You are reading an old version of the documentation (v3.1.1). For the latest version see
Version 3.1.2
Fork me on GitHub

Table of Contents

New in Matplotlib 2.2

Constrained Layout Manager


Constrained Layout is experimental. The behaviour and API are subject to change, or the whole functionality may be removed without a deprecation period.

A new method to automatically decide spacing between subplots and their organizing GridSpec instances has been added. It is meant to replace the venerable tight_layout method. It is invoked via a new constrained_layout=True kwarg to Figure or subplots.

There are new rcParams for this package, and spacing can be more finely tuned with the new set_constrained_layout_pads.

Features include:

  • Automatic spacing for subplots with a fixed-size padding in inches around subplots and all their decorators, and space between as a fraction of subplot size between subplots.
  • Spacing for suptitle, and colorbars that are attached to more than one axes.
  • Nested GridSpec layouts using GridSpecFromSubplotSpec.

For more details and capabilities please see the new tutorial: Constrained Layout Guide

Note the new API to access this:

New plt.figure and plt.subplots kwarg: constrained_layout

figure() and subplots() can now be called with constrained_layout=True kwarg to enable constrained_layout.

New ax.set_position behaviour

set_position() now makes the specified axis no longer responsive to constrained_layout, consistent with the idea that the user wants to place an axis manually.

Internally, this means that old ax.set_position calls inside the library are changed to private ax._set_position calls so that constrained_layout will still work with these axes.

New figure kwarg for GridSpec

In order to facilitate constrained_layout, GridSpec now accepts a figure keyword. This is backwards compatible, in that not supplying this will simply cause constrained_layout to not operate on the subplots orgainzed by this GridSpec instance. Routines that use GridSpec (e.g. fig.subplots) have been modified to pass the figure to GridSpec.

xlabels and ylabels can now be automatically aligned

Subplot axes ylabels can be misaligned horizontally if the tick labels are very different widths. The same can happen to xlabels if the ticklabels are rotated on one subplot (for instance). The new methods on the Figure class: Figure.align_xlabels and Figure.align_ylabels will now align these labels horizontally or vertically. If the user only wants to align some axes, a list of axes can be passed. If no list is passed, the algorithm looks at all the labels on the figure.

Only labels that have the same subplot locations are aligned. i.e. the ylabels are aligned only if the subplots are in the same column of the subplot layout.

Alignemnt is persistent and automatic after these are called.

A convenience wrapper Figure.align_labels calls both functions at once.

(Source code, png, pdf)


Axes legends now included in tight_bbox

Legends created via ax.legend can sometimes overspill the limits of the axis. Tools like fig.tight_layout() and fig.savefig(bbox_inches='tight') would clip these legends. A change was made to include them in the tight calculations.

Cividis colormap

A new dark blue/yellow colormap named 'cividis' was added. Like viridis, cividis is perceptually uniform and colorblind friendly. However, cividis also goes a step further: not only is it usable by colorblind users, it should actually look effectively identical to colorblind and non-colorblind users. For more details see Nuñez J, Anderton C, and Renslow R: "Optimizing colormaps with consideration for color vision deficiency to enable accurate interpretation of scientific data".

(Source code, png, pdf)


New style colorblind-friendly color cycle

A new style defining a color cycle has been added, tableau-colorblind10, to provide another option for colorblind-friendly plots. A demonstration of this new style can be found in the reference of style sheets. To load this color cycle in place of the default one:

import matplotlib.pyplot as plt'tableau-colorblind10')

Support for numpy.datetime64

Matplotlib has supported datetime.datetime dates for a long time in matplotlib.dates. We now support numpy.datetime64 dates as well. Anywhere that dateime.datetime could be used, numpy.datetime64 can be used. eg:

time = np.arange('2005-02-01', '2005-02-02', dtype='datetime64[h]')

Writing animations with Pillow

It is now possible to use Pillow as an animation writer. Supported output formats are currently gif (Pillow>=3.4) and webp (Pillow>=5.0). Use e.g. as

from __future__ import division

from matplotlib import pyplot as plt
from matplotlib.animation import FuncAnimation, PillowWriter

fig, ax = plt.subplots()
line, = plt.plot([0, 1])

def animate(i):
   line.set_ydata([0, i / 20])
   return [line]

anim = FuncAnimation(fig, animate, 20, blit=True)"movie.gif", writer=PillowWriter(fps=24))

Slider UI widget can snap to discrete values

The slider UI widget can take the optional argument valstep. Doing so forces the slider to take on only discrete values, starting from valmin and counting up to valmax with steps of size valstep.

If closedmax==True, then the slider will snap to valmax as well.

capstyle and joinstyle attributes added to Collection

The Collection class now has customizable capstyle and joinstyle attributes. This allows the user for example to set the capstyle of errorbars.

pad kwarg added to ax.set_title

The method axes.set_title now has a pad kwarg, that specifies the distance from the top of an axes to where the title is drawn. The units of pad is points, and the default is the value of the (already-existing) rcParams['axes.titlepad'].

Comparison of 2 colors in Matplotlib

As the colors in Matplotlib can be specified with a wide variety of ways, the matplotlib.colors.same_color method has been added which checks if two colors are the same.

Autoscaling a polar plot snaps to the origin

Setting the limits automatically in a polar plot now snaps the radial limit to zero if the automatic limit is nearby. This means plotting from zero doesn't automatically scale to include small negative values on the radial axis.

The limits can still be set manually in the usual way using set_ylim.

PathLike support

On Python 3.6+, savefig, imsave, imread, and animation writers now accept os.PathLikes as input.

Axes.tick_params can set gridline properties

Tick objects hold gridlines as well as the tick mark and its label. Axis.set_tick_params, Axes.tick_params and pyplot.tick_params now have keyword arguments 'grid_color', 'grid_alpha', 'grid_linewidth', and 'grid_linestyle' for overriding the defaults in rcParams: 'grid.color', etc.

Axes.imshow clips RGB values to the valid range

When Axes.imshow is passed an RGB or RGBA value with out-of-range values, it now logs a warning and clips them to the valid range. The old behaviour, wrapping back in to the range, often hid outliers and made interpreting RGB images unreliable.

Properties in matplotlibrc to place xaxis and yaxis tick labels

Introducing four new boolean properties in matplotlibrc for default positions of xaxis and yaxis tick labels, namely, xtick.labeltop, xtick.labelbottom, ytick.labelright and ytick.labelleft. These can also be changed in rcParams.

PGI bindings for gtk3

The GTK3 backends can now use PGI instead of PyGObject. PGI is a fairly incomplete binding for GObject, thus its use is not recommended; its main benefit is its availability on Travis (thus allowing CI testing for the gtk3agg and gtk3cairo backends).

The binding selection rules are as follows: - if gi has already been imported, use it; else - if pgi has already been imported, use it; else - if gi can be imported, use it; else - if pgi can be imported, use it; else - error out.

Thus, to force usage of PGI when both bindings are installed, import it first.

Cairo rendering for Qt, WX, and Tk canvases

The new Qt4Cairo, Qt5Cairo, WXCairo, and TkCairo backends allow Qt, Wx, and Tk canvases to use Cairo rendering instead of Agg.

Added support for QT in new ToolManager

Now it is possible to use the ToolManager with Qt5 For example

import matplotlib

matplotlib.use('QT5AGG') matplotlib.rcParams['toolbar'] = 'toolmanager' import matplotlib.pyplot as plt


Treat the new Tool classes experimental for now, the API will likely change and perhaps the rcParam as well

The main example examples/user_interfaces/ shows more details, just adjust the header to use QT instead of GTK3

TkAgg backend reworked to support PyPy

PyPy can now plot using the TkAgg backend, supported on PyPy 5.9 and greater (both PyPy for python 2.7 and PyPy for python 3.5).

Python logging library used for debug output

Matplotlib has in the past (sporadically) used an internal verbose-output reporter. This version converts those calls to using the standard python logging library.

Support for the old rcParams verbose.level and verbose.fileo is dropped.

The command-line options --verbose-helpful and --verbose-debug are still accepted, but deprecated. They are now equivalent to setting logging.INFO and logging.DEBUG.

The logger's root name is matplotlib and can be accessed from programs as:

import logging
mlog = logging.getLogger('matplotlib')

Instructions for basic usage are in Troubleshooting and for developers in Contributing.

Improved repr for Transforms

Transforms now indent their reprs in a more legible manner:

In [1]: l, = plt.plot([]); l.get_transform()
               Bbox(x0=-0.05500000000000001, y0=-0.05500000000000001, x1=0.05500000000000001, y1=0.05500000000000001),
               Bbox(x0=0.125, y0=0.10999999999999999, x1=0.9, y1=0.88),
                        Bbox(x0=0.0, y0=0.0, x1=6.4, y1=4.8),
                           [[ 100.    0.    0.]
                           [   0.  100.    0.]
                           [   0.    0.    1.]])))))))