You are reading an old version of the documentation (v1.4.0). For the latest version see

Table Of Contents

Previous topic

Our Favorite Recipes

Next topic

Github stats

This Page

What’s new in matplotlib

This page just covers the highlights – for the full story, see the CHANGELOG

For a list of all of the issues and pull requests since the last revision, see the github-stats.


matplotlib 1.4 supports Python 2.6, 2.7, 3.3, and 3.4

matplotlib 1.3 supports Python 2.6, 2.7, 3.2, and 3.3

matplotlib 1.2 supports Python 2.6, 2.7, and 3.1

matplotlib 1.1 supports Python 2.4 to 2.7

Table of Contents

new in matplotlib-1.4

Thomas A. Caswell served as the release manager for the 1.4 release.

New colormap

In heatmaps, a green-to-red spectrum is often used to indicate intensity of activity, but this can be problematic for the red/green colorblind. A new, colorblind-friendly colormap is now available at This colormap maintains the red/green symbolism while achieving deuteranopic legibility through brightness variations. See here for more information.

Documentation changes

Phil Elson rewrote of the documentation and userguide for both Legend and PathEffects (links needed).

New plotting features

Power-law normalization

Ben Gamari added a power-law normalization method, PowerNorm. This class maps a range of values to the interval [0,1] with power-law scaling with the exponent provided by the constructor’s gamma argument. Power law normalization can be useful for, e.g., emphasizing small populations in a histogram.

Fully customizable boxplots

Paul Hobson overhauled the boxplot() method such that it is now completely customizable in terms of the styles and positions of the individual artists. Under the hood, boxplot() relies on a new function (boxplot_stats()), which accepts any data structure currently compatible with boxplot(), and returns a list of dictionaries containing the positions for each element of the boxplots. Then a second method, bxp() is called to draw the boxplots based on the stats.

The boxplot() function can be used as before to generate boxplots from data in one step. But now the user has the flexibility to generate the statistics independently, or to modify the output of boxplot_stats() prior to plotting with bxp().

Lastly, each artist (e.g., the box, outliers, cap, notches) can now be toggled on or off and their styles can be passed in through individual kwargs. See the examples: statistics example code: and statistics example code:

Added a bool kwarg, manage_xticks, which if False disables the management of the ticks and limits on the x-axis by bxp().

Support for datetime axes in 2d plots

Andrew Dawson added support for datetime axes to contour(), contourf(), pcolormesh() and pcolor().

Support for additional spectrum types

Todd Jennings added support for new types of frequency spectrum plots: magnitude_spectrum(), phase_spectrum(), and angle_spectrum(), as well as corresponding functions in mlab.

He also added these spectrum types to specgram(), as well as adding support for linear scaling there (in addition to the existing dB scaling). Support for additional spectrum types was also added to specgram().

He also increased the performance for all of these functions and plot types.

Support for detrending and windowing 2D arrays in mlab

Todd Jennings added support for 2D arrays in the detrend_mean(), detrend_none(), and detrend(), as well as adding apply_window() which support windowing 2D arrays.

Support for strides in mlab

Todd Jennings added some functions to mlab to make it easier to use numpy strides to create memory-efficient 2D arrays. This includes stride_repeat(), which repeats an array to create a 2D array, and stride_windows(), which uses a moving window to create a 2D array from a 1D array.

Formatter for new-style formatting strings

Added FormatStrFormatterNewStyle which does the same job as FormatStrFormatter, but accepts new-style formatting strings instead of printf-style formatting strings

Consistent grid sizes in streamplots

streamplot() uses a base grid size of 30x30 for both density=1 and density=(1, 1). Previously a grid size of 30x30 was used for density=1, but a grid size of 25x25 was used for density=(1, 1).

Get a list of all tick labels (major and minor)

Added the kwarg ‘which’ to get_xticklabels(), get_yticklabels() and get_ticklabels(). ‘which’ can be ‘major’, ‘minor’, or ‘both’ select which ticks to return, like set_ticks_position(). If ‘which’ is None then the old behaviour (controlled by the bool minor).

Separate horizontal/vertical axes padding support in ImageGrid

The kwarg ‘axes_pad’ to mpl_toolkits.axes_grid1.ImageGrid can now be a tuple if separate horizontal/vertical padding is needed. This is supposed to be very helpful when you have a labelled legend next to every subplot and you need to make some space for legend’s labels.

Support for skewed transformations

The Affine2D gained additional methods skew and skew_deg to create skewed transformations. Additionally, matplotlib internals were cleaned up to support using such transforms in Axes. This transform is important for some plot types, specifically the Skew-T used in meteorology.

(Source code, png, hires.png, pdf)


Support for specifying properties of wedge and text in pie charts.

Added the kwargs ‘wedgeprops’ and ‘textprops’ to pie() to accept properties for wedge and text objects in a pie. For example, one can specify wedgeprops = {‘linewidth’:3} to specify the width of the borders of the wedges in the pie. For more properties that the user can specify, look at the docs for the wedge and text objects.

Fixed the direction of errorbar upper/lower limits

Larry Bradley fixed the errorbar() method such that the upper and lower limits (lolims, uplims, xlolims, xuplims) now point in the correct direction.

More consistent add-object API for Axes

Added the Axes method add_image() to put image handling on a par with artists, collections, containers, lines, patches, and tables.

Violin Plots

Per Parker, Gregory Kelsie, Adam Ortiz, Kevin Chan, Geoffrey Lee, Deokjae Donald Seo, and Taesu Terry Lim added a basic implementation for violin plots. Violin plots can be used to represent the distribution of sample data. They are similar to box plots, but use a kernel density estimation function to present a smooth approximation of the data sample used. The added features are:

violin() - Renders a violin plot from a collection of statistics. violin_stats() - Produces a collection of statistics suitable for rendering a violin plot. violinplot() - Creates a violin plot from a set of sample data. This method makes use of violin_stats() to process the input data, and violin_stats() to do the actual rendering. Users are also free to modify or replace the output of violin_stats() in order to customize the violin plots to their liking.

This feature was implemented for a software engineering course at the University of Toronto, Scarborough, run in Winter 2014 by Anya Tafliovich.

More markevery options to show only a subset of markers

Rohan Walker extended the markevery property in Line2D. You can now specify a subset of markers to show with an int, slice object, numpy fancy indexing, or float. Using a float shows markers at approximately equal display-coordinate-distances along the line.

Fixed the mouse coordinates giving the wrong theta value in Polar graph

Added code to transform_non_affine() to ensure that the calculated theta value was between the range of 0 and 2 * pi since the problem was that the value can become negative after applying the direction and rotation to the theta calculation.

Simple quiver plot for mplot3d toolkit

A team of students in an Engineering Large Software Systems course, taught by Prof. Anya Tafliovich at the University of Toronto, implemented a simple version of a quiver plot in 3D space for the mplot3d toolkit as one of their term project. This feature is documented in quiver(). The team members are: Ryan Steve D’Souza, Victor B, xbtsw, Yang Wang, David, Caradec Bisesar and Vlad Vassilovski.

(Source code, png, hires.png, pdf)


polar-plot r-tick locations

Added the ability to control the angular position of the r-tick labels on a polar plot via set_rlabel_position().

Date handling

n-d array support for date conversion

Andrew Dawson added support for n-d array handling to matplotlib.dates.num2date(), matplotlib.dates.date2num() and matplotlib.dates.datestr2num(). Support is also added to the unit conversion interfaces matplotlib.dates.DateConverter and matplotlib.units.Registry.

Configuration (rcParams)

savefig.transparent added

Controls whether figures are saved with a transparent background by default. Previously savefig always defaulted to a non-transparent background.


Added rcParam to control the weight of the title

axes.formatter.useoffset added

Controls the default value of useOffset in ScalarFormatter. If True and the data range is much smaller than the data average, then an offset will be determined such that the tick labels are meaningful. If False then the full number will be formatted in all conditions.

XDG compliance

Matplotlib now looks for configuration files (both rcparams and style) in XDG compliant locations.

style package added

You can now easily switch between different styles using the new style package:

>>> from matplotlib import style
>>> style.use('dark_background')

Subsequent plots will use updated colors, sizes, etc. To list all available styles, use:

>>> print style.available

You can add your own custom <style name>.mplstyle files to ~/.matplotlib/stylelib or call use with a URL pointing to a file with matplotlibrc settings.

Note that this is an experimental feature, and the interface may change as users test out this new feature.


Qt5 backend

Martin Fitzpatrick and Tom Badran implemented a Qt5 backend. The differences in namespace locations between Qt4 and Qt5 was dealt with by shimming Qt4 to look like Qt5, thus the Qt5 implementation is the primary implementation. Backwards compatibility for Qt4 is maintained by wrapping the Qt5 implementation.

The Qt5Agg backend currently does not work with IPython’s %matplotlib magic.

Qt4 backend

Rudolf Höfler changed the appearance of the subplottool. All sliders are vertically arranged now, buttons for tight layout and reset were added. Furthermore, the the subplottool is now implemented as a modal dialog. It was previously a QMainWindow, leaving the SPT open if one closed the plot window.

In the figure options dialog one can now choose to (re-)generate a simple automatic legend. Any explicitly set legend entries will be lost, but changes to the curves’ label, linestyle, et cetera will now be updated in the legend.

Interactive performance of the Qt4 backend has been dramatically improved under windows.

The mapping of key-signals from Qt to values matplotlib understands was greatly improved (For both Qt4 and Qt5).

Cairo backends

The Cairo backends are now able to use the cairocffi bindings which are more actively maintained than the pycairo bindings.

Gtk3Agg backend

The Gtk3Agg backend now works on Python 3.x, if the cairocffi bindings are installed.

PDF backend

Added context manager for saving to multi-page PDFs.


Text URLs supported by SVG backend

The svg backend will now render Text objects’ url as a link in output SVGs. This allows one to make clickable text in saved figures using the url kwarg of the Text class.

Anchored sizebar font

Added the fontproperties kwarg to AnchoredSizeBar to control the font properties.

Sphinx extensions

The :context: directive in the plot_directive Sphinx extension can now accept an optional reset setting, which will cause the context to be reset. This allows more than one distinct context to be present in documentation. To enable this option, use :context: reset instead of :context: any time you want to reset the context.


Span Selector

Added an option span_stays to the SpanSelector which makes the selector rectangle stay on the axes after you release the mouse.

