What's new in Matplotlib 3.3.0

For a list of all of the issues and pull requests since the last revision, see the GitHub Stats.

Table of Contents

Figure and Axes creation / management

Provisional API for composing semantic axes layouts from text or nested lists

The Figure class has a provisional method to generate complex grids of named axes.Axes based on nested list input or ASCII art:

axd = plt.figure(constrained_layout=True).subplot_mosaic(
    [['.', 'histx'],
     ['histy', 'scat']]
for k, ax in axd.items():
    ax.text(0.5, 0.5, k,
            ha='center', va='center', fontsize=36,

(Source code, png, pdf)


or as a string (with single-character Axes labels):

axd = plt.figure(constrained_layout=True).subplot_mosaic(
for k, ax in axd.items():
    ax.text(0.5, 0.5, k,
            ha='center', va='center', fontsize=36,

(Source code, png, pdf)


See Complex and semantic figure composition for more details and examples.


The GridSpec class gained a subplots method, so that one can write

fig.add_gridspec(2, 2, height_ratios=[3, 1]).subplots()

as an alternative to

fig.subplots(2, 2, gridspec_kw={"height_ratios": [3, 1]})

New Axes.sharex, Axes.sharey methods

These new methods allow sharing axes immediately after creating them. Note that behavior is indeterminate if axes are not shared immediately after creation.

For example, they can be used to selectively link some axes created all together using subplot_mosaic:

fig = plt.figure(constrained_layout=True)
axd = fig.subplot_mosaic([['.', 'histx'], ['histy', 'scat']],
                         gridspec_kw={'width_ratios': [1, 7],
                                      'height_ratios': [2, 7]})


(Source code, png, pdf)


tight_layout now supports suptitle

Previous versions did not consider Figure.suptitle, so it may overlap with other artists after calling tight_layout:

(Source code, png, pdf)


From now on, the suptitle will be considered:

(Source code, png, pdf)


Setting axes box aspect

It is now possible to set the aspect of an axes box directly via set_box_aspect. The box aspect is the ratio between axes height and axes width in physical units, independent of the data limits. This is useful to, e.g., produce a square plot, independent of the data it contains, or to have a non-image plot with the same axes dimensions next to an image plot with fixed (data-)aspect.

For use cases check out the Axes box aspect example.

Colors and colormaps

Turbo colormap

Turbo is an improved rainbow colormap for visualization, created by the Google AI team for computer vision and machine learning. Its purpose is to display depth and disparity data. Please see the Google AI Blog for further details.

(Source code, png, pdf)


colors.BoundaryNorm supports extend keyword argument

BoundaryNorm now has an extend keyword argument, analogous to extend in contourf. When set to 'both', 'min', or 'max', it maps the corresponding out-of-range values to Colormap lookup-table indices near the appropriate ends of their range so that the colors for out-of range values are adjacent to, but distinct from, their in-range neighbors. The colorbar inherits the extend argument from the norm, so with extend='both', for example, the colorbar will have triangular extensions for out-of-range values with colors that differ from adjacent in-range colors.

Text color for legend labels

The text color of legend labels can now be set by passing a parameter labelcolor to legend. The labelcolor keyword can be:

  • A single color (either a string or RGBA tuple), which adjusts the text color of all the labels.
  • A list or tuple, allowing the text color of each label to be set individually.
  • linecolor, which sets the text color of each label to match the corresponding line color.
  • markerfacecolor, which sets the text color of each label to match the corresponding marker face color.
  • markeredgecolor, which sets the text color of each label to match the corresponding marker edge color.

(Source code, png, pdf)


Pcolor and Pcolormesh now accept shading='nearest' and 'auto'

Previously axes.Axes.pcolor and axes.Axes.pcolormesh handled the situation where x and y have the same (respective) size as C by dropping the last row and column of C, and x and y are regarded as the edges of the remaining rows and columns in C. However, many users want x and y centered on the rows and columns of C.

To accommodate this, shading='nearest' and shading='auto' are new allowed strings for the shading keyword argument. 'nearest' will center the color on x and y if x and y have the same dimensions as C (otherwise an error will be thrown). shading='auto' will choose 'flat' or 'nearest' based on the size of X, Y, C.

If shading='flat' then X, and Y should have dimensions one larger than C. If X and Y have the same dimensions as C, then the previous behavior is used and the last row and column of C are dropped, and a DeprecationWarning is emitted.

Users can also specify this by the new rcParams["pcolor.shading"] (default: 'flat') in their .matplotlibrc or via rcParams.

See pcolormesh for examples.

Titles, ticks, and labels

Align labels to Axes edges

set_xlabel, set_ylabel and ColorbarBase.set_label support a parameter loc for simplified positioning. For the xlabel, the supported values are 'left', 'center', or 'right'. For the ylabel, the supported values are 'bottom', 'center', or 'top'.

The default is controlled via rcParams["xaxis.labelposition"] and rcParams["yaxis.labelposition"]; the Colorbar label takes the rcParam based on its orientation.

