Seaborn: 'darkgrid' style has trasparent background

[tritemio]

Hi,

I'm not sure this is a seaborn issue but I just update to ipython 3 and the seaborn style darkgrid shows a transparent (white) background and no grid (probably invisible due to the white notebook background).

System info: Windows 7 x64, anaconda 64bits, matplotlib 1.4.3, seaborn 0.5.1, ipython 3.

[mwaskom]

Are you calling %matplotlib inline after you import seaborn?

[tritemio]

Yes, in a few points have to switch to the QT backend for some interactive plot and then I switch back to inline.

[mwaskom]

I think they changed something about how setting inline plots work that overrides the configuration seaborn does; you can call sns.set() or sns.set_style after you change your backend.

[tritemio]

I already do set the style with set_style just after the plot but the background is still transparent (I can tell this because I have a large figure that spills outside the central panel where the background changes to gray).

[mwaskom]

It won't work if you do it after the plot, the style parameters become a property of the figure/axes object at the time it's instantiated, so it doesn't matter what the global parameters are at draw time.

[tritemio]

Ok, thanks. If I do

%matplotlib inline
seaborn.set_style('darkgrid')

it works. I was inverting the order of these two lines. It used to work in ipython 2.x even with inverted order but this solution is good enough to me.

As far as I am concerned you can close the issue, thanks.

[tritemio]

Wait, I still have some redundant %matplotlib inline here and there and the background switches back to transparent.

I mean:

%matplotlib inline
seaborn.set_style('darkgrid')

works, then a cell like

%matplotlib inline

(even without changing to QT in between) switches back to transparent background. I can work around it but just to let you know.

[mwaskom]

Yes, %matplotlib is changing more rc parameters in 3.0 than it did in the past. I doubt there's anything that seaborn could do to work around that (automatically), but I haven't had time to figure out if it's required/intended behavior in IPython or a bug that can be fixed on their end.

[mwaskom]

Opened an issue on IPython, so we'll wee what they do. Closing here as nothing to do on the seaborn side of things.