GAE integration

Matplotlib will now run on google app engine.

new in matplotlib-1.3

New in 1.3.1

1.3.1 is a bugfix release, primarily dealing with improved setup and handling of dependencies, and correcting and enhancing the documentation.

The following changes were made in 1.3.1 since 1.3.0.


Bug fixes

  • Histogram plots now contain the endline.
  • Fixes to the Molleweide projection.
  • Handling recent fonts from Microsoft and Macintosh-style fonts with non-ascii metadata is improved.
  • Hatching of fill between plots now works correctly in the PDF backend.
  • Tight bounding box support now works in the PGF backend.
  • Transparent figures now display correctly in the Qt4Agg backend.
  • Drawing lines from one subplot to another now works.
  • Unit handling on masked arrays has been improved.

Setup and dependencies

  • Now works with any version of pyparsing 1.5.6 or later, without displaying hundreds of warnings.
  • Now works with 64-bit versions of Ghostscript on MS-Windows.
  • When installing from source into an environment without Numpy, Numpy will first be downloaded and built and then used to build matplotlib.
  • Externally installed backends are now always imported using a fully-qualified path to the module.
  • Works with newer version of wxPython.
  • Can now build with a PyCXX installed globally on the system from source.
  • Better detection of Gtk3 dependencies.


  • Tests should now work in non-English locales.
  • PEP8 conformance tests now report on locations of issues.

New plotting features

xkcd-style sketch plotting

To give your plots a sense of authority that they may be missing, Michael Droettboom (inspired by the work of many others in PR #1329) has added an xkcd-style sketch plotting mode. To use it, simply call matplotlib.pyplot.xkcd() before creating your plot. For really fine control, it is also possible to modify each artist’s sketch parameters individually with matplotlib.artist.Artist.set_sketch_params().

(Source code)

Updated Axes3D.contour methods

Damon McDougall updated the tricontour() and tricontourf() methods to allow 3D contour plots on abitrary unstructured user-specified triangulations.

(Source code, png, hires.png, pdf)


New eventplot plot type

Todd Jennings added a eventplot() function to create multiple rows or columns of identical line segments

(Source code, png, hires.png, pdf)


As part of this feature, there is a new EventCollection class that allows for plotting and manipulating rows or columns of identical line segments.

Triangular grid interpolation

Geoffroy Billotey and Ian Thomas added classes to perform interpolation within triangular grids: (LinearTriInterpolator and CubicTriInterpolator) and a utility class to find the triangles in which points lie (TrapezoidMapTriFinder). A helper class to perform mesh refinement and smooth contouring was also added (UniformTriRefiner). Finally, a class implementing some basic tools for triangular mesh improvement was added (TriAnalyzer).

(Source code, png, hires.png, pdf)


Baselines for stackplot

Till Stensitzki added non-zero baselines to stackplot(). They may be symmetric or weighted.

(Source code, png, hires.png, pdf)


Rectangular colorbar extensions

Andrew Dawson added a new keyword argument extendrect to colorbar() to optionally make colorbar extensions rectangular instead of triangular.

More robust boxplots

Paul Hobson provided a fix to the boxplot() method that prevent whiskers from being drawn inside the box for oddly distributed data sets.

Calling subplot() without arguments

A call to subplot() without any arguments now acts the same as subplot(111) or subplot(1,1,1) – it creates one axes for the whole figure. This was already the behavior for both axes() and subplots(), and now this consistency is shared with subplot().


Independent alpha values for face and edge colors

Wes Campaigne modified how Patch objects are drawn such that (for backends supporting transparency) you can set different alpha values for faces and edges, by specifying their colors in RGBA format. Note that if you set the alpha attribute for the patch object (e.g. using set_alpha() or the alpha keyword argument), that value will override the alpha components set in both the face and edge colors.

Path effects on lines

Thanks to Jae-Joon Lee, path effects now also work on plot lines.

(Source code, png, hires.png, pdf)


Easier creation of colormap and normalizer for levels with colors

Phil Elson added the matplotlib.colors.from_levels_and_colors() function to easily create a colormap and normalizer for representation of discrete colors for plot types such as matplotlib.pyplot.pcolormesh(), with a similar interface to that of contourf().

Full control of the background color

Wes Campaigne and Phil Elson fixed the Agg backend such that PNGs are now saved with the correct background color when fig.patch.get_alpha() is not 1.

Improved bbox_inches="tight" functionality

Passing bbox_inches="tight" through to now takes into account all artists on a figure - this was previously not the case and led to several corner cases which did not function as expected.

Initialize a rotated rectangle

Damon McDougall extended the Rectangle constructor to accept an angle kwarg, specifying the rotation of a rectangle in degrees.


Anchored text support

The svg and pgf backends are now able to save text alignment information to their output formats. This allows to edit text elements in saved figures, using Inkscape for example, while preserving their intended position. For svg please note that you’ll have to disable the default text-to-path conversion (mpl.rc('svg', fonttype='none')).

Better vertical text alignment and multi-line text

The vertical alignment of text is now consistent across backends. You may see small differences in text placement, particularly with rotated text.

If you are using a custom backend, note that the draw_text renderer method is now passed the location of the baseline, not the location of the bottom of the text bounding box.

