Grouped Bar-Chart with customized DateTime Index using pandas and Matplotlib

Question

I'd like to create a grouped bar chart that shows a customized Date-Time Index - just showing Month and year instead of the full dates. I want the bars to be grouped and not stacked.

I assumed pandas could handle this easily, using:

import pandas as pd
import matplotlib.pylab as plt
import matplotlib.dates as mdates

testdata = pd.DataFrame({"A": [1, 2, 3]
                       ,"B": [2, 3, 1]
                       , "C": [2, 3, 1]}  
                       ,index=pd.to_datetime(pd.DatetimeIndex(
                            data=["2019-03-02", "2019-04-01","2019-05-01"])))
ax = testdata.plot.bar()

This creates the plot that I want, I'd just like to change to date into something more simple, like March 2019, April 2019, May 2019.

I assumed using a Custom Date Formatter would work, therefore I tried

ax.xaxis.set_major_locator(mdates.MonthLocator())
ax.xaxis.set_major_formatter(mdates.DateFormatter('%b %Y'))

But than my labels are gone completely. And this question implies that pandas and the DateFormatter have a bit of a difficult relationship. Therefore I tried to do it with Matplotlib basics:

fig, ax = plt.subplots()
width = 0.8
ax.bar(testdata.index, testdata["A"]) 
ax.bar(testdata.index, testdata["B"])
ax.bar(testdata.index, testdata["C"])
ax.xaxis.set_major_locator(mdates.MonthLocator())
ax.xaxis.set_major_formatter(mdates.DateFormatter('%b %Y'))
plt.show()

Now the date representation is as expected (although the whitespace could be reduced), but the data overlap, which doesn't help.

Defining a width and subtracting it from the x values (as suggested normally) won't help due to the DateTime-Index I use. I get an error that subtracting DatetimeIndes and float is unsupported.

fig, ax = plt.subplots()
width = 0.8
ax.bar(testdata.index-width, testdata["A"]) 
ax.bar(testdata.index, testdata["B"])
ax.bar(testdata.index+width, testdata["C"])
ax.xaxis.set_major_locator(mdates.MonthLocator())
ax.xaxis.set_major_formatter(mdates.DateFormatter('%b %Y'))
plt.show()

So now I'm running out of ideas and hope for input

Answer 1

Pandas barplots are categorical. So maybe you're overthinking this and just want to use the string you want to see as a category label on the axis as index?

import pandas as pd
import matplotlib.pyplot as plt

df = pd.DataFrame({"A": [1, 2, 3]
                       ,"B": [2, 3, 1]
                       , "C": [2, 3, 1]}  
                       ,index=pd.to_datetime(pd.DatetimeIndex(
                            data=["2019-03-02", "2019-04-01","2019-05-01"])))

df.index = [d.strftime("%b %Y") for d in df.index]
ax = df.plot.bar()
plt.show()

Answer 2

The reason ax.xaxis.set_major_locator(mdates.MonthLocator()) fails because under the hood, pandas plots the bars against range(len(df)) , then rename the ticks accordingly.

You can grab the xticklabels after you plot, and reformat it:

ax = testdata.plot.bar()

ticks = [tick.get_text() for tick in ax.get_xticklabels()]
ticks = pd.to_datetime(ticks).strftime('%b %Y')
ax.set_xticklabels(ticks)

which gives the same result as ImpotanceOfBeingErnest's:

Another, probably better way is to shift the bars of each columns. This works better when you have many columns and want to reduce the number of xticks.

fig, ax = plt.subplots()

# define the shift
shift = pd.to_timedelta('1D')

# modify the base of each columns, can do with a for loop
ax.bar(testdata.index + shift, testdata["A"]) 
ax.bar(testdata.index, testdata["B"])
ax.bar(testdata.index - shift, testdata["C"])
ax.xaxis.set_major_locator(mdates.MonthLocator())
ax.xaxis.set_major_formatter(mdates.DateFormatter('%b %Y'))
plt.show()

Output: