Pandas: BUG: register_matplotlib_converters leads to wrong datetime interpretation with matplotlib 3.3

[ZahlGraf]

• [x] I have checked that this issue has not already been reported.

• [x] I have confirmed this bug exists on the latest version of pandas.

• [ ] (optional) I have confirmed this bug exists on the master branch of pandas.

---

Code Sample, a copy-pastable example

import datetime
import matplotlib
import matplotlib.pyplot as plt
from matplotlib.dates import DateFormatter

from pandas.plotting import register_matplotlib_converters
register_matplotlib_converters()

t = [
    datetime.datetime(2022, 12, 31, 0, 0), 
    datetime.datetime(2022, 11, 30, 0, 0), 
    datetime.datetime(2022, 10, 31, 0, 0),
]
s = [0.0, 0.1, 0.2]

fig, ax = plt.subplots()
ax.plot_date(x=t, y=s)
ax.xaxis.set_major_formatter(DateFormatter("%b '%y"))

plt.xticks(rotation=90)
plt.show()

Problem description

The years in the plot are now 91 and 92.

Note: With matplotlib 3.2.2 (the last stable release) the error is not there. So I guess this is a bug due to an api change inside matplotlib 3.3.0.

Expected Output

I expects the year 22 and 23 like in this image:

The second image can be produced by deleting the call to register_matplotlib_converters()

Output of pd.show_versions()

INSTALLED VERSIONS
------------------
commit           : None
python           : 3.7.6.final.0
python-bits      : 64
OS               : Windows
OS-release       : 10
machine          : AMD64
processor        : Intel64 Family 6 Model 78 Stepping 3, GenuineIntel
byteorder        : little
LC_ALL           : None
LANG             : None
LOCALE           : None.None

pandas           : 1.0.5
numpy            : 1.19.0
pytz             : 2020.1
dateutil         : 2.8.1
pip              : 19.2.3
setuptools       : 41.2.0
Cython           : None
pytest           : None
hypothesis       : None
sphinx           : None
blosc            : None
feather          : None
xlsxwriter       : None
lxml.etree       : None
html5lib         : None
pymysql          : None
psycopg2         : None
jinja2           : None
IPython          : None
pandas_datareader: None
bs4              : None
bottleneck       : None
fastparquet      : None
gcsfs            : None
lxml.etree       : None
matplotlib       : 3.3.0
numexpr          : None
odfpy            : None
openpyxl         : None
pandas_gbq       : None
pyarrow          : None
pytables         : None
pytest           : None
pyxlsb           : None
s3fs             : None
scipy            : None
sqlalchemy       : None
tables           : None
tabulate         : None
xarray           : None
xlrd             : None
xlwt             : None
xlsxwriter       : None
numba            : None

[TomAugspurger]

@ZahlGraf do you see anything in the release notes that indicates what might have changed?

[ZahlGraf]

Unfortunately not really. There are some changes to ticks, but they are not really related to dates:
https://matplotlib.org/3.3.0/users/whats_new.html

I also remember a deprecation warning about epoch2num, coming from pandas code. But there is already an issue for that (https://github.com/pandas-dev/pandas/issues/34850) here and furthermore it is just a warning about upcoming changes.

[TomAugspurger]

cc @jklymak on the hunch that this is related to the variable epoch changes in https://github.com/matplotlib/matplotlib/pull/15008, but that's just a guess.

I'll hopefully have time to look more closely later in the week.

[jklymak]

Yes our epoch changed from 0000-12-31T00:00:00 to 1970-01-01T00:00:00. You can check for the old behaviour via
plt.rcParams['date.epoch'] = '0000-12-31T00:00:00' and see if that fixes things.

[jklymak]

I don't know why pandas relies on matplotlib's epoch - presumably you use mdates.date2num and its now incorrect. However, if at all possible its helpful to use a more modern epoch to give more resolution in floating point times.

[ZahlGraf]

I can confirm, that using plt.rcParams['date.epoch'] = '0000-12-31T00:00:00' as a workaround fixes the issue.

import datetime
import matplotlib
import matplotlib.pyplot as plt
from matplotlib.dates import DateFormatter

from pandas.plotting import register_matplotlib_converters
register_matplotlib_converters()

plt.rcParams['date.epoch'] = '0000-12-31T00:00:00'

t = [
    datetime.datetime(2022, 12, 31, 0, 0), 
    datetime.datetime(2022, 11, 30, 0, 0), 
    datetime.datetime(2022, 10, 31, 0, 0),
]
s = [0.0, 0.1, 0.2]

fig, ax = plt.subplots()
ax.plot_date(x=t, y=s)
ax.xaxis.set_major_formatter(DateFormatter("%b '%y"))

plt.xticks(rotation=90)
plt.show()

[jklymak]

Glad there is a workaround, and sorry for the inconvenience. I should have pinged pandas when the change went in. https://matplotlib.org/3.3.0/users/whats_new.html#dates-use-a-modern-epoch