DateParseError on bar plot

[vfilimonov]

DateParseError is raised when Series with DateTimeIndex is plotted using bar-plot on axis that already contains another plot:

import pandas as pd
df = pd.Series([4,2,1,3,5], index=pd.to_datetime(['2004','2005','2006','2007','2008']))
df.plot()
df.plot(kind='bar')

Both of plots work independently, but being called one after another they result in DateParseError: day is out of range for month

If called in different order:

df.plot(kind='bar')
df.plot()

exception is not raised, but x-axis scale is incorrect

pandas version: 0.15.1
numpy verson: 1.9.1

[jreback]

cc @onesandzeroes
@TomAugspurger

@vfilimonov hmm, can you tell if this is coming from matplotlib or pandas? (e.g. I don't think pandas knows that it is multiple plotting).

[jorisvandenbossche]

It are series, so they are plotting on the current existing (so in this case the same) axis.

I think the reason for the error is (simplified) that they are plotted differently: df.plot() is plotted as a timeseries (which has custom axis formatting, index values are converted to periods if regular), while with df.plot(kind='label') it just plots the actual values in the index. And these both kinds are not compatible, and give rise to that error.

[vfilimonov]

Right, its difference in plotting, and xvalues for these plots are different:

df.plot()
print plt.gca().get_xlim()

results in (34.0, 38.0) and

df.plot(kind='bar')
print plt.gca().get_xlim()

results in (-0.5, 4.5)

[TomAugspurger]

@vfilimonov what version of matplotlib are you using? I don't get a ValueError, but the scale is off.

[TomAugspurger]

As far as a solution, I'm going to hopefully be cleaning up a bunch of DatetimeIndex plotting stuff soon.

[vfilimonov]

My matplotlib's version is 1.4.1

I get ValueError if I plot df.plot(kind='bar') after df.plot(). If I plot it in other way round (df.plot() after df.plot(kind='bar')), there's no error, but the scale is off

[wesm]

This is still a bit borked. The first example works correctly now, but the reverse order plotting is not correct (it is still a bar plot after the second df.plot())

[eipiminus1]

@wabu I can not reproduce what you are saying. While I don't get an error I don't see the expected behaviour, running the example provided above:

While there is no error, I expect to see both the line and the bar chart. Maybe there has been an improvement in the latest releases? I am running pandas 0.23.1 right now.

I had a look into the code and I assume the core problem is in line https://github.com/pandas-dev/pandas/blob/master/pandas/plotting/_core.py#L1197

So the x axis of a bar chart is just enumerated instead of keeping the actual scale. This is useful especially if the index of data is trivial to map to a real axis. It also simplifies the calculation of the position and width of the bars significantly. But it is annoying if you try to join the plot with a line plot. Also, this only works if the original index is equidistant (something you usually want in a bar chart).

The best I could come up with, to join a multicolumn DataFrame bar plot with a line plot is to rescale the line plot to match the enumeration of the bar chart.

df = pd.DataFrame({'a':[4,2,1,3,5], 'b':[4,5,2,3,1]}, index=pd.to_datetime(['2004','2005','2006','2007','2008']))
ax = df.plot(kind='bar')
ax.plot(np.arange(len(df)), df.a, 'g')

So I guess we need to handle data with a DatetimeIndex separately. I don't have the time (and expertise) to come up with a PR right now but I would be interested in your thoughts on how to tackle this, to make any effort easier.