Axis Ticks#

The x and y Axis on each Axes have default tick "locators" and "formatters" that depend on the scale being used (see Axis scales). It is possible to customize the ticks and tick labels with either high-level methods like set_xticks or set the locators and formatters directly on the axis.

Manual location and formats#

The simplest method to customize the tick locations and formats is to use set_xticks and set_yticks. These can be used on either the major or the minor ticks.

import numpy as np
import matplotlib.pyplot as plt

import matplotlib.ticker as ticker


fig, axs = plt.subplots(2, 1, figsize=(5.4, 5.4), layout='constrained')
x = np.arange(100)
for nn, ax in enumerate(axs):
    ax.plot(x, x)
    if nn == 1:
        ax.set_title('Manual ticks')
        ax.set_yticks(np.arange(0, 100.1, 100/3))
        xticks = np.arange(0.50, 101, 20)
        xlabels = [f'\\${x:1.2f}' for x in xticks]
        ax.set_xticks(xticks, labels=xlabels)
    else:
        ax.set_title('Automatic ticks')
Automatic ticks, Manual ticks

Note that the length of the labels argument must have the same length as the array used to specify the ticks.

By default set_xticks and set_yticks act on the major ticks of an Axis, however it is possible to add minor ticks:

fig, axs = plt.subplots(2, 1, figsize=(5.4, 5.4), layout='constrained')
x = np.arange(100)
for nn, ax in enumerate(axs):
    ax.plot(x, x)
    if nn == 1:
        ax.set_title('Manual ticks')
        ax.set_yticks(np.arange(0, 100.1, 100/3))
        ax.set_yticks(np.arange(0, 100.1, 100/30), minor=True)
    else:
        ax.set_title('Automatic ticks')
Automatic ticks, Manual ticks

Locators and Formatters#

Manually setting the ticks as above works well for specific final plots, but does not adapt as the user interacts with the axes. At a lower level, Matplotlib has Locators that are meant to automatically choose ticks depending on the current view limits of the axis, and Formatters that are meant to format the tick labels automatically.

The full list of locators provided by Matplotlib are listed at Tick locating, and the formatters at Tick formatting.

def setup(ax, title):
    """Set up common parameters for the Axes in the example."""
    # only show the bottom spine
    ax.yaxis.set_major_locator(ticker.NullLocator())
    ax.spines[['left', 'right', 'top']].set_visible(False)

    ax.xaxis.set_ticks_position('bottom')
    ax.tick_params(which='major', width=1.00, length=5)
    ax.tick_params(which='minor', width=0.75, length=2.5)
    ax.set_xlim(0, 5)
    ax.set_ylim(0, 1)
    ax.text(0.0, 0.2, title, transform=ax.transAxes,
            fontsize=14, fontname='Monospace', color='tab:blue')


fig, axs = plt.subplots(8, 1, layout='constrained')

# Null Locator
setup(axs[0], title="NullLocator()")
axs[0].xaxis.set_major_locator(ticker.NullLocator())
axs[0].xaxis.set_minor_locator(ticker.NullLocator())

# Multiple Locator
setup(axs[1], title="MultipleLocator(0.5)")
axs[1].xaxis.set_major_locator(ticker.MultipleLocator(0.5))
axs[1].xaxis.set_minor_locator(ticker.MultipleLocator(0.1))

# Fixed Locator
setup(axs[2], title="FixedLocator([0, 1, 5])")
axs[2].xaxis.set_major_locator(ticker.FixedLocator([0, 1, 5]))
axs[2].xaxis.set_minor_locator(ticker.FixedLocator(np.linspace(0.2, 0.8, 4)))

# Linear Locator
setup(axs[3], title="LinearLocator(numticks=3)")
axs[3].xaxis.set_major_locator(ticker.LinearLocator(3))
axs[3].xaxis.set_minor_locator(ticker.LinearLocator(31))

# Index Locator
setup(axs[4], title="IndexLocator(base=0.5, offset=0.25)")
axs[4].plot(range(0, 5), [0]*5, color='white')
axs[4].xaxis.set_major_locator(ticker.IndexLocator(base=0.5, offset=0.25))

# Auto Locator
setup(axs[5], title="AutoLocator()")
axs[5].xaxis.set_major_locator(ticker.AutoLocator())
axs[5].xaxis.set_minor_locator(ticker.AutoMinorLocator())

# MaxN Locator
setup(axs[6], title="MaxNLocator(n=4)")
axs[6].xaxis.set_major_locator(ticker.MaxNLocator(4))
axs[6].xaxis.set_minor_locator(ticker.MaxNLocator(40))

# Log Locator
setup(axs[7], title="LogLocator(base=10, numticks=15)")
axs[7].set_xlim(10**3, 10**10)
axs[7].set_xscale('log')
axs[7].xaxis.set_major_locator(ticker.LogLocator(base=10, numticks=15))
plt.show()
axes ticks

Similarly, we can specify "Formatters" for the major and minor ticks on each axis.

The tick format is configured via the function set_major_formatter or set_minor_formatter. It accepts:

See Tick formatting for the complete list.

def setup(ax, title):
    """Set up common parameters for the Axes in the example."""
    # only show the bottom spine
    ax.yaxis.set_major_locator(ticker.NullLocator())
    ax.spines[['left', 'right', 'top']].set_visible(False)

    # define tick positions
    ax.xaxis.set_major_locator(ticker.MultipleLocator(1.00))
    ax.xaxis.set_minor_locator(ticker.MultipleLocator(0.25))

    ax.xaxis.set_ticks_position('bottom')
    ax.tick_params(which='major', width=1.00, length=5)
    ax.tick_params(which='minor', width=0.75, length=2.5, labelsize=10)
    ax.set_xlim(0, 5)
    ax.set_ylim(0, 1)
    ax.text(0.0, 0.2, title, transform=ax.transAxes,
            fontsize=14, fontname='Monospace', color='tab:blue')


fig = plt.figure(figsize=(8, 8), layout='constrained')
fig0, fig1, fig2 = fig.subfigures(3, height_ratios=[1.5, 1.5, 7.5])

fig0.suptitle('String Formatting', fontsize=16, x=0, ha='left')
ax0 = fig0.subplots()

setup(ax0, title="'{x} km'")
ax0.xaxis.set_major_formatter('{x} km')

fig1.suptitle('Function Formatting', fontsize=16, x=0, ha='left')
ax1 = fig1.subplots()

setup(ax1, title="def(x, pos): return str(x-5)")
ax1.xaxis.set_major_formatter(lambda x, pos: str(x-5))

fig2.suptitle('Formatter Object Formatting', fontsize=16, x=0, ha='left')
axs2 = fig2.subplots(7, 1)

setup(axs2[0], title="NullFormatter()")
axs2[0].xaxis.set_major_formatter(ticker.NullFormatter())

setup(axs2[1], title="StrMethodFormatter('{x:.3f}')")
axs2[1].xaxis.set_major_formatter(ticker.StrMethodFormatter("{x:.3f}"))

setup(axs2[2], title="FormatStrFormatter('#%d')")
axs2[2].xaxis.set_major_formatter(ticker.FormatStrFormatter("#%d"))


def fmt_two_digits(x, pos):
    return f'[{x:.2f}]'


setup(axs2[3], title='FuncFormatter("[{:.2f}]".format)')
axs2[3].xaxis.set_major_formatter(ticker.FuncFormatter(fmt_two_digits))

setup(axs2[4], title="FixedFormatter(['A', 'B', 'C', 'D', 'E', 'F'])")
# FixedFormatter should only be used together with FixedLocator.
# Otherwise, one cannot be sure where the labels will end up.
positions = [0, 1, 2, 3, 4, 5]
labels = ['A', 'B', 'C', 'D', 'E', 'F']
axs2[4].xaxis.set_major_locator(ticker.FixedLocator(positions))
axs2[4].xaxis.set_major_formatter(ticker.FixedFormatter(labels))

setup(axs2[5], title="ScalarFormatter()")
axs2[5].xaxis.set_major_formatter(ticker.ScalarFormatter(useMathText=True))

setup(axs2[6], title="PercentFormatter(xmax=5)")
axs2[6].xaxis.set_major_formatter(ticker.PercentFormatter(xmax=5))
axes ticks

Styling ticks (tick parameters)#

The appearance of ticks can be controlled at a low level by finding the individual Tick on the axis. However, usually it is simplest to use tick_params to change all the objects at once.

The tick_params method can change the properties of ticks:

  • length

  • direction (in or out of the frame)

  • colors

  • width and length

  • and whether the ticks are drawn at the bottom, top, left, or right of the Axes.

It also can control the tick labels:

  • labelsize (fontsize)

  • labelcolor (color of the label)

  • labelrotation

  • labelbottom, labeltop, labelleft, labelright

In addition there is a pad keyword argument that specifies how far the tick label is from the tick.

Finally, the grid linestyles can be set:

  • grid_color

  • grid_alpha

  • grid_linewidth

  • grid_linestyle

All these properties can be restricted to one axis, and can be applied to just the major or minor ticks

fig, axs = plt.subplots(1, 2, figsize=(6.4, 3.2), layout='constrained')

for nn, ax in enumerate(axs):
    ax.plot(np.arange(100))
    if nn == 1:
        ax.grid('on')
        ax.tick_params(right=True, left=False, axis='y', color='r', length=16,
                       grid_color='none')
        ax.tick_params(axis='x', color='m', length=4, direction='in', width=4,
                       labelcolor='g', grid_color='b')
axes ticks

Total running time of the script: (0 minutes 2.153 seconds)

Gallery generated by Sphinx-Gallery