Plot area legend rendered incorrectly when moved out of the plot

[jldiaz]

Code Sample, a copy-pastable example if possible

In area plots, the handles in the legend are made with filled rectangles, as for example in:

df = pd.DataFrame(np.random.rand(20, 5), columns=['A', 'B', 'C', 'D', 'E'])
df.plot(kind='area', linewidth=0.1)

However, if method legend() is used to alter the position of the legend, the handles are rendered as lines, which poses a problem if the line width is small, as in the following example:

df = pd.DataFrame(np.random.rand(20, 5), columns=['A', 'B', 'C', 'D', 'E'])
df.plot(kind='area', linewidth=0.1).legend(loc="center left", bbox_to_anchor=(1.02, 0.5))

output of pd.show_versions()

INSTALLED VERSIONS
------------------
commit: None
python: 3.5.1.final.0
python-bits: 64
OS: Linux
OS-release: 3.19.0-37-generic
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8

pandas: 0.17.1
nose: 1.3.7
pip: 8.0.2
setuptools: 19.6.2
Cython: None
numpy: 1.10.4
scipy: 0.16.1
statsmodels: None
IPython: 4.0.1
sphinx: None
patsy: None
dateutil: 2.4.2
pytz: 2015.7
blosc: None
bottleneck: None
tables: 3.2.2
numexpr: 2.4.6
matplotlib: 1.5.1
openpyxl: None
xlrd: 0.9.4
xlwt: None
xlsxwriter: None
lxml: None
bs4: 4.4.1
html5lib: 0.999
httplib2: None
apiclient: None
sqlalchemy: None
pymysql: None
psycopg2: None
Jinja2: None

[TomAugspurger]

IIRC the area plot legend is a bit of a hack. It's done here

Interested in seeing how much it would be to get this working?

[jldiaz]

Yes, I was already digging pandas' source and arrived at the conclusion that the function you pointed out is the one wihch builds the legend, at least when it is placed at its default location. However, for some reason which I don't understand (since I'm not familiar with pandas nor matplotlib internal architecture), that is not the one used when you use the method legend() to alter its position. For this case, apparently, the one used is the implemented in LinePlot class (which happens to be the base class of AreaPlot, btw)

I tried to study the code of BarPlot, which works fine also when the legend is positioned via legend() method, but I got lost, because BarPlot does not implement _add_legend_handle(), but uses the one in its base class MPLPlot.

Some hints about how to debug this issue?

[jldiaz]

I've found a workaround, just in case anyone has the same problem:

# Generate and plot the areas
df = pd.DataFrame(np.random.rand(20, 5), columns=['A', 'B', 'C', 'D', 'E'])
a = df.plot(kind='area', linewidth=0.1, alpha=0.6)

# Recreate the legend (reversing labels to preserve the stacked order of the areas)
# and put it at the right of the plot
import matplotlib.patches as mpatches
handles, labels = a.get_legend_handles_labels()
a.legend(reversed([mpatches.Rectangle((0, 0), 1, 1, fc=handle.get_color(), alpha=handle.get_alpha()) 
                   for handle in handles]), reversed(labels), 
         loc="center left", bbox_to_anchor=(1.02, 0.5))

[tacaswell]

What is the underlying artists that area uses?

[sinhrks]

@tacaswell it's PolyCollection (fill_between).

It may be a timing to remove the legend hack which @TomAugspurger pointed, showing a warning if mpl < 1.5.0 which supports PolyCollection legend.

[ResidentMario]

This issue should be closed (?) as it was addressed in the aforementioned pull.