Using `seaborn.swarmplot` with proplot axes

[spencerkclark]

Description

Using seaborn.swarmplot with proplot axes currently leads to an error. I realize that the documentation states that most seaborn plotting methods should work with proplot axes and therefore we should not expect all of them to work, but I'm mainly curious if there might be a simple workaround.

Proplot is great by the way -- thanks for the effort to create/maintain it.

Steps to reproduce

A "Minimal, Complete and Verifiable Example" will make it much easier for maintainers to help you.

import proplot as pplt
import seaborn as sns

tips = sns.load_dataset("tips")
fig = pplt.figure()
ax = fig.subplot()
sns.swarmplot(ax=ax, x="day", y="total_bill", data=tips)

Expected behavior: [What you expected to happen]

I would expect this to produce a plot similar to the one produced when using just seaborn and matplotlib axes.

Actual behavior: [What actually happened]

Instead the following error is raised in addition to some warnings:

/home/spencerc/miniconda3/envs/fv3net/lib/python3.7/site-packages/seaborn/categorical.py:1376: ProPlotWarning: Failed to restrict automatic colormap normalization algorithm to in-bounds data only. Error message: boolean index did not match indexed array along dimension 1; dimension is 3 but corresponding boolean dimension is 62
  points = ax.scatter(cat_pos, swarm_data, s=s, **kws)
/home/spencerc/miniconda3/envs/fv3net/lib/python3.7/site-packages/seaborn/categorical.py:1376: ProPlotWarning: Failed to restrict automatic colormap normalization algorithm to in-bounds data only. Error message: boolean index did not match indexed array along dimension 1; dimension is 3 but corresponding boolean dimension is 19
  points = ax.scatter(cat_pos, swarm_data, s=s, **kws)
/home/spencerc/miniconda3/envs/fv3net/lib/python3.7/site-packages/seaborn/categorical.py:1376: ProPlotWarning: Failed to restrict automatic colormap normalization algorithm to in-bounds data only. Error message: boolean index did not match indexed array along dimension 1; dimension is 3 but corresponding boolean dimension is 87
  points = ax.scatter(cat_pos, swarm_data, s=s, **kws)
/home/spencerc/miniconda3/envs/fv3net/lib/python3.7/site-packages/seaborn/categorical.py:1376: ProPlotWarning: Failed to restrict automatic colormap normalization algorithm to in-bounds data only. Error message: boolean index did not match indexed array along dimension 1; dimension is 3 but corresponding boolean dimension is 76
  points = ax.scatter(cat_pos, swarm_data, s=s, **kws)
---------------------------------------------------------------------------
AttributeError                            Traceback (most recent call last)
<ipython-input-22-5939fc904540> in <module>
      1 fig = proplot.figure()
      2 ax = fig.subplot()
----> 3 sns.swarmplot(ax=ax, x="day", y="total_bill", data=tips)

~/miniconda3/envs/fv3net/lib/python3.7/site-packages/seaborn/_decorators.py in inner_f(*args, **kwargs)
     44             )
     45         kwargs.update({k: arg for k, arg in zip(sig.parameters, args)})
---> 46         return f(**kwargs)
     47     return inner_f
     48 

~/miniconda3/envs/fv3net/lib/python3.7/site-packages/seaborn/categorical.py in swarmplot(x, y, hue, data, order, hue_order, dodge, orient, color, palette, size, edgecolor, linewidth, ax, **kwargs)
   3022 
   3023     This function is similar to :func:`stripplot`, but the points are adjusted
-> 3024     (only along the categorical axis) so that they don't overlap. This gives a
   3025     better representation of the distribution of values, but it does not scale
   3026     well to large numbers of observations. This style of plot is sometimes

~/miniconda3/envs/fv3net/lib/python3.7/site-packages/seaborn/categorical.py in plot(self, ax, kws)
   1426 
   1427 class _CategoricalStatPlotter(_CategoricalPlotter):
-> 1428 
   1429     require_numeric = True
   1430 