(Source code)


(png, pdf)


(png, pdf)

Allow tick formatters to be set with str or function inputs

set_major_formatter and set_minor_formatter now accept str or function inputs in addition to Formatter instances. For a str a StrMethodFormatter is automatically generated and used. For a function a FuncFormatter is automatically generated and used. In other words,

ax.xaxis.set_major_formatter('{x} km')
ax.xaxis.set_minor_formatter(lambda x, pos: str(x-5))

are shortcuts for:

import matplotlib.ticker as mticker

ax.xaxis.set_major_formatter(mticker.StrMethodFormatter('{x} km'))
    mticker.FuncFormatter(lambda x, pos: str(x-5))

(Source code, png, pdf)


Axes.set_title gains a y keyword argument to control auto positioning

set_title tries to auto-position the title to avoid any decorators on the top x-axis. This is not always desirable so now y is an explicit keyword argument of set_title. It defaults to None which means to use auto-positioning. If a value is supplied (i.e. the pre-3.0 default was y=1.0) then auto-positioning is turned off. This can also be set with the new rcParameter rcParams["axes.titley"] (default: None).

(Source code, png, pdf)


Offset text is now set to the top when using axis.tick_top()

Solves the issue that the power indicator (e.g., 1e4) stayed on the bottom, even if the ticks were on the top.

Set zorder of contour labels

clabel now accepts a zorder keyword argument making it easier to set the zorder of contour labels. If not specified, the default zorder of clabels used to always be 3 (i.e. the default zorder of Text) irrespective of the zorder passed to contour/contourf. The new default zorder for clabels has been changed to (2 + zorder passed to contour / contourf).

Other changes

New Axes.axline method

A new axline method has been added to draw infinitely long lines that pass through two points.

fig, ax = plt.subplots()

ax.axline((.1, .1), slope=5, color='C0', label='by slope')
ax.axline((.1, .2), (.8, .7), color='C3', label='by points')


(Source code, png, pdf)


imshow now coerces 3D arrays with depth 1 to 2D

Starting from this version arrays of size MxNx1 will be coerced into MxN for displaying. This means commands like plt.imshow(np.random.rand(3, 3, 1)) will no longer return an error message that the image shape is invalid.

Better control of Axes.pie normalization

Previously, Axes.pie would normalize its input x if sum(x) > 1, but would do nothing if the sum were less than 1. This can be confusing, so an explicit keyword argument normalize has been added. By default, the old behavior is preserved.

By passing normalize, one can explicitly control whether any rescaling takes place or whether partial pies should be created. If normalization is disabled, and sum(x) > 1, then an error is raised.

(Source code, png, pdf)


Dates use a modern epoch

Matplotlib converts dates to days since an epoch using dates.date2num (via matplotlib.units). Previously, an epoch of 0000-12-31T00:00:00 was used so that 0001-01-01 was converted to 1.0. An epoch so distant in the past meant that a modern date was not able to preserve microseconds because 2000 years times the 2^(-52) resolution of a 64-bit float gives 14 microseconds.

Here we change the default epoch to the more reasonable UNIX default of 1970-01-01T00:00:00 which for a modern date has 0.35 microsecond resolution. (Finer resolution is not possible because we rely on datetime.datetime for the date locators). Access to the epoch is provided by get_epoch, and there is a new rcParams["date.epoch"] (default: '1970-01-01T00:00:00') rcParam. The user may also call set_epoch, but it must be set before any date conversion or plotting is used.

If you have data stored as ordinal floats in the old epoch, you can convert them to the new ordinal using the following formula:

new_ordinal = old_ordinal + mdates.date2num(np.datetime64('0000-12-31'))

Lines now accept MarkerStyle instances as input

Similar to scatter, plot and Line2D now accept MarkerStyle instances as input for the marker parameter:

plt.plot(..., marker=matplotlib.markers.MarkerStyle("D"))


Simple syntax to select fonts by absolute path

Fonts can now be selected by passing an absolute pathlib.Path to the font keyword argument of Text.

Improved font weight detection

Matplotlib is now better able to determine the weight of fonts from their metadata, allowing to differentiate between fonts within the same family more accurately.

rcParams improvements

matplotlib.rc_context can be used as a decorator

matplotlib.rc_context can now be used as a decorator (technically, it is now implemented as a contextlib.contextmanager), e.g.,

@rc_context({"lines.linewidth": 2})
def some_function(...):

rcParams for controlling default "raise window" behavior

The new config option rcParams["figure.raise_window"] (default: True) allows disabling of the raising of the plot window when calling show or pause. The MacOSX backend is currently not supported.

Add generalized mathtext.fallback to rcParams

New rcParams["mathtext.fallback"] (default: 'cm') rcParam. Takes "cm", "stix", "stixsans" or "none" to turn fallback off. The rcParam mathtext.fallback_to_cm is deprecated, but if used, will override new fallback.

Add contour.linewidth to rcParams

The new config option rcParams["contour.linewidth"] (default: None) allows to control the default line width of contours as a float. When set to None, the line widths fall back to rcParams["lines.linewidth"] (default: 1.5). The config value is overridden as usual by the linewidths argument passed to contour when it is not set to None.

