bug fix: hue in sc.pl.violin

[AnnaChristina]

closses #1174

hue key is now added to keys if hue is in kwds. before obs_tidy did not include this keyword, so this resulted in an ValueError

an alternative would be to add hue as a function parameter to sc.pl.violin

[codecov[bot]]

Codecov Report

Merging #1822 (62e48bc) into master (c488909) will increase coverage by 0.01%. The diff coverage is 50.00%.

@@            Coverage Diff             @@
##           master    #1822      +/-   ##
==========================================
+ Coverage   71.18%   71.20%   +0.01%     
==========================================
  Files          92       92              
  Lines       11190    11192       +2     
==========================================
+ Hits         7966     7969       +3     
+ Misses       3224     3223       -1     

Impacted Files	Coverage Δ
scanpy/plotting/_anndata.py	84.39% <50.00%> (-0.08%)	:arrow_down:
scanpy/logging.py	98.38% <0.00%> (+1.61%)	:arrow_up:

[ivirshup]

Thanks for the PR.

I think if this is added it should be added as a function parameter. It should also get some tests, I'm a bit concerned how just passing hue will work with the groupby and palette parameters.

[AnnaChristina]

would it then make generally sense to add an optional hue parameter in seaborn based plotting functions? was checking if hue is anywhere used but does not seem to be the case.

will check behaviour for this wrt groupby and palette.

[ivirshup]

would it then make generally sense to add an optional hue parameter in seaborn based plotting functions? was checking if hue is anywhere used but does not seem to be the case.

I'm not sure. I don't know off the top of my head how well we can make current functions work with hue, since seaborn uses it to group observations in many cases.

[ivirshup]

Looking at this again, my current reading is that

sc.pl.violin(adata, keys=keys, groupby=groupby, hue=hue)

should translate into something like:

df = sc.get.obs_df(adata, [groupby, hue] + keys)
for k in keys:
    sns.violinplot(data=df, x=groupby, y=k, hue=hue)

sns.catplot could also be a good model, but I don't think it supports the same range of attributes.

While also passing through the right color palette to use for hue.

• We would need to address where the legend goes, since I believe this would currently repeat the same legend for each plot
• palette handling, which I believe is currently resulting in a KeyError
• Axis labeling, we would probably want it per groupby, and leave hue to the legend.