pandas: plot hourly ticks with datetime.time index

Question

I have a pandas DataFrame indexed by a DatetimeIndex that holds a time series, i.e. some data as a function of time. Now I would like to plot the behavior over the day regardless of the date (cf. this question):

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.dates as mdates

idx = pd.date_range('2017-01-01 05:03', '2017-01-05 18:03', freq = '30min')
df = pd.Series(np.random.randn(len(idx)),  index = idx)

hours = mdates.HourLocator(interval = 1)
h_fmt = mdates.DateFormatter('%H:%M:%S')

for date, group in df.groupby(by = df.index.date):
  group.index = group.index.timetz
  group.name  = date # for legend
  ax = group.plot()
plt.ion()
plt.show()

This works but the labels on the x-axis have peculiar spacings:

I'd prefer, e.g., to have ticks every hour on the hour. Based on this SO answer I found a solution that works, setting x_compat and using the HourLocator:

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.dates as mdates

idx = pd.date_range('2017-01-01 05:03', '2017-01-01 18:03', freq = '30min')
df = pd.Series(np.random.randn(len(idx)),  index = idx)

hours = mdates.HourLocator(interval = 1)
h_fmt = mdates.DateFormatter('%H:%M:%S')

with pd.plotting.plot_params.use('x_compat', True): 
  ax = df.plot() 
  ax.xaxis.set_major_locator(hours)
  ax.xaxis.set_major_formatter(h_fmt)

plt.ion()
plt.show()

This gives the following plot (note I have reduced the date_range to one day here):

It still works when splitting with groupby and plotting more data:

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.dates as mdates

idx = pd.date_range('2017-01-01 05:03', '2017-01-05 18:03', freq = '30min')
df = pd.Series(np.random.randn(len(idx)),  index = idx)

hours = mdates.HourLocator(interval = 1)
h_fmt = mdates.DateFormatter('%H:%M:%S')

with pd.plotting.plot_params.use('x_compat', True): 
  for date, group in df.groupby(by = df.index.date):
    ax = group.plot()
    ax.xaxis.set_major_locator(hours)
    ax.xaxis.set_major_formatter(h_fmt)

plt.ion()
plt.show()

Of course, I still need to wrap around (drop) the date here. But once I do that, my solution no longer works:

import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.dates as mdates

idx = pd.date_range('2017-01-01 05:03', '2017-01-05 18:03', freq = '30min')
df = pd.Series(np.random.randn(len(idx)),  index = idx)

hours = mdates.HourLocator(interval = 1)
h_fmt = mdates.DateFormatter('%H:%M:%S')

with pd.plotting.plot_params.use('x_compat', True): 
  for date, group in df.groupby(by = df.index.date):
    group.index = group.index.timetz
    group.name  = date # for legend
    ax = group.plot()
    ax.xaxis.set_major_locator(hours)
    ax.xaxis.set_major_formatter(h_fmt)

plt.ion()
plt.show()

After computing for a while, this throws an error, maybe it's trying to make ticks starting at pandas 0 for timestamps?

RuntimeError: Locator attempting to generate 2030401 ticks from 180.0 to 84780.0: exceeds Locator.MAXTICKS

Answer 1

We can use ax.axis to return the x- and y-axis ranges. This reveals that the x-axis range is likely represented as seconds internally:

In [11]: ax.axis()
Out[11]: (180.0, 84780.0, -3.96605612012256, 3.4854575601641957)

So we can use the MultipleLocator:

import matplotlib.ticker as ticker
...
n = 2
ax.xaxis.set_major_locator(ticker.MultipleLocator(3600*n))

to have ticks every n hours: