DataFrame.plot() produces incorrect legend labels when plotting multiple series on the same axis

[scottlawsonbc]

Code Sample, a copy-pastable example if possible

import pandas as pd
import matplotlib.pyplot as plt
df = pd.DataFrame(data=[[1, 1, 1, 1], [2, 2, 4, 8]], columns=['x', 'r', 'g', 'b'])
fig, ax = plt.subplots(nrows=1, ncols=3)
# Left plot
df.plot(x='x', y='r', linewidth=0, marker='o', color='r', ax=ax[0])
df.plot(x='x', y='g', linewidth=1, marker='x', color='g', ax=ax[0])
df.plot(x='x', y='b', linewidth=1, marker='o', color='b', ax=ax[0])
# Center plot
df.plot(x='x', y='b', linewidth=1, marker='o', color='b', ax=ax[1])
df.plot(x='x', y='r', linewidth=0, marker='o', color='r', ax=ax[1])
df.plot(x='x', y='g', linewidth=1, marker='x', color='g', ax=ax[1])
# Right plot
df.plot(x='x', y='g', linewidth=1, marker='x', color='g', ax=ax[2])
df.plot(x='x', y='b', linewidth=1, marker='o', color='b', ax=ax[2])
df.plot(x='x', y='r', linewidth=0, marker='o', color='r', ax=ax[2])
# Produces correct output when uncommented:
# ax[0].legend()
# ax[1].legend()
# ax[2].legend()
plt.show()

Problem description

Note: Possibly related to #14958, #17939, #14563, however this issue discusses how the behaviour depends on the order in which df.plot() is called

• In certain situations, df.plot() may generate incorrect legend labels (see example)
• Incorrect legend labels may appear when df.plot() plots multiple series on the same axis
• Plotting only one series per axis always appears to produce the correct legend label
• When multiple series are plotted on the same axis:
  • Only the last df.plot() call will produce the correct legend label
  • Calling ax.legend() after the final df.plot() will produce the correct legend entry
• Calling df.plot() with markerstyle='' always appears produces the correct legend label for that series regardless of the order in which df.plot() is called, or the number of other series on the same axis

Expected Output

This is the incorrect output produced by the sample code
Note that in each plot, only the last series to be plotted has a correct legend label

This is the expected output

Output of pd.show_versions()

[paste the output of pd.show_versions() here below this line]
INSTALLED VERSIONS

commit: None
python: 3.5.4.final.0
python-bits: 64
OS: Linux
OS-release: 4.10.0-37-generic
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: None
LANG: en_US.UTF-8
LOCALE: en_US.UTF-8

pandas: 0.21.0
pytest: None
pip: 9.0.1
setuptools: 36.5.0.post20170921
Cython: None
numpy: 1.13.3
scipy: 0.19.1
pyarrow: None
xarray: None
IPython: 6.1.0
sphinx: None
patsy: None
dateutil: 2.6.1
pytz: 2017.3
blosc: None
bottleneck: None
tables: None
numexpr: None
feather: None
matplotlib: 2.1.0
openpyxl: None
xlrd: None
xlwt: None
xlsxwriter: None
lxml: None
bs4: None
html5lib: None
sqlalchemy: None
pymysql: None
psycopg2: None
jinja2: None
s3fs: None
fastparquet: None
pandas_gbq: None
pandas_datareader: None

[gfyoung]

cc @TomAugspurger

@scottlawsonbc : Could you provide snapshots of what you're seeing currently?

[scottlawsonbc]

@gfyoung there are two screenshots under the expected output section

[gfyoung]

there are two screenshots under the expected output section

Yes, I see that, but which, if any, represent what you are seeing currently? Or are both snapshots representative of what you would expect to see?

[scottlawsonbc]

@gfyoung I understand your question now. I see how the plots are confusing and I should have explained their meaning in a way that is more clear.

At the bottom of my example code, I commented out several lines:

# Produces correct output when uncommented:
# ax[0].legend()
# ax[1].legend()
# ax[2].legend()

When these lines are commented out, the sample code produces a plot with incorrect legend labels. If you run the sample code exactly as I provided in the original issue, then it will produce the erroneous incorrect output.

When these lines are uncommented, the sample code produces a plot with the correct legend labels. However, the issue is that pandas should be producing this output without requiring these lines to be run.

In the expected output section of my issue, the first plot shows the output which is incorrect, and the second plot shows what the correct output should be (intended as a reference).

Does this answer your question? I will update my issue to make this more clear.

[gfyoung]

Ah, okay. That makes sense. If you could update the issue to reflect your answer, that would be great.

[TomAugspurger]

@scottlawsonbc do you have time to debug where things are going wrong?

pandas/pandas/plotting/_core.py

Line 495 in 7f4c960

def _make_legend(self):

pandas/pandas/plotting/_core.py

Line 384 in 7f4c960

def _post_plot_logic_common(self, ax, data):

may be good starting points (or maybe not, haven't touched that code in a while)

[timisid]

if you still using multiple series with one axis (like barplot with x,y, and hue options) and the legend still incorect, you can modified the legend using get_texts() after df.plot().
in this case df.plots() going to change it with sns.barplot()

example:
first line ; sns.barplot(data=data, x=..., y=..., hue=..., ax=ax)
next line :
L=ax.legend()
L.set_title('modified legend')
L.get_texts()[0].set_text('b')
L.get_texts()[1].set_text('r')
L.get_texts()[2].set_text('g')

[comready]

If you need another example of this problem let me know

[MarcoGorelli]

Hi @jzlcdh - if you're aware of a new bug then please open a new issue, thanks