Bug report

Bug summary

MaxNLocator changes the scientific notation with different number of tick labels.

Code for reproduction

I encountered this unexpected behavior when I changed the y-ticklabel size. Going from size 13 to 14 changes the scientific notation from 1e9 to 1e10, and adds an decimal offset to all the ticklabels.

import matplotlib.pyplot as plt

sci_nums = [10**9, 8*10**9]
plt.rcParams['ytick.labelsize'] = 13
fig, ax = plt.subplots()
ax.scatter(range(len(sci_nums)), sci_nums)

plt.rcParams['ytick.labelsize'] = 14
fig, ax = plt.subplots()
ax.scatter(range(len(sci_nums)), sci_nums)

This is related to that the larger font size reduces the number of tickslabels. MaxNLocator apparently adjusts the scientific notation depending on the number of ticklabels.

plt.rcParams['ytick.labelsize'] = 13
fig, ax = plt.subplots()
ax.scatter(range(len(sci_nums)), sci_nums)

import matplotlib.ticker as ticker
ax.yaxis.set_major_locator(ticker.MaxNLocator(4))

I dont' quite understand the rule that MaxNLocator uses. Setting 1, 2, 4 or 5 ticklabels results in 1e10. Setting 3, 6, 7, 8, or 9 ticklabels results in 1e9.

Expected outcome
I would expect that changing the number of ticklabels (either directly or via another parameter such as the font size) would not result in a change in the formatting of the scientific notation. In this case, I would expect it to remain 1e9 for all the examples above.

Matplotlib version

• Operating system: Linux
• Matplotlib version: 2.2.2
• Matplotlib backend (print(matplotlib.get_backend())): module://ipykernel.pylab.backend_inline
• Python version: 3.6.6 (conda-forge)
• Jupyter version (if applicable): jupyterlab=0.34.3, jupyter_core=4.4.0
• Other libraries: