UserWarning: FixedFormatter should only be used together with FixedLocator

Each Answer to this Q is separated by one/two green lines.

I have used for a long time small subroutines to format axes of charts I’m plotting. A couple of examples:

def format_y_label_thousands(): # format y-axis tick labels formats
    ax = plt.gca()
    label_format="{:,.0f}"
    ax.set_yticklabels([label_format.format(x) for x in ax.get_yticks().tolist()])

def format_y_label_percent(): # format y-axis tick labels formats
    ax = plt.gca()
    label_format="{:.1%}"
    ax.set_yticklabels([label_format.format(x) for x in ax.get_yticks().tolist()])

However, after an update to matplotlib yesterday, I get the following warning when calling any of these two functions:

UserWarning: FixedFormatter should only be used together with FixedLocator
  ax.set_yticklabels([label_format.format(x) for x in ax.get_yticks().tolist()])

What is the reason for such a warning? I couldn’t figure it out looking into matplotlib’s documentation.

WORKAROUND:

The way to avoid the warning is to use FixedLocator (that is part of matplotlib.ticker). Below I show a code to plot three charts. I format their axes in different ways. Note that the “set_ticks” silence the warning, but it changes the actual ticks locations/labels (it took me some time to figure out that FixedLocator uses the same info but keeps the ticks locations intact). You can play with the x/y’s to see how each solution might affect the output.

import matplotlib as mpl
import matplotlib.pyplot as plt
import numpy as np
import matplotlib.ticker as mticker

mpl.rcParams['font.size'] = 6.5

x = np.array(range(1000, 5000, 500))
y = 37*x

fig, [ax1, ax2, ax3] = plt.subplots(1,3)

ax1.plot(x,y, linewidth=5, color="green")
ax2.plot(x,y, linewidth=5, color="red")
ax3.plot(x,y, linewidth=5, color="blue")

label_format="{:,.0f}"

# nothing done to ax1 as it is a "control chart."

# fixing yticks with "set_yticks"
ticks_loc = ax2.get_yticks().tolist()
ax2.set_yticks(ax1.get_yticks().tolist())
ax2.set_yticklabels([label_format.format(x) for x in ticks_loc])

# fixing yticks with matplotlib.ticker "FixedLocator"
ticks_loc = ax3.get_yticks().tolist()
ax3.yaxis.set_major_locator(mticker.FixedLocator(ticks_loc))
ax3.set_yticklabels([label_format.format(x) for x in ticks_loc])

# fixing xticks with FixedLocator but also using MaxNLocator to avoid cramped x-labels
ax3.xaxis.set_major_locator(mticker.MaxNLocator(3))
ticks_loc = ax3.get_xticks().tolist()
ax3.xaxis.set_major_locator(mticker.FixedLocator(ticks_loc))
ax3.set_xticklabels([label_format.format(x) for x in ticks_loc])

fig.tight_layout()
plt.show()

OUTPUT CHARTS:

Sample charts

Obviously, having a couple of idle lines of code like the one above (I’m basically getting the yticks or xticks and setting them again) only adds noise to my program. I would prefer that the warning was removed. However, look into some of the “bug reports” (from links on the comments above/below; the issue is not actually a bug: it is an update that is generating some issues), and the contributors that manage matplotlib have their reasons to keep the warning.

OLDER VERSION OF MATPLOTLIB:
If you use your Console to control critical outputs of your code (as I do), the warning messages might be problematic. Therefore, a way to delay having to deal with the issue is to downgrade matplotlib to version 3.2.2. I use Anaconda to manage my Python packages, and here is the command used to downgrade matplotlib:

conda install matplotlib=3.2.2

Not all listed versions might be available. For instance, couldn’t install matplotlib 3.3.0 although it is listed on matplotlib’s releases page: https://github.com/matplotlib/matplotlib/releases

If someone comes here using the function axes.xaxis.set_ticklabels() (or yaxis equivalent), you don’t need to use FixedLocator, you can avoid this warning using axes.xaxis.set_ticks(values_list) BEFORE axes.xaxis.set_ticklabels(labels_list).

According to this matplotlib page

# FixedFormatter should only be used together with FixedLocator. 
# Otherwise, one cannot be sure where the labels will end up.

This means one should do

positions = [0, 1, 2, 3, 4, 5]
labels = ['A', 'B', 'C', 'D', 'E', 'F']
ax.xaxis.set_major_locator(ticker.FixedLocator(positions))
ax.xaxis.set_major_formatter(ticker.FixedFormatter(labels))

But the issue also persisted with the ticker.LogLocator even if the labels were passed to ticker.FixedFormatter.
So the solution in this case was

  1. Define a formatter function

    # FuncFormatter can be used as a decorator
    @ticker.FuncFormatter
    def major_formatter(x, pos):
        return f'{x:.2f}'
    
  2. and pass the formatter function to the FixedFormatter

    ax.xaxis.set_major_locator(ticker.LogLocator(base=10, numticks=5))
    ax.xaxis.set_major_formatter(major_formatter)
    

See the above link for details.

Simplest workaround is to suppress warnings (this includes UserWarning):

import warnings
warnings.filterwarnings("ignore")

The use-case would be if you don’t want your jupyter notebook on github to look trashed with warning messages. Unlike most warnings, this warning keeps repeating if you’re in a loop (python 3.7).

I had the same problem as I tried to rotate the tick labels on the X-axis with located date ticks:

ax.set_xticklabels(ax.get_xticklabels(), rotation=45)
ax.xaxis.set_major_locator(dates.DayLocator())

It worked out using the ‘tick_params()’ method:

ax.tick_params(axis="x", labelrotation = 45)

just use axis.set_xticks([labels])


The answers/resolutions are collected from stackoverflow, are licensed under cc by-sa 2.5 , cc by-sa 3.0 and cc by-sa 4.0 .