direction.

[gfairbro]

I just thought i would add that actually the legend is there, but it is being cut off (apologies if this is obvious). you can just barely see the edge above in the saved file. It's more obvious in my example:

[jalsti]

This seems to be related to #3130. When the plot gets created with pyplot during show(), the legend location gets set fixed to "center left", which is above the plot's graph area. During save() pyplot is not used, which results in the hardcoded position "center right", which is more or less on the outside because of the hardcoded bbox_to_anchor value of (.98, .55) for the legend. With the PR #3128, related to the issue, the legend could be set explicitly.

[leodarrigade]

I could solve this issue by passing bbox_inches='tight' to the save function. It worked when I used the Seaborn save function as

so.Plot(...).save('test.png', bbox_inches='tight)

or when I used Matplotlib:

f = plt.figure()
p.on(f)
f.savefig("test.png", bbox_inches='tight)

Note that in the latter case, if I did not specified bbox_inches='tight' to savefig, the legend was cutoff (contrary to what @evanhorn said).

[madSec]

Hi,
Just found that when showing a plot rather than saving one can use the "legend" method with the "engine" argument as shown below to move the legend into the frame of the plot.

fig = plt.figure()
(
so.Plot()
.add(so.Dot())
.layout(engine="tight")
.on(fig)
.show()
)

Strangely this only affects showing the image in the notebook for example, when saving the image I still need to add "bbox_inches='tight'" to the save methode, and even then in the saved image the legend is located outside the frame.