Stacked histogram with datetime in matplotlib

Question

I am trying to create a stacked histogram with datetime objects, but I get the following error:

 TypeError: unorderable types: datetime.datetime() < float()

The code does work when I either convert the objects to timestamps or when I use one range of data (no stacking).

import datetime
import matplotlib.pyplot as plt

data = [[datetime.datetime(2015, 12, 24, 21, 13, 45), datetime.datetime(2015, 12, 30, 23, 37, 8), datetime.datetime(2015, 12, 30, 19, 43, 18), datetime.datetime(2015, 12, 30, 16, 14, 12), datetime.datetime(2015, 12, 30, 11, 32, 8), datetime.datetime(2015, 12, 29, 6, 29, 25), datetime.datetime(2015, 12, 28, 22, 54, 49), datetime.datetime(2015, 12, 28, 18, 41, 50), datetime.datetime(2015, 12, 28, 14, 25, 42), datetime.datetime(2015, 12, 28, 3, 1, 34), datetime.datetime(2015, 12, 27, 21, 10, 20), datetime.datetime(2015, 12, 27, 11, 29, 38), datetime.datetime(2015, 12, 26, 20, 34, 14), datetime.datetime(2015, 12, 26, 16, 58, 47), datetime.datetime(2015, 12, 26, 10, 54, 40), datetime.datetime(2015, 12, 25, 18, 17, 42), datetime.datetime(2015, 12, 24, 15, 44, 58), datetime.datetime(2015, 12, 25, 17, 25, 9), datetime.datetime(2015, 12, 25, 12, 33, 7), datetime.datetime(2015, 12, 30, 19, 26, 15), datetime.datetime(2015, 12, 30, 12, 46, 13), datetime.datetime(2015, 12, 30, 3, 38, 24), datetime.datetime(2015, 12, 25, 21, 11, 59), datetime.datetime(2015, 12, 25, 13, 30, 34), datetime.datetime(2015, 12, 24, 14, 6, 20)], [datetime.datetime(2015, 12, 28, 20, 59, 53), datetime.datetime(2015, 12, 27, 14, 3, 41), datetime.datetime(2015, 12, 26, 9, 37, 17)], [datetime.datetime(2015, 12, 29, 17, 18, 32)], [datetime.datetime(2015, 12, 29, 23, 15, 24)]]

fig, histograms = plt.subplots(5, 1, sharex=True, squeeze=False)
h = histograms[1][0]
h.hist(data, stacked=True)

This is the code without stacking:

import datetime
import matplotlib.pyplot as plt

data = [datetime.datetime(2015, 12, 24, 21, 13, 45), datetime.datetime(2015, 12, 30, 23, 37, 8), datetime.datetime(2015, 12, 30, 19, 43, 18), datetime.datetime(2015, 12, 30, 16, 14, 12), datetime.datetime(2015, 12, 30, 11, 32, 8), datetime.datetime(2015, 12, 29, 6, 29, 25), datetime.datetime(2015, 12, 28, 22, 54, 49), datetime.datetime(2015, 12, 28, 18, 41, 50), datetime.datetime(2015, 12, 28, 14, 25, 42), datetime.datetime(2015, 12, 28, 3, 1, 34), datetime.datetime(2015, 12, 27, 21, 10, 20), datetime.datetime(2015, 12, 27, 11, 29, 38), datetime.datetime(2015, 12, 26, 20, 34, 14), datetime.datetime(2015, 12, 26, 16, 58, 47), datetime.datetime(2015, 12, 26, 10, 54, 40), datetime.datetime(2015, 12, 25, 18, 17, 42), datetime.datetime(2015, 12, 24, 15, 44, 58), datetime.datetime(2015, 12, 25, 17, 25, 9), datetime.datetime(2015, 12, 25, 12, 33, 7), datetime.datetime(2015, 12, 30, 19, 26, 15), datetime.datetime(2015, 12, 30, 12, 46, 13), datetime.datetime(2015, 12, 30, 3, 38, 24), datetime.datetime(2015, 12, 25, 21, 11, 59), datetime.datetime(2015, 12, 25, 13, 30, 34), datetime.datetime(2015, 12, 24, 14, 6, 20), datetime.datetime(2015, 12, 28, 20, 59, 53), datetime.datetime(2015, 12, 27, 14, 3, 41), datetime.datetime(2015, 12, 26, 9, 37, 17), datetime.datetime(2015, 12, 29, 17, 18, 32), datetime.datetime(2015, 12, 29, 23, 15, 24)]

fig, histograms = plt.subplots(5, 1, sharex=True, squeeze=False)
h = histograms[1][0]
h.hist(data, stacked=True)

NOTE: As per answers, this is considered a bug. For future visitors I have filed a bug report https://github.com/matplotlib/matplotlib/issues/5898 in case you want to track progress

Answer 1

This is a bug, revealed by version 1.5.x supporting histograms of single series of datetime type data. Previous versions of matplotlib would not histogram datetime data whether stacked or not, failing with a similar error that said datetime could not be compared with float.

The Exception is thrown by this line of code. As you can see, that is called only when bin edges are not specified and is trying to find the minimum in the time series (comparing it with np.inf and taking the minimum of those). You can workaround this by specifying bin edges in the call, but then that leads to a further failure as the numpy histogram function called under the hood looks for less than zero width bins.

"Under the hood" when a single list of datetime.datetime objects is passed to the pyplot.hist() function, the data are actually converted to UNIX epoch timestamps (you can guess this from the labels of the x axis). This is not done when the input is a list of lists of datetime.datetime objects.

At that stage, I think we have to call it a bug and you will have to use timestamp as you have already discovered - e.g. h.hist([[t.timestamp() for t in s] for s in data], stacked=True). You can still give the bin labels in date format, even though the actual data being histogrammed are timestamps, thus this should be transparent to the user.

I'll have a look to see whether I can find a nicer workaround / fix and possibly raise an issue on the matplotlib github.

Code that works (matplotlib 1.5.1, Python 3), albeit a bit messy

import datetime
import matplotlib.pyplot as plt

data = [[datetime.datetime(2015, 12, 24, 21, 13, 45), datetime.datetime(2015, 12, 30, 23, 37, 8), datetime.datetime(2015, 12, 30, 19, 43, 18), datetime.datetime(2015, 12, 30, 16, 14, 12), datetime.datetime(2015, 12, 30, 11, 32, 8), datetime.datetime(2015, 12, 29, 6, 29, 25), datetime.datetime(2015, 12, 28, 22, 54, 49), datetime.datetime(2015, 12, 28, 18, 41, 50), datetime.datetime(2015, 12, 28, 14, 25, 42), datetime.datetime(2015, 12, 28, 3, 1, 34), datetime.datetime(2015, 12, 27, 21, 10, 20), datetime.datetime(2015, 12, 27, 11, 29, 38), datetime.datetime(2015, 12, 26, 20, 34, 14), datetime.datetime(2015, 12, 26, 16, 58, 47), datetime.datetime(2015, 12, 26, 10, 54, 40), datetime.datetime(2015, 12, 25, 18, 17, 42), datetime.datetime(2015, 12, 24, 15, 44, 58), datetime.datetime(2015, 12, 25, 17, 25, 9), datetime.datetime(2015, 12, 25, 12, 33, 7), datetime.datetime(2015, 12, 30, 19, 26, 15), datetime.datetime(2015, 12, 30, 12, 46, 13), datetime.datetime(2015, 12, 30, 3, 38, 24), datetime.datetime(2015, 12, 25, 21, 11, 59), datetime.datetime(2015, 12, 25, 13, 30, 34), datetime.datetime(2015, 12, 24, 14, 6, 20)], [datetime.datetime(2015, 12, 28, 20, 59, 53), datetime.datetime(2015, 12, 27, 14, 3, 41), datetime.datetime(2015, 12, 26, 9, 37, 17)], [datetime.datetime(2015, 12, 29, 17, 18, 32)], [datetime.datetime(2015, 12, 29, 23, 15, 24)]]

fig, histograms = plt.subplots(5, 1, sharex=True, squeeze=False)
h = histograms[1][0]

h.hist([[t.timestamp() for t in l] for l in data], stacked=True)

locs, labels = plt.xticks()
plt.xticks(locs,[datetime.datetime.fromtimestamp(t) for t in locs], rotation='vertical')
plt.gcf().subplots_adjust(bottom=0.4)
fig.set_size_inches(4, 15)

plt.show()

Produces