3D Axes improvements

Axes3D no longer distorts the 3D plot to match the 2D aspect ratio

Plots made with Axes3D were previously stretched to fit a square bounding box. As this stretching was done after the projection from 3D to 2D, it resulted in distorted images if non-square bounding boxes were used. As of 3.3, this no longer occurs.

Currently, modes of setting the aspect (via set_aspect) in data space are not supported for Axes3D but may be in the future. If you want to simulate having equal aspect in data space, set the ratio of your data limits to match the value of get_box_aspect. To control these ratios use the set_box_aspect method which accepts the ratios as a 3-tuple of X:Y:Z. The default aspect ratio is 4:4:3.

3D axes now support minor ticks

ax = plt.figure().add_subplot(projection='3d')

ax.scatter([0, 1, 2], [1, 3, 5], [30, 50, 70])

ax.set_xticks([0.25, 0.75, 1.25, 1.75], minor=True)
ax.set_xticklabels(['a', 'b', 'c', 'd'], minor=True)

ax.set_yticks([1.5, 2.5, 3.5, 4.5], minor=True)
ax.set_yticklabels(['A', 'B', 'C', 'D'], minor=True)

ax.set_zticks([35, 45, 55, 65], minor=True)
ax.set_zticklabels([r'$\alpha$', r'$\beta$', r'$\delta$', r'$\gamma$'],

ax.tick_params(which='major', color='C0', labelcolor='C0', width=5)
ax.tick_params(which='minor', color='C1', labelcolor='C1', width=3)

(Source code, png, pdf)


Interactive tool improvements

More consistent toolbar behavior across backends

Toolbar features are now more consistent across backends. The history buttons will auto-disable when there is no further action in a direction. The pan and zoom buttons will be marked active when they are in use.

In NbAgg and WebAgg, the toolbar buttons are now grouped similarly to other backends. The WebAgg toolbar now uses the same icons as other backends.

Toolbar icons are now styled for dark themes

On dark themes, toolbar icons will now be inverted. When using the GTK3Agg backend, toolbar icons are now symbolic, and both foreground and background colors will follow the theme. Tooltips should also behave correctly.

Cursor text now uses a number of significant digits matching pointing precision

Previously, the x/y position displayed by the cursor text would usually include far more significant digits than the mouse pointing precision (typically one pixel). This is now fixed for linear scales.

GTK / Qt zoom rectangle now black and white

This makes it visible even over a dark background.

Event handler simplifications

The backend_bases.key_press_handler and backend_bases.button_press_handler event handlers can now be directly connected to a canvas with canvas.mpl_connect("key_press_event", key_press_handler) and canvas.mpl_connect("button_press_event", button_press_handler), rather than having to write wrapper functions that fill in the (now optional) canvas and toolbar parameters.

Functions to compute a Path's size

Various functions were added to BezierSegment and Path to allow computation of the shape/size of a Path and its composite Bezier curves.

In addition to the fixes below, BezierSegment has gained more documentation and usability improvements, including properties that contain its dimension, degree, control_points, and more.

Better interface for Path segment iteration

iter_bezier iterates through the BezierSegment's that make up the Path. This is much more useful typically than the existing iter_segments function, which returns the absolute minimum amount of information possible to reconstruct the Path.

Fixed bug that computed a Path's Bbox incorrectly

Historically, get_extents has always simply returned the Bbox of a curve's control points, instead of the Bbox of the curve itself. While this is a correct upper bound for the path's extents, it can differ dramatically from the Path's actual extents for non-linear Bezier curves.

Backend-specific improvements

savefig() gained a backend keyword argument

The backend keyword argument to savefig can now be used to pick the rendering backend without having to globally set the backend; e.g., one can save PDFs using the pgf backend with savefig("file.pdf", backend="pgf").

The SVG backend can now render hatches with transparency

The SVG backend now respects the hatch stroke alpha. Useful applications are, among others, semi-transparent hatches as a subtle way to differentiate columns in bar plots.

SVG supports URLs on more artists

URLs on more artists (i.e., from Artist.set_url) will now be saved in SVG files, namely, Ticks and Line2Ds are now supported.

Images in SVG will no longer be blurred in some viewers

A style is now supplied to images without interpolation (imshow(..., interpolation='none') so that SVG image viewers will no longer perform interpolation when rendering themselves.

Saving SVG now supports adding metadata

When saving SVG files, metadata can now be passed which will be saved in the file using Dublin Core and RDF. A list of valid metadata can be found in the documentation for FigureCanvasSVG.print_svg.

Saving PDF metadata via PGF now consistent with PDF backend

When saving PDF files using the PGF backend, passed metadata will be interpreted in the same way as with the PDF backend. Previously, this metadata was only accepted by the PGF backend when saving a multi-page PDF with backend_pgf.PdfPages, but is now allowed when saving a single figure, as well.

NbAgg and WebAgg no longer use jQuery & jQuery UI

Instead, they are implemented using vanilla JavaScript. Please report any issues with browsers.