theme_xkcd() + save(..., dpi=int) -> AttributeError

[TyberiusPrime]

Minimal Example

import pandas as pd
import plotnine
df = pd.DataFrame(patsy.demo_data("a",'b','x','y', 'z'))
p = plotnine.ggplot(df)
p += plotnine.geom_point(plotnine.aes('x','y'))
p += plotnine.facet_wrap('b')
p += plotnine.theme_xkcd()
p.save('test.png') # succeeds
p.save('test2.png', dpi=72) # fails

fails with the following stack trace:

/martha/ffs/e/plotnine_exploration/code3/plotnine/plotnine/ggplot.py in save(self, filename, format, path, width, height, units, dpi, limitsize, verbose, **kwargs)
    726         except Exception as err:
    727             figure[0] and plt.close(figure[0])
--> 728             raise err
    729         else:
    730             figure[0] and plt.close(figure[0])

/martha/ffs/e/plotnine_exploration/code3/plotnine/plotnine/ggplot.py in save(self, filename, format, path, width, height, units, dpi, limitsize, verbose, **kwargs)
    723 
    724         try:
--> 725             _save()
    726         except Exception as err:
    727             figure[0] and plt.close(figure[0])

/martha/ffs/e/plotnine_exploration/code3/plotnine/plotnine/ggplot.py in _save()
    720                 fig_kwargs['frameon'] = True
    721 
--> 722             fig.savefig(filename, **fig_kwargs)
    723 
    724         try:

/usr/local/lib/python3.6/dist-packages/matplotlib/figure.py in savefig(self, fname, **kwargs)
   2033             self.set_frameon(frameon)
   2034 
-> 2035         self.canvas.print_figure(fname, **kwargs)
   2036 
   2037         if frameon:

/usr/local/lib/python3.6/dist-packages/matplotlib/backend_bases.py in print_figure(self, filename, dpi, facecolor, edgecolor, orientation, format, **kwargs)
   2210                     orientation=orientation,
   2211                     dryrun=True,
-> 2212                     **kwargs)
   2213                 renderer = self.figure._cachedRenderer
   2214                 bbox_inches = self.figure.get_tightbbox(renderer)

/usr/local/lib/python3.6/dist-packages/matplotlib/backends/backend_agg.py in print_png(self, filename_or_obj, *args, **kwargs)
    511 
    512     def print_png(self, filename_or_obj, *args, **kwargs):
--> 513         FigureCanvasAgg.draw(self)
    514         renderer = self.get_renderer()
    515         original_dpi = renderer.dpi

/usr/local/lib/python3.6/dist-packages/matplotlib/backends/backend_agg.py in draw(self)
    431             # if toolbar:
    432             #     toolbar.set_cursor(cursors.WAIT)
--> 433             self.figure.draw(self.renderer)
    434             # A GUI class may be need to update a window using this draw, so
    435             # don't forget to call the superclass.

/usr/local/lib/python3.6/dist-packages/matplotlib/artist.py in draw_wrapper(artist, renderer, *args, **kwargs)
     53                 renderer.start_filter()
     54 
---> 55             return draw(artist, renderer, *args, **kwargs)
     56         finally:
     57             if artist.get_agg_filter() is not None:

/usr/local/lib/python3.6/dist-packages/matplotlib/figure.py in draw(self, renderer)
   1470 
   1471             if self.frameon:
-> 1472                 self.patch.draw(renderer)
   1473 
   1474             mimage._draw_list_compositing_images(

/usr/local/lib/python3.6/dist-packages/matplotlib/artist.py in draw_wrapper(artist, renderer, *args, **kwargs)
     53                 renderer.start_filter()
     54 
---> 55             return draw(artist, renderer, *args, **kwargs)
     56         finally:
     57             if artist.get_agg_filter() is not None:

/usr/local/lib/python3.6/dist-packages/matplotlib/patches.py in draw(self, renderer)
    556             renderer = PathEffectRenderer(self.get_path_effects(), renderer)
    557 
--> 558         renderer.draw_path(gc, tpath, affine, rgbFace)
    559 
    560         gc.restore()

/usr/local/lib/python3.6/dist-packages/matplotlib/patheffects.py in draw_path(self, gc, tpath, affine, rgbFace)
    108         for path_effect in self._path_effects:
    109             path_effect.draw_path(self._renderer, gc, tpath, affine,
--> 110                                   rgbFace)
    111 
    112     def draw_markers(self, gc, marker_path, marker_trans, path, *args,

/usr/local/lib/python3.6/dist-packages/matplotlib/patheffects.py in draw_path(self, renderer, gc, tpath, affine, rgbFace)
    209     """
    210     def draw_path(self, renderer, gc, tpath, affine, rgbFace):
--> 211         Stroke.draw_path(self, renderer, gc, tpath, affine, rgbFace)
    212         renderer.draw_path(gc, tpath, affine, rgbFace)
    213 

/usr/local/lib/python3.6/dist-packages/matplotlib/patheffects.py in draw_path(self, renderer, gc, tpath, affine, rgbFace)
    196         gc0.copy_properties(gc)
    197 
--> 198         gc0 = self._update_gc(gc0, self._gc)
    199         trans = self._offset_transform(renderer, affine)
    200         renderer.draw_path(gc0, tpath, trans, rgbFace)

AttributeError: 'withStroke' object has no attribute '_gc'

[TyberiusPrime]

Changing plotnine.options.dpi works as expected though.

Debugging leads me to line 701 of ggplot.py where
self.theme = self.theme + theme(dpi=dpi) triggers the bug,
but
self.theme += theme(dpi=dpi)
does not.
So it's so likely to be somewhere in the theme deep copy?

[TyberiusPrime]

hm, rcParams claims the following:

       try:
            rcParams = deepcopy(self._rcParams)
        except NotImplementedError:
            # deepcopy raises an error for objects that are drived from or
            # composed of matplotlib.transform.TransformNode.
            # Not desirable, but probably requires upstream fix.
            # In particular, XKCD uses matplotlib.patheffects.withStrok
            rcParams = copy(self._rcParams)
``` - so perhaps the upstream fix was not 100%?

[has2k1]

Thanks for the debugging, but I think this is an upstream quirk. That means in Matplotlib, if you have an xkcd plot and you savefig(dpi=123), you should run into the same issue.

[TyberiusPrime]

I agree that it's an upstream quirk, but your intuition is mistaken -
I can plot xkcd-plots at different DPIs in either matplotlib or plotnine,
but I must not use the dpi argument on save().

Let me illustrate,
first maplotlib

import matplotlib
import pandas as pd
x = [1,2,3]
y = [4,5,6]
from matplotlib import pyplot as plt
plt.xkcd()

fig = plt.figure()
plt.plot(x,y)
plt.savefig("test.png",dpi=12)

leads to this (very tiny, 12 dpi) image:

Now, same thing in plotnine which works:

import pandas as pd
import plotnine
df = pd.DataFrame({'x': [1,2,3], 'y': [4,5,6], 'b': ['a','a','b']})
p = plotnine.ggplot(df)
p += plotnine.geom_point(plotnine.aes('x','y'))
p += plotnine.facet_wrap('b')
p += plotnine.theme_xkcd() + plotnine.theme(dpi=24)
p.save('test2.png')

And the one that fails

import pandas as pd
import plotnine
df = pd.DataFrame({'x': [1,2,3], 'y': [4,5,6], 'b': ['a','a','b']})
p = plotnine.ggplot(df)
p += plotnine.geom_point(plotnine.aes('x','y'))
p += plotnine.facet_wrap('b')
p += plotnine.theme_xkcd() 
p.save('test3.png', dpi=24)

Notice how the difference is only where I specify the dpi!
It's also noteworthy that adding the theme right away (p += (theme_xkcd() + theme(dpi=12))
works, but doing it in save() fails:

import pandas as pd
import plotnine
df = pd.DataFrame({'x': [1,2,3], 'y': [4,5,6], 'b': ['a','a','b']})
p = plotnine.ggplot(df)
p += plotnine.geom_point(plotnine.aes('x','y'))
p += plotnine.facet_wrap('b')
p += plotnine.theme_xkcd()
p.theme = p.theme + plotnine.theme(dpi=24)
p.save('test3.png')

is ok, but doing essentiale the same thing:
self.theme = self.theme + theme(dpi=dpi)
in line 701 of ggplot.py leads to the exception.

This is because the self in save() is a copy - done in line 671.
This can be seen by this example:

import pandas as pd
import plotnine
from copy import copy
df = pd.DataFrame({'x': [1,2,3], 'y': [4,5,6], 'b': ['a','a','b']})
p = plotnine.ggplot(df)
p += plotnine.geom_point(plotnine.aes('x','y'))
p += plotnine.facet_wrap('b')
p += plotnine.theme_xkcd()
p = copy(p) # the magic line!
p.theme = p.theme + plotnine.theme(dpi=24)
p.save('test4.png')

Which triggers the same exception.

The problem disappears under two things: Either
by replacing line 701 with the in place modifying one,
or by doing a deepcopy in line 671 instead of a copy.