Using pandas.TimeSeries_DateFormatter in bar plot?

[sinhrks]

xref #5774

When time series is set as index, line plot uses the TimeSeries_DateFormatter for recognizable date formatting. However, bar plot use the FixedFormatter which outputs the datetime as it is. Can bar plot also use the DateFormatter?

In [1]: from pandas import *
In [2]: from numpy.random import *
In [3]: ts = Series(randn(100), index=date_range('1/1/2000', periods=100))
In [4]: df = DataFrame(randn(100, 4), index=ts.index, columns=['A', 'B', 'C', 'D'])
In [5]: df = df.cumsum()

In [7]: a = df.plot(kind='bar')
In [9]: a.get_xaxis().get_major_formatter()
Out[9]: <matplotlib.ticker.FixedFormatter instance at 0x108700f38>

In [11]: a = df.plot()
In [12]: a.get_xaxis().get_major_formatter()
Out[12]: <pandas.tseries.converter.TimeSeries_DateFormatter instance at 0x10877ccf8>

[datapythonista]

Just in case it's useful, I found a bug that looks related to this issue to me. This code from an ipython notebook shows the problem:

%matplotlib inline
import numpy as np
import pandas as pd
import matplotlib
ts = pd.Series(np.random.randn(100), index=pd.date_range('1/1/2000', periods=100))
ax = ts.plot(kind='bar')
ax.xaxis.set_major_formatter(matplotlib.dates.DateFormatter('%m %d %Y'))

The first time it's executed it shows this warning, and the plot does not show:

~/.pydata/lib/python2.7/site-packages/IPython/core/formatters.py:239: FormatterWarning: Exception in image/png formatter: DateFormatter found a value of x=0, which is an illegal date. This usually occurs because you have not informed the axis that it is plotting dates, eg with ax.xaxis_date()
FormatterWarning,

Next executions the warning does not appear, neither the plot.

Some things that makes the plot appear:

• Removing the kind='bar'
• Removing the last line of code, where the formatter is set
• Changing the DateFormatter by another like the FormatStrFormatter
• Using plt.bar instead of pandas Series plot method

Hope this helps

[jorisvandenbossche]

Related with #8803

[ehsteve]

I've just run into this problem as well. Would very much appreciate a fix.

[tharibo]

Same problem here. Is there a workaround?

[pwaller]

Funny, I subscribed to this issue yesterday and someone replies to it! I too have had lots of problems with plot(kind="bar") with time series data. A lot of things are screwed up. The x-axis dates aren't shown correctly, ax.twinx() doesn't work properly. I've resorted to using plt.bar(df.index, df.values) with much greater success.

[pwaller]

Just hit this again. Can't figure out how to get the bar plot to show dates sensibly at all, it's showing a complete timestamp "%d/%m/%Y %H:%M:%S" for something which is a 2-month granularity. I just want it to show 04/2017, but the obvious ax.xaxis.set_major_formatter(plt.DateFormatter('%m/%Y')) does not work.

[pwaller]

Exception for reference:

/home/pwaller/.local/lib/python3.5/site-packages/matplotlib/dates.py in __call__(self, x, pos)
    460     def __call__(self, x, pos=0):
    461         if x == 0:
--> 462             raise ValueError('DateFormatter found a value of x=0, which is '
    463                              'an illegal date.  This usually occurs because '
    464                              'you have not informed the axis that it is '

ValueError: DateFormatter found a value of x=0, which is an illegal date.  This usually occurs because you have not informed the axis that it is plotting dates, e.g., with ax.xaxis_date()

[pwaller]

I found a workaround, by applying this at the end:

In this case, d is the DataFrame I'm plotting, and d.index is the DatetimeIndex.

plt.gca().xaxis.set_major_formatter(plt.FixedFormatter(d.index.to_series().dt.strftime("%b %Y")))

[shankari]

#1918 (comment) works!
Just FYI, if you have stored the axis for the plot already (as in #1918 (comment)), you can also do

ax.xaxis.set_major_formatter(plt.FixedFormatter(ts.index.to_series().dt.strftime("%d %b %Y")))

[pwaller]

I'm not a huge fan of the workaround and hit this quite frequently still. It also doesn't produce x-axis labels which are as good a data.plot() does, because the latter does Year labels on a separate row than Month labels, and I don't know how to achieve the same effect. I'd love for bar chars to be able to display time series data in the same way that line plots do, sometimes it is a less misleading way to present the data.