Multi-line text will now leave enough room for the height of very tall or very low text, such as superscripts and subscripts.

Left and right side axes titles

Andrew Dawson added the ability to add axes titles flush with the left and right sides of the top of the axes using a new keyword argument loc to title().

Improved manual contour plot label positioning

Brian Mattern modified the manual contour plot label positioning code to interpolate along line segments and find the actual closest point on a contour to the requested position. Previously, the closest path vertex was used, which, in the case of straight contours was sometimes quite distant from the requested location. Much more precise label positioning is now possible.

Configuration (rcParams)

Quickly find rcParams

Phil Elson made it easier to search for rcParameters by passing a valid regular expression to matplotlib.RcParams.find_all(). matplotlib.RcParams now also has a pretty repr and str representation so that search results are printed prettily:

>>> import matplotlib
>>> print(matplotlib.rcParams.find_all('\.size'))
RcParams({'font.size': 12,
          'xtick.major.size': 4,
          'xtick.minor.size': 2,
          'ytick.major.size': 4,
          'ytick.minor.size': 2})

axes.xmargin and axes.ymargin added to rcParams

rcParam values (axes.xmargin and axes.ymargin) were added to configure the default margins used. Previously they were hard-coded to default to 0, default value of both rcParam values is 0.

Changes to font rcParams

The font.* rcParams now affect only text objects created after the rcParam has been set, and will not retroactively affect already existing text objects. This brings their behavior in line with most other rcParams.

savefig.jpeg_quality added to rcParams

rcParam value savefig.jpeg_quality was added so that the user can configure the default quality used when a figure is written as a JPEG. The default quality is 95; previously, the default quality was 75. This change minimizes the artifacting inherent in JPEG images, particularly with images that have sharp changes in color as plots often do.


WebAgg backend

Michael Droettboom, Phil Elson and others have developed a new backend, WebAgg, to display figures in a web browser. It works with animations as well as being fully interactive.


Future versions of matplotlib will integrate this backend with the IPython notebook for a fully web browser based plotting frontend.

Remember save directory

Martin Spacek made the save figure dialog remember the last directory saved to. The default is configurable with the new rcParam in matplotlibrc.

Documentation and examples

Numpydoc docstrings

Nelle Varoquaux has started an ongoing project to convert matplotlib’s docstrings to numpydoc format. See MEP10 for more information.

Example reorganization

Tony Yu has begun work reorganizing the examples into more meaningful categories. The new gallery page is the fruit of this ongoing work. See MEP12 for more information.

Examples now use subplots()

For the sake of brevity and clarity, most of the examples now use the newer subplots(), which creates a figure and one (or multiple) axes object(s) in one call. The old way involved a call to figure(), followed by one (or multiple) subplot() calls.



A number of features that were deprecated in 1.2 or earlier, or have not been in a working state for a long time have been removed. Highlights include removing the Qt version 3 backends, and the FltkAgg and Emf backends. See Changes in 1.3.x for a complete list.

New setup script

matplotlib 1.3 includes an entirely rewritten setup script. We now ship fewer dependencies with the tarballs and installers themselves. Notably, pytz, dateutil, pyparsing and six are no longer included with matplotlib. You can either install them manually first, or let pip install them as dependencies along with matplotlib. It is now possible to not include certain subcomponents, such as the unit test data, in the install. See setup.cfg.template for more information.

XDG base directory support

On Linux, matplotlib now uses the XDG base directory specification to find the matplotlibrc configuration file. matplotlibrc should now be kept in config/matplotlib, rather than matplotlib. If your configuration is found in the old location, it will still be used, but a warning will be displayed.

Catch opening too many figures using pyplot

Figures created through pyplot.figure are retained until they are explicitly closed. It is therefore common for new users of matplotlib to run out of memory when creating a large series of figures in a loop without closing them.

matplotlib will now display a RuntimeWarning when too many figures have been opened at once. By default, this is displayed for 20 or more figures, but the exact number may be controlled using the figure.max_num_figures rcParam.

new in matplotlib 1.2.2

Improved collections

The individual items of a collection may now have different alpha values and be rendered correctly. This also fixes a bug where collections were always filled in the PDF backend.

Multiple images on same axes are correctly transparent

When putting multiple images onto the same axes, the background color of the axes will now show through correctly.

new in matplotlib-1.2

Python 3.x support

Matplotlib 1.2 is the first version to support Python 3.x, specifically Python 3.1 and 3.2. To make this happen in a reasonable way, we also had to drop support for Python versions earlier than 2.6.

This work was done by Michael Droettboom, the Cape Town Python Users’ Group, many others and supported financially in part by the SAGE project.

The following GUI backends work under Python 3.x: Gtk3Cairo, Qt4Agg, TkAgg and MacOSX. The other GUI backends do not yet have adequate bindings for Python 3.x, but continue to work on Python 2.6 and 2.7, particularly the Qt and QtAgg backends (which have been deprecated). The non-GUI backends, such as PDF, PS and SVG, work on both Python 2.x and 3.x.

Features that depend on the Python Imaging Library, such as JPEG handling, do not work, since the version of PIL for Python 3.x is not sufficiently mature.

