pandas plotting backend?

Plotly has toplevel .plot function which allows for a pandas plotting backend to exist:

https://github.com/plotly/plotly.py/blob/4363c51448cda178463277ff3c12becf35dbd3b8/packages/python/plotly/plotly/__init__.py

Like this, if people have plotly installed, they can do:

pd.set_option('plotting.backend', 'plotly')

and then df.plot.line(x=x, y=y) will defer to plotly.express.line(data_frame=df, x=x, y=y):

It’d be nice to be able to do

pd.set_option('plotting.backend', 'seaborn')

and then have df.plot.line(x=x, y=y) defer to seaborn.line(data=df, x=x, y=y)

Would you be open to these ~150 lines of code or so to allow seaborn to be set as a plotting backend in pandas? Check the link above to see what it looks like in plotly. I’d be happy to implement this, just checking if it’d be welcome

[nicolaskruchten]

In fairness, though, the plotly layer is light in part because a lot of stuff just doesn’t work.

Oh yes, the plotly backend deliberately does not aim to be arg-compatible with the builtin backend! This is what I put in the docs:

[mwaskom]

I wouldn’t ask seaborn to add any new plots (like pie) - if it’s not in seaborn, I’d argue that it probably shouldn’t be in pandas either

That’s tricky because sometimes things aren’t in seaborn because they’re in pandas, e.g. “area plots” https://github.com/mwaskom/seaborn/issues/2410. In cases where a “seaborn version” of a plot type wouldn’t add anything beyond what an existing pandas (or matplotlib) implementation can do, I’m usually reluctant to add it, especially if it doesn’t fit nicely into seaborn’s existing plot taxonomy. It’s hard to see a case for pandas fully deprecating area and (alas) pie charts — they’re fundamental for the kinds of applications that many people use pandas for.

The layer in seaborn should really be very light, just like it is in plotly

In fairness, though, the plotly layer is light in part because a lot of stuff just doesn’t work. e.g. for a few quick things i tried, both wide data and subplots raise. And in the example you linked to above, the by and bins (bins!!) parameters for histograms are simply ignored, as with s and c for scatter. I suppose it’s fine for the plotly backend to not be feature-complete with the matplotlib backend, but you’d be talking about deprecation of major functionality if you took the same approach and didn’t include anything in the “seaborn” backend that wasn’t a simple parameter name translation.