~/miniconda3/envs/fv3net/lib/python3.7/site-packages/seaborn/categorical.py in draw_swarmplot(self, ax, kws)
   1419         """Make the full plot."""
   1420         self.draw_swarmplot(ax, kws)
-> 1421         self.add_legend_data(ax)
   1422         self.annotate_axes(ax)
   1423         if self.orient == "h":

~/miniconda3/envs/fv3net/lib/python3.7/site-packages/seaborn/categorical.py in swarm_points(self, ax, points, center, width, s, **kws)
   1309         # We'll figure out the swarm positions in the latter
   1310         # and then convert back to data coordinates and replot
-> 1311         orig_xy = ax.transData.transform(points.get_offsets())
   1312 
   1313         # Order the variables so that x is the categorical axis

AttributeError: 'list' object has no attribute 'get_offsets'

Equivalent steps in matplotlib

Please make sure this bug is related to a specific proplot feature. If you're not sure, try to replicate it with the native matplotlib API. Matplotlib bugs belong on the matplotlib github page.

import seaborn as sns

tips = sns.load_dataset("tips")
sns.swarmplot(x="day", y="total_bill", data=tips)

Proplot version

Paste the results of import matplotlib; print(matplotlib.__version__); import proplot; print(proplot.version)here.

3.2.2
0.8.1

[lukelbd]

This one is tricky... proplot iterates over 2D columns of input data for all "1D" plotting commands (bar, scatter, area, line), returns lists of artists if there is >1 column, and uses the property cycle for each column. By contrast, matplotlib flattens 2D input to scatter and returns a single PathCollection. It appears this seaborn example is relying on that matplotlib behavior. Not sure if this can be fixed... iterating over columns is a feature I really want to keep. But maybe there's a workaround.

I'd like to support more seaborn functions but it's very preliminary, my note about seaborn in the docs was definitely optimistic. The next documentation update will be less optimistic.

P.S. In recent matplotlib versions I noticed importing seaborn emits a bunch of warnings due to overwriting existing colormaps. I've added a quick patch to correct this.

[lukelbd]

Actually this is simpler than I thought. Seaborn was not passing 2D input arrays -- instead it was passing an Nx3 RGB style color array, and proplot was interpreting that as a 2D array of scalar color data instead of explicit RGB tuples. Now when Nx3 or Nx4 color data is passed scatter it is properly interpreted as RGB[A] tuples. Will push the fix to master soon.

[spencerkclark]

Cool, thanks for investigating and finding a fix!

[spencerkclark]

Sorry to bother you again. This is perhaps a separate issue, but still something I ran into when using proplot with seaborn.swarmplot (after your nice update in dd2e03e). It's a bit more of a subtle issue. I find that -- specifically when using the "default" matplotlib style and a particular configuration of swarmplot -- I get an error when seaborn tries to create a legend.

Here's an example using the "tips" dataset from before:

import proplot as pplt
import seaborn as sns

# This line appears to be key in eventually triggering the error.
pplt.config.use_style("default")

tips = sns.load_dataset("tips")
fig = pplt.figure()
ax = fig.subplot()

# This call works (even with pplt.config.use_style("default"))
sns.swarmplot(ax=ax, x="day", y="total_bill", data=tips)

# This call does not work with pplt.config.use_style("default") but does without it
sns.swarmplot(ax=ax, x="day", y="tip", hue="smoker", data=tips, dodge=True)

The traceback in the error case looks like this:

---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-61-114426113da6> in <module>
      1 fig = pplt.figure()
      2 ax = fig.subplot()
----> 3 sns.swarmplot(ax=ax, x="day", y="tip", hue="smoker", data=tips, dodge=True)

~/Software/miniconda3/envs/xarray-tests/lib/python3.7/site-packages/seaborn/categorical.py in swarmplot(x, y, hue, data, order, hue_order, dodge, orient, color, palette, size, edgecolor, linewidth, ax, **kwargs)
   2989                        linewidth=linewidth))
   2990 
-> 2991     plotter.plot(ax, kwargs)
   2992     return ax
   2993 

