Add global config for FacetGrid to always follow matplotlib.rcParams['figure.figsize']

[dgkuester]

Is your feature request related to a problem?

Most plots that I generate need to fit specific sizes depending on strict fixed formatting constraints for publication or presentation, particularly in IEEE publciations. To this end, I invested some effort in writing matplotlib style sheets that match the proportions needed for these venues. I was surprised when I noticed that passing col and/or row to the various .plot[.*](...) methods led to different figure sizes.

After some experimentation and reading through the docs (thanks for all the detail in those!), I eventually learned that this is because col and row lead to the FacetGrid layout mechanism. I understand and respect from the plotting docs that this behavior is deliberate in order to match seaborn. I also understand that I can force a specific figure size by explicitly passing a figsize argument.

In terms of API usage and discovery, however, I found it counterintuitive these keyword arguments would trigger changes in figure size, since they seem unrelated; I spent longer than I'd care to admit diagnosing why some of my plots were coming out different sizes.

Describe the solution you'd like

I had the idea that it could be nice if there were a global option (via xarray.set_option?) to adhere more strictly to matplotlib's RC parameters, rather than seaborn API conventions. That way, it would be easier for me to feel confident when autogenerating nice-looking figures that are subject to fixed size constraints.

I'm not sure if there are other style sheet parameters that would make sense to follow more strictly as well in a more general "publication mode"/"stylesheet mode", or if this parameter is an outlier.

Describe alternatives you've considered

For now, I've started explicitly passing in figsize=plt.rcParams['figure.figsize'] in many of my calls to .plot[.*]. That approach is working for now, but it feels like a workaround because it

• Adds boilerplate
• Requires careful attention to whether axis was also passed in, since they are mutually exclusive
• Seems to defeat the basic purpose of defining the parameter in stylesheets

Additional context

No response

[lurgyman63]

Looking back at this, sorry for the rambling and opinionated text!

I’d just like to add that if there’s a sense that this is desired, I’d be happy to try drafting a PR.

[dcherian]

I think the core issue is that the default figsize is too small for useful FacetGrids in many cases. Not sure what a good solution would be.