df.plot produces wrong result when x vals are unsorted. (Dupe #2726)

[akhmerov]

Could be related to #2923. The following code produces two different plots for me in pandas 10.1.

import pandas
from matplotlib import pyplot
from numpy.random import randn

data = pandas.DataFrame(randn(6, 2))
pyplot.plot(data[0], data[1])
data.plot(0, 1)

One can actually implement a test for this by comparing the collections produced by the plot.

Does this work for you?

import matplotlib.pyplot as plt
import pandas as pd

data = pandas.DataFrame(randn(6, 2))
for i in range(len(data.columns)):
    data[i].plot()
plt.show()

The color cycle issue in v0.10.1 was fixed by #2836 for the upcoming v0.11:

[akhmerov]

That works. I'm sorry I wasn't clear enough. The issue here isn't about the color cycle. The thing is that the two plots that I make in the example are supposed to be the same, or at least go through the same points. So one should see either one line, or two lines with the same vertices.

Is the issue that the two plots dont appear in the same view, or that the two lines don't coincide,
or that when you try the former, the latter doesn't happen?

In any case, I can't seem to reproduce with git master:

[akhmerov]

I confirm: the issue does not appear in master. Can you double-check with 0.10.1? This is a severe bug, since it shows completely wrong output without any error messages or warnings, and unless 0.11 is going to be released soon, I think, backporting it would be important.

Pandas doesn't currently maintain multiple release branches - limited resources.
But release cycles are also comparatively short compared to other projects, usually
a few months between.
I think master gets commits quite responsibly, and regression testing is always part
of the process, so it's also usually pretty stable.

I can confirm the plots look different in v0.10.1. will look into it.

Ok, It's a sorted order thing:

data = pd.DataFrame([[0,0],[2,2],[1,1]])
data.plot(0, 1)
pyplot.plot(data[0], data[1])

v0.10.1:

v0.11:

fixed in #2726 (dupe). closing now.

[akhmerov]

Would adding a test for this particular issue make sense?

Always glad to have more tests and new contributors. please!