~/Software/miniconda3/envs/xarray-tests/lib/python3.7/site-packages/seaborn/categorical.py in plot(self, ax, kws)
   1446         self.draw_swarmplot(ax, kws)
   1447         self.add_legend_data(ax)
-> 1448         self.annotate_axes(ax)
   1449         if self.orient == "h":
   1450             ax.invert_yaxis()

~/Software/miniconda3/envs/xarray-tests/lib/python3.7/site-packages/seaborn/categorical.py in annotate_axes(self, ax)
    414 
    415         if self.hue_names is not None:
--> 416             leg = ax.legend(loc="best")
    417             if self.hue_title is not None:
    418                 leg.set_title(self.hue_title)

~/Software/miniconda3/envs/xarray-tests/lib/python3.7/site-packages/proplot/axes/base.py in legend(self, handles, labels, loc, location, queue, **kwargs)
   2908             self._register_guide('legend', (handles, labels), loc, **kwargs)
   2909         else:
-> 2910             leg = self._draw_legend(handles, labels, loc=loc, **kwargs)
   2911             return leg
   2912 

~/Software/miniconda3/envs/xarray-tests/lib/python3.7/site-packages/proplot/axes/base.py in _draw_legend(self, handles, labels, loc, width, pad, space, align, frame, frameon, ncol, ncols, alphabetize, center, order, label, title, fontsize, fontweight, fontcolor, titlefontsize, titlefontweight, titlefontcolor, handle_kw, handler_map, **kwargs)
   2814             kwargs.update({key: kw_frame.pop(key) for key in ('shadow', 'fancybox')})
   2815             objs = [lax._parse_ordinary_legend(pairs, ncol=ncol, order=order, **kwargs)]
-> 2816             objs[0].legendPatch.update(kw_frame)
   2817         for obj in objs:
   2818             if hasattr(lax, 'legend_') and lax.legend_ is None:

~/Software/miniconda3/envs/xarray-tests/lib/python3.7/site-packages/matplotlib/artist.py in update(self, props)
   1062                         raise AttributeError(f"{type(self).__name__!r} object "
   1063                                              f"has no property {k!r}")
-> 1064                     ret.append(func(v))
   1065         if ret:
   1066             self.pchanged()

~/Software/miniconda3/envs/xarray-tests/lib/python3.7/site-packages/matplotlib/patches.py in set_facecolor(self, color)
    366         """
    367         self._original_facecolor = color
--> 368         self._set_facecolor(color)
    369 
    370     def set_color(self, c):

~/Software/miniconda3/envs/xarray-tests/lib/python3.7/site-packages/matplotlib/patches.py in _set_facecolor(self, color)
    354             color = mpl.rcParams['patch.facecolor']
    355         alpha = self._alpha if self._fill else 0
--> 356         self._facecolor = colors.to_rgba(color, alpha)
    357         self.stale = True
    358 

~/Software/miniconda3/envs/xarray-tests/lib/python3.7/site-packages/matplotlib/colors.py in to_rgba(c, alpha)
    204         rgba = None
    205     if rgba is None:  # Suppress exception chaining of cache lookup failure.
--> 206         rgba = _to_rgba_no_colorcycle(c, alpha)
    207         try:
    208             _colors_full_map.cache[c, alpha] = rgba

~/Software/miniconda3/envs/xarray-tests/lib/python3.7/site-packages/matplotlib/colors.py in _to_rgba_no_colorcycle(c, alpha)
    275                     f"Value must be within 0-1 range")
    276             return c, c, c, alpha if alpha is not None else 1.
--> 277         raise ValueError(f"Invalid RGBA argument: {orig_c!r}")
    278     # turn 2-D array into 1-D array
    279     if isinstance(c, np.ndarray):

ValueError: Invalid RGBA argument: 'inherit'