PGF/TikZ backend

Peter Würtz wrote a backend that allows matplotlib to export figures as drawing commands for LaTeX. These can be processed by PdfLaTeX, XeLaTeX or LuaLaTeX using the PGF/TikZ package. Usage examples and documentation are found in Typesetting With XeLaTeX/LuaLaTeX.


Locator interface

Philip Elson exposed the intelligence behind the tick Locator classes with a simple interface. For instance, to get no more than 5 sensible steps which span the values 10 and 19.5:

>>> import matplotlib.ticker as mticker
>>> locator = mticker.MaxNLocator(nbins=5)
>>> print(locator.tick_values(10, 19.5))
[ 10.  12.  14.  16.  18.  20.]

Tri-Surface Plots

Damon McDougall added a new plotting method for the mplot3d toolkit called plot_trisurf().

(Source code, png, hires.png, pdf)


Control the lengths of colorbar extensions

Andrew Dawson added a new keyword argument extendfrac to colorbar() to control the length of minimum and maximum colorbar extensions.

(Source code, png, hires.png, pdf)


Figures are picklable

Philip Elson added an experimental feature to make figures picklable for quick and easy short-term storage of plots. Pickle files are not designed for long term storage, are unsupported when restoring a pickle saved in another matplotlib version and are insecure when restoring a pickle from an untrusted source. Having said this, they are useful for short term storage for later modification inside matplotlib.

Set default bounding box in matplotlibrc

Two new defaults are available in the matplotlibrc configuration file: savefig.bbox, which can be set to ‘standard’ or ‘tight’, and savefig.pad_inches, which controls the bounding box padding.

New Boxplot Functionality

Users can now incorporate their own methods for computing the median and its confidence intervals into the boxplot() method. For every column of data passed to boxplot, the user can specify an accompanying median and confidence interval.

(Source code, png, hires.png, pdf)


New RC parameter functionality

Matthew Emmett added a function and a context manager to help manage RC parameters: rc_file() and rc_context. To load RC parameters from a file:

>>> mpl.rc_file('mpl.rc')

To temporarily use RC parameters:

>>> with mpl.rc_context(fname='mpl.rc', rc={'text.usetex': True}):
>>>     ...


Tom Flannaghan and Tony Yu have added a new streamplot() function to plot the streamlines of a vector field. This has been a long-requested feature and complements the existing quiver() function for plotting vector fields. In addition to simply plotting the streamlines of the vector field, streamplot() allows users to map the colors and/or line widths of the streamlines to a separate parameter, such as the speed or local intensity of the vector field.

(Source code)

New hist functionality

Nic Eggert added a new stacked kwarg to hist() that allows creation of stacked histograms using any of the histogram types. Previously, this functionality was only available by using the barstacked histogram type. Now, when stacked=True is passed to the function, any of the histogram types can be stacked. The barstacked histogram type retains its previous functionality for backwards compatibility.

Updated shipped dependencies

The following dependencies that ship with matplotlib and are optionally installed alongside it have been updated:

  • pytz 2012d
  • dateutil 1.5 on Python 2.x, and 2.1 on Python 3.x

Face-centred colors in tripcolor plots

Ian Thomas extended tripcolor() to allow one color value to be specified for each triangular face rather than for each point in a triangulation.

(Source code)

Hatching patterns in filled contour plots, with legends

Phil Elson added support for hatching to contourf(), together with the ability to use a legend to identify contoured ranges.

(Source code)

Known issues in the matplotlib-1.2 release

  • When using the Qt4Agg backend with IPython 0.11 or later, the save dialog will not display. This should be fixed in a future version of IPython.

new in matplotlib-1.1

Sankey Diagrams

Kevin Davies has extended Yannick Copin’s original Sankey example into a module (sankey) and provided new examples (api example code:, api example code:, api example code:

(Source code, png, hires.png, pdf)



Ryan May has written a backend-independent framework for creating animated figures. The animation module is intended to replace the backend-specific examples formerly in the Matplotlib Examples listings. Examples using the new framework are in animation Examples; see the entrancing double pendulum which uses to create the movie below.

This should be considered as a beta release of the framework; please try it and provide feedback.

Tight Layout

A frequent issue raised by users of matplotlib is the lack of a layout engine to nicely space out elements of the plots. While matplotlib still adheres to the philosophy of giving users complete control over the placement of plot elements, Jae-Joon Lee created the tight_layout module and introduced a new command tight_layout() to address the most common layout issues.

(Source code, png, hires.png, pdf)


(png, hires.png, pdf)


The usage of this functionality can be as simple as


and it will adjust the spacing between subplots so that the axis labels do not overlap with neighboring subplots. A Tight Layout guide has been created to show how to use this new tool.

PyQT4, PySide, and IPython

Gerald Storer made the Qt4 backend compatible with PySide as well as PyQT4. At present, however, PySide does not support the PyOS_InputHook mechanism for handling gui events while waiting for text input, so it cannot be used with the new version 0.11 of IPython. Until this feature appears in PySide, IPython users should use the PyQT4 wrapper for QT4, which remains the matplotlib default.

An rcParam entry, “backend.qt4”, has been added to allow users to select PyQt4, PyQt4v2, or PySide. The latter two use the Version 2 Qt API. In most cases, users can ignore this rcParam variable; it is available to aid in testing, and to provide control for users who are embedding matplotlib in a PyQt4 or PySide app.


Jae-Joon Lee has improved plot legends. First, legends for complex plots such as stem() plots will now display correctly. Second, the ‘best’ placement of a legend has been improved in the presence of NANs.

See legend-complex-plots for more detailed explanation and examples.

(Source code, png, hires.png, pdf)



In continuing the efforts to make 3D plotting in matplotlib just as easy as 2D plotting, Ben Root has made several improvements to the mplot3d module.

  • Axes3D has been improved to bring the class towards feature-parity with regular Axes objects
  • Documentation for mplot3d was significantly expanded
  • Axis labels and orientation improved
  • Most 3D plotting functions now support empty inputs
  • Ticker offset display added:

(Source code, png, hires.png, pdf)

  • contourf() gains zdir and offset kwargs. You can now do this:

(Source code, png, hires.png, pdf)


Numerix support removed

After more than two years of deprecation warnings, Numerix support has now been completely removed from matplotlib.


The list of available markers for plot() and scatter() has now been merged. While they were mostly similar, some markers existed for one function, but not the other. This merge did result in a conflict for the ‘d’ diamond marker. Now, ‘d’ will be interpreted to always mean “thin” diamond while ‘D’ will mean “regular” diamond.

Thanks to Michael Droettboom for this effort.

Other improvements

  • Unit support for polar axes and arrow()
  • PolarAxes gains getters and setters for “theta_direction”, and “theta_offset” to allow for theta to go in either the clock-wise or counter-clockwise direction and to specify where zero degrees should be placed. set_theta_zero_location() is an added convenience function.
  • Fixed error in argument handling for tri-functions such as tripcolor()
  • axes.labelweight parameter added to rcParams.
  • For imshow(), interpolation=’nearest’ will now always perform an interpolation. A “none” option has been added to indicate no interpolation at all.
  • An error in the Hammer projection has been fixed.
  • clabel for contour() now accepts a callable. Thanks to Daniel Hyams for the original patch.
  • Jae-Joon Lee added the HBox and VBox classes.
  • Christoph Gohlke reduced memory usage in imshow().
  • scatter() now accepts empty inputs.
  • The behavior for ‘symlog’ scale has been fixed, but this may result in some minor changes to existing plots. This work was refined by ssyr.
  • Peter Butterworth added named figure support to figure().
  • Michiel de Hoon has modified the MacOSX backend to make its interactive behavior consistent with the other backends.
  • Pim Schellart added a new colormap called “cubehelix”. Sameer Grover also added a colormap called “coolwarm”. See it and all other colormaps here.
  • Many bug fixes and documentation improvements.

new in matplotlib-1.0

HTML5/Canvas backend

Simon Ratcliffe and Ludwig Schwardt have released an HTML5/Canvas backend for matplotlib. The backend is almost feature complete, and they have done a lot of work comparing their html5 rendered images with our core renderer Agg. The backend features client/server interactive navigation of matplotlib figures in an html5 compliant browser.

Sophisticated subplot grid layout

Jae-Joon Lee has written gridspec, a new module for doing complex subplot layouts, featuring row and column spans and more. See Customizing Location of Subplot Using GridSpec for a tutorial overview.

(Source code, png, hires.png, pdf)


Easy pythonic subplots

Fernando Perez got tired of all the boilerplate code needed to create a figure and multiple subplots when using the matplotlib API, and wrote a subplots() helper function. Basic usage allows you to create the figure and an array of subplots with numpy indexing (starts with 0). e.g.:

fig, axarr = plt.subplots(2, 2)
axarr[0,0].plot([1,2,3])   # upper, left

See pylab_examples example code: for several code examples.

Contour fixes and and triplot

Ian Thomas has fixed a long-standing bug that has vexed our most talented developers for years. contourf() now handles interior masked regions, and the boundaries of line and filled contours coincide.

Additionally, he has contributed a new module tri and helper function triplot() for creating and plotting unstructured triangular grids.

(Source code)

multiple calls to show supported

A long standing request is to support multiple calls to show(). This has been difficult because it is hard to get consistent behavior across operating systems, user interface toolkits and versions. Eric Firing has done a lot of work on rationalizing show across backends, with the desired behavior to make show raise all newly created figures and block execution until they are closed. Repeated calls to show should raise newly created figures since the last call. Eric has done a lot of testing on the user interface toolkits and versions and platforms he has access to, but it is not possible to test them all, so please report problems to the mailing list and bug tracker.

mplot3d graphs can be embedded in arbitrary axes

You can now place an mplot3d graph into an arbitrary axes location, supporting mixing of 2D and 3D graphs in the same figure, and/or multiple 3D graphs in a single figure, using the “projection” keyword argument to add_axes or add_subplot. Thanks Ben Root.

(Source code, png, hires.png, pdf)



Eric Firing wrote tick_params, a convenience method for changing the appearance of ticks and tick labels. See pyplot function tick_params() and associated Axes method tick_params().

Lots of performance and feature enhancements

  • Faster magnification of large images, and the ability to zoom in to a single pixel
  • Local installs of documentation work better
  • Improved “widgets” – mouse grabbing is supported
  • More accurate snapping of lines to pixel boundaries
  • More consistent handling of color, particularly the alpha channel, throughout the API

Much improved software carpentry

The matplotlib trunk is probably in as good a shape as it has ever been, thanks to improved software carpentry. We now have a buildbot which runs a suite of nose regression tests on every svn commit, auto-generating a set of images and comparing them against a set of known-goods, sending emails to developers on failures with a pixel-by-pixel image comparison. Releases and release bugfixes happen in branches, allowing active new feature development to happen in the trunk while keeping the release branches stable. Thanks to Andrew Straw, Michael Droettboom and other matplotlib developers for the heavy lifting.

Bugfix marathon

Eric Firing went on a bug fixing and closing marathon, closing over 100 bugs on the bug tracker with help from Jae-Joon Lee, Michael Droettboom, Christoph Gohlke and Michiel de Hoon.

new in matplotlib-0.99

New documentation

Jae-Joon Lee has written two new guides Legend guide and Annotating Axes. Michael Sarahan has written Image tutorial. John Hunter has written two new tutorials on working with paths and transformations: Path Tutorial and Transformations Tutorial.


Reinier Heeres has ported John Porter’s mplot3d over to the new matplotlib transformations framework, and it is now available as a toolkit mpl_toolkits.mplot3d (which now comes standard with all mpl installs). See mplot3d Examples and mplot3d tutorial

(Source code, png, hires.png, pdf)


axes grid toolkit

Jae-Joon Lee has added a new toolkit to ease displaying multiple images in matplotlib, as well as some support for curvilinear grids to support the world coordinate system. The toolkit is included standard with all new mpl installs. See axes_grid Examples and The Matplotlib AxesGrid Toolkit User’s Guide.

(Source code, png, hires.png, pdf)


Axis spine placement

Andrew Straw has added the ability to place “axis spines” – the lines that denote the data limits – in various arbitrary locations. No longer are your axis lines constrained to be a simple rectangle around the figure – you can turn on or off left, bottom, right and top, as well as “detach” the spine to offset it away from the data. See pylab_examples example code: and matplotlib.spines.Spine.

(Source code, png, hires.png, pdf)


new in 0.98.4

It’s been four months since the last matplotlib release, and there are a lot of new features and bug-fixes.

Thanks to Charlie Moad for testing and preparing the source release, including binaries for OS X and Windows for python 2.4 and 2.5 (2.6 and 3.0 will not be available until numpy is available on those releases). Thanks to the many developers who contributed to this release, with contributions from Jae-Joon Lee, Michael Droettboom, Ryan May, Eric Firing, Manuel Metz, Jouni K. Seppänen, Jeff Whitaker, Darren Dale, David Kaplan, Michiel de Hoon and many others who submitted patches

Legend enhancements

Jae-Joon has rewritten the legend class, and added support for multiple columns and rows, as well as fancy box drawing. See legend() and matplotlib.legend.Legend.

(Source code, png, hires.png, pdf)


Fancy annotations and arrows

Jae-Joon has added lots of support to annotations for drawing fancy boxes and connectors in annotations. See annotate() and BoxStyle, ArrowStyle, and ConnectionStyle.

(Source code, png, hires.png, pdf)


Native OS X backend

Michiel de Hoon has provided a native Mac OSX backend that is almost completely implemented in C. The backend can therefore use Quartz directly and, depending on the application, can be orders of magnitude faster than the existing backends. In addition, no third-party libraries are needed other than Python and NumPy. The backend is interactive from the usual terminal application on Mac using regular Python. It hasn’t been tested with ipython yet, but in principle it should to work there as well. Set ‘backend : macosx’ in your matplotlibrc file, or run your script with:

> python -dmacosx

psd amplitude scaling

Ryan May did a lot of work to rationalize the amplitude scaling of psd() and friends. See pylab_examples example code: and pylab_examples example code: The changes should increase MATLAB compatibility and increase scaling options.

Fill between

Added a fill_between() function to make it easier to do shaded region plots in the presence of masked data. You can pass an x array and a ylower and yupper array to fill between, and an optional where argument which is a logical mask where you want to do the filling.

(Source code, png, hires.png, pdf)


Lots more

Here are the 0.98.4 notes from the CHANGELOG:

Added mdehoon's native macosx backend from sf patch 2179017 - JDH

Removed the prints in the set_*style commands.  Return the list of
pretty-printed strings instead - JDH

Some of the changes Michael made to improve the output of the
property tables in the rest docs broke of made difficult to use
some of the interactive doc helpers, eg setp and getp.  Having all
the rest markup in the ipython shell also confused the docstrings.
I added a new rc param docstring.harcopy, to format the docstrings
differently for hardcopy and other use.  The ArtistInspector
could use a little refactoring now since there is duplication of
effort between the rest out put and the non-rest output - JDH

Updated spectral methods (psd, csd, etc.) to scale one-sided
densities by a factor of 2 and, optionally, scale all densities by
the sampling frequency.  This gives better MATLAB
compatibility. -RM

Fixed alignment of ticks in colorbars. -MGD

drop the deprecated "new" keyword of np.histogram() for numpy 1.2
or later.  -JJL

Fixed a bug in svg backend that new_figure_manager() ignores
keywords arguments such as figsize, etc. -JJL

Fixed a bug that the handlelength of the new legend class set too
short when numpoints=1 -JJL

