Changes to major/minor locator on a DateTime index leads to messed up axis

[niklassemmler]

Code Sample

import matplotlib
matplotlib.use('Agg')
import matplotlib.ticker as ticker
import matplotlib.dates as mdates
from matplotlib import pyplot as plt

from datetime import datetime
import pandas

# Creates problems
hours = range(24)

# Works fine
# hours = list(range(12)) + list(range(13, 24))

times = []
for hour in hours:
     times.append(datetime(year=2019, month=5, day=6, hour=hour))
df_example = pandas.DataFrame({'column':range(len(times))}, index=times)

f, ax = plt.subplots()

# Works fine
# ax.plot(df_example.index, df_example['column'])

# Creates problems
df_example.plot(ax=ax)

hour_locator = mdates.HourLocator(byhour=range(0, 24, 6))
ax.xaxis.set_minor_formatter(mdates.DateFormatter('%H:%M'))
ax.xaxis.set_minor_locator(hour_locator)
ax.xaxis.set_major_locator(ticker.NullLocator())
ax.xaxis.set_major_formatter(ticker.NullFormatter())
f.savefig('example.png')

Problem description

I was plotting a DataFrame with a DateTime index and tried to change the major and minor ticks afterwards. I expected an x-axis with ticks every 6-hours. Instead the x-axis became filled with ticks at what I assume to be hourly intervals.

This behavior does not occur, when either matplotlib is used to plot the columns ax.plot(...) or when the DateTime index contains gaps. Both cases are shown in the comments above.

The issue appeared both with pandas versions 0.24.1 and 0.24.2.

Expected Output

Output

[gfyoung]

cc @TomAugspurger @mroeschke

[TomAugspurger]

Pandas tick formatting is quite a bit more complex that that. I would recommend just using matplotlib entirely, rather than trying to customize a pandas-produced axes.

[TomAugspurger]

If you can identify a specific issue, then we can leave this open. Otherwise I worry that it's just going to sit around in an unclear state.

[niklassemmler]

Hi Tom, in the end I used matplotlib directly. I think it would help to at least a sentence somewhere that states your recommendation. That would have at least helped me. Maybe a comment in a documentation would be good or maybe this issue is enough.

[TomAugspurger]

A clarification in http://pandas-docs.github.io/pandas-docs-travis/user_guide/visualization.html#plotting-directly-with-matplotlib would be welcome if you can think of a good phrasing.

…

On Fri, May 17, 2019 at 2:53 AM metaswirl ***@***.***> wrote: Hi Tom, in the end I used matplotlib directly. I think it would help to at least a sentence somewhere that states your recommendation. That would have at least helped me. Maybe a comment in a documentation would be good or maybe this issue is enough. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#26293?email_source=notifications&email_token=AAKAOIQZ4QJI4T2YGUXAAXLPVZQBFA5CNFSM4HLARK22YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODVUBARY#issuecomment-493359175>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAKAOIWIONFLP2IWZGRIQLTPVZQBFANCNFSM4HLARK2Q> .

[niklassemmler]

Well, the wording could be either "Don't use pandas for changing ticks in a timeseries plot." to "While ticks in a timeseries plot can be changed in pandas, it is still an experimental feature. Use matplotlib when you experience troubles." Or if you have more insight, in when the problem occurs, then it would help to mention that.

[TomAugspurger]

Something like that sounds reasonable, though I might not use "experimental". More so not easily customizable after the fact, compared to if you used matplotlib directly. And this is a general recommendation with DataFrame.plot, not timeseries-specific. Are you interested in making a PR adding that to the documentation?

…

On Wed, May 22, 2019 at 6:43 AM metaswirl ***@***.***> wrote: Well, the wording could be either "Don't use pandas for changing ticks in a timeseries plot." to "While ticks in a timeseries plot can be changed in pandas, it is still an experimental feature. Use matplotlib when you experience troubles." Or if you have more insight, in when the problem occurs, then it would help to mention that. — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#26293?email_source=notifications&email_token=AAKAOIU2ZJSUYWGG6UAUGJLPWUWWJA5CNFSM4HLARK22YY3PNVWWK3TUL52HS4DFVREXG43VMVBW63LNMVXHJKTDN5WW2ZLOORPWSZGODV6Y25Q#issuecomment-494767478>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAKAOIU4G4Z3ADUH7GHDXIDPWUWWJANCNFSM4HLARK2Q> .

[niklassemmler]

Sorry for the delay, above you find a pull request to add a line to the docs.