numpy and pandas datetime handling in matplotlib

Question

I'm trying to fit the curve to a time series as it explained here: https://stackoverflow.com/a/17639070/12012445

If I do not import pandas module - everything works like a charm. If I do, and I have to, - I get "'datetime.timezone' object has no attribute '_utcoffset'"

it seems to me that the issue is with plotting step were it returns the coefficients for a polynomial p(x) which in turn depends on datetime...

I've gone through a number of hints, including Converting between datetime, Timestamp and datetime64

I'm quite desperate (5th-hour researching)

import pandas as pd #note that I'm not even calling it anywhere... so confused

import numpy as np
import matplotlib.pyplot as plt
import matplotlib.dates as mdates
import datetime as dt
import time


dates = [dt.datetime(1978, 7, 7),
     dt.datetime(1980, 9, 26),
     dt.datetime(1983, 8, 1),
     dt.datetime(1985, 8, 8)]

y = [0.00134328779552718,
     0.00155187668863844,
     0.0039431374327427,
     0.00780037563783297]


yerr = [0.0000137547160254577,
        0.0000225670232594083,
        0.000105623642510075,
        0.00011343121508]

x = mdates.date2num(dates)

z4 = np.polyfit(x, y, 3)
p4 = np.poly1d(z4)

fig, cx = plt.subplots()

xx = np.linspace(x.min(), x.max(), 100)
dd = mdates.num2date(xx)

cx.plot(dd, p4(xx), '-g')
cx.plot(dates, y, '+', color='b', label='blub')
cx.errorbar(dates, y,
             yerr,
             marker='.',
             color='k',
             ecolor='b',
             markerfacecolor='b',
             label="series 1",
             capsize=0,
             linestyle='')

cx.grid()
cx.set_ylim(0, 0.03)
plt.show()

expecting:

Can you share the version of pandas and matplotlib you are using? — Madhuri Sangaraju, Sep 03 '19 at 05:40
I dont understand your question. The code works fine without pandas library. — Humi, Sep 03 '19 at 06:20
@Madhuri Sangaraju pandas 0.24.2, numpy 1.15.4, matplotlib 3.0.3 — Gudaev, Sep 03 '19 at 17:28
@Hue, that's right - the code works fine w/o pandas, but i have to have pandas in other areas.. i'm going to try [link](https://stackoverflow.com/a/57769554/12012445) answer below, will let you know — Gudaev, Sep 03 '19 at 17:31
[Answer](https://stackoverflow.com/a/57769554/12012445) below worked! Thanks! — Gudaev, Sep 03 '19 at 17:39

score 1 · Answer 1 · answered Apr 28 '21 at 17:10

The answer above works. On the other hand, if you want to keep use the pandas converter. However, the format of mdates is different. It counts number of days from year 0. Therefore, one can try to convert the mdates format into the standard datetime format by the following:

dd = mdates.num2date(xx)
t0 = mdates.date2num(pd.to_datetime(0)) # obtain 1970-01-01 in mdates
new_dates = pd.to_datetime(((dd-t0)*24*3600).astype(int), unit='s')
# explicitly register pandas converter to avoid a future warning
pd.plotting.register_matplotlib_converters()

Note that the problem appears to be gone by pandas version 0.25.1

score -1 · Accepted Answer · answered Sep 03 '19 at 10:11

If you import pandas, it would register its datetime converters to be used by matplotlib. Apparently there is a problem with those somewhere. And in case that happens, you may deregister them via

import pandas as pd
pd.plotting.deregister_matplotlib_converters()

This will ensure matplotlib converters are used and those seem to work as expected.

numpy and pandas datetime handling in matplotlib

2 Answers2