Added support for data with units (e.g., dates) to
Axes.fill_between. -RM

Added fancybox keyword to legend. Also applied some changes for
better look, including baseline adjustment of the multiline texts
so that it is center aligned. -JJL

The transmuter classes in the are reorganized as
subclasses of the Style classes. A few more box and arrow styles
are added. -JJL

Fixed a bug in the new legend class that didn't allowed a tuple of
coordinate values as loc. -JJL

Improve checks for external dependencies, using subprocess
(instead of deprecated popen*) and distutils (for version
checking) - DSD

Reimplementation of the legend which supports baseline alignment,
multi-column, and expand mode. - JJL

Fixed histogram autoscaling bug when bins or range are given
explicitly (fixes Debian bug 503148) - MM

Added rcParam axes.unicode_minus which allows plain hyphen for
minus when False - JDH

Added scatterpoints support in Legend. patch by Erik Tollerud -

Fix crash in log ticking. - MGD

Added static helper method BrokenHBarCollection.span_where and
Axes/pyplot method fill_between.  See
examples/pylab/ - JDH

Add x_isdata and y_isdata attributes to Artist instances, and use
them to determine whether either or both coordinates are used when
updating dataLim.  This is used to fix autoscaling problems that
had been triggered by axhline, axhspan, axvline, axvspan. - EF

Update the psd(), csd(), cohere(), and specgram() methods of Axes
and the csd() cohere(), and specgram() functions in mlab to be in
sync with the changes to psd().  In fact, under the hood, these
all call the same core to do computations. - RM

Add 'pad_to' and 'sides' parameters to mlab.psd() to allow
controlling of zero padding and returning of negative frequency
components, respectively.  These are added in a way that does not
change the API. - RM

Fix handling of c kwarg by scatter; generalize is_string_like to
accept numpy and string array scalars. - RM and EF

Fix a possible EINTR problem in dviread, which might help when
saving pdf files from the qt backend. - JKS

Fix bug with zoom to rectangle and twin axes - MGD

Added Jae Joon's fancy arrow, box and annotation enhancements --
see examples/pylab_examples/

Autoscaling is now supported with shared axes - EF

Fixed exception in dviread that happened with Minion - JKS

set_xlim, ylim now return a copy of the viewlim array to avoid
modify inplace surprises

Added image thumbnail generating function
matplotlib.image.thumbnail.  See examples/misc/

Applied scatleg patch based on ideas and work by Erik Tollerud and
Jae-Joon Lee. - MM

Fixed bug in pdf backend: if you pass a file object for output
instead of a filename, e.g., in a wep app, we now flush the object
at the end. - JKS

Add path simplification support to paths with gaps. - EF

Fix problem with AFM files that don't specify the font's full name
or family name. - JKS

Added 'scilimits' kwarg to Axes.ticklabel_format() method, for
easy access to the set_powerlimits method of the major
ScalarFormatter. - EF

Experimental new kwarg borderpad to replace pad in legend, based
on suggestion by Jae-Joon Lee.  - EF

Allow spy to ignore zero values in sparse arrays, based on patch
by Tony Yu.  Also fixed plot to handle empty data arrays, and
fixed handling of markers in figlegend. - EF

Introduce drawstyles for lines. Transparently split linestyles
like 'steps--' into drawstyle 'steps' and linestyle '--'.  Legends
always use drawstyle 'default'. - MM

Fixed quiver and quiverkey bugs (failure to scale properly when
resizing) and added additional methods for determining the arrow
angles - EF

Fix polar interpolation to handle negative values of theta - MGD

Reorganized cbook and mlab methods related to numerical
calculations that have little to do with the goals of those two
modules into a separate module Also, added
ability to select points and stop point selection with keyboard in
ginput and manual contour labeling code.  Finally, fixed contour
labeling bug. - DMK

Fix backtick in Postscript output. - MGD

[ 2089958 ] Path simplification for vector output backends
Leverage the simplification code exposed through path_to_polygons
to simplify certain well-behaved paths in the vector backends
(PDF, PS and SVG).  "path.simplify" must be set to True in
matplotlibrc for this to work.  - MGD

Add "filled" kwarg to Path.intersects_path and
Path.intersects_bbox. - MGD

Changed full arrows slightly to avoid an xpdf rendering problem
reported by Friedrich Hagedorn. - JKS

Fix conversion of quadratic to cubic Bezier curves in PDF and PS
backends. Patch by Jae-Joon Lee. - JKS

Added 5-point star marker to plot command q- EF

Fix hatching in PS backend - MGD

Fix log with base 2 - MGD

Added support for bilinear interpolation in
NonUniformImage; patch by Gregory Lielens. - EF

Added support for multiple histograms with data of
different length - MM

Fix step plots with log scale - MGD

Fix masked arrays with markers in non-Agg backends - MGD

Fix clip_on kwarg so it actually works correctly - MGD

Fix locale problems in SVG backend - MGD

fix quiver so masked values are not plotted - JSW

improve interactive pan/zoom in qt4 backend on windows - DSD

Fix more bugs in NaN/inf handling.  In particular, path
simplification (which does not handle NaNs or infs) will be turned
off automatically when infs or NaNs are present.  Also masked
arrays are now converted to arrays with NaNs for consistent
handling of masks and NaNs - MGD and EF

Added support for arbitrary rasterization resolutions to the SVG
backend. - MW