You are reading documentation for the unreleased version of Matplotlib. Try searching for the released version of this page instead?
Version 2.1.1.post1111+gac2da0e
Fork me on GitHub

Table Of Contents

This Page

What’s new in Matplotlib

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

Table of Contents

What’s new in unreleased Matplotlib?

Please place new portions of whats_new.rst in the next_whats_new directory.

When adding an entry please look at the currently existing files to see if you can extend any of them. If you create a file, name it something like cool_new_feature.rst if you have added a brand new feature or something like updated_feature.rst for extensions of existing features. Include contents of the form:

Section Title for Feature

A bunch of text about how awesome the new feature is and examples of how
to use it.

A sub-section

New in Matplotlib 2.1


The examples have been migrated to use sphinx gallery. This allows better mixing of prose and code in the examples, provides links to download the examples as both a Python script and a Jupyter notebook, and improves the thumbnail galleries. The examples have been re-organized into Tutorials and a Gallery.

Many docstrings and examples have been clarified and improved.

New features

String categorical values

All plotting functions now support string categorical values as input. For example:

data = {'apples': 10, 'oranges': 15, 'lemons': 5, 'limes': 20}
fig, ax = plt.subplots(), data.values(), color='lightgray')

(Source code, png, pdf)


Interactive JS widgets for animation

Jake Vanderplas’ JSAnimation package has been merged into Matplotlib. This adds to Matplotlib the HTMLWriter class for generating a JavaScript HTML animation, suitable for the IPython notebook. This can be activated by default by setting the animation.html rc parameter to jshtml. One can also call the to_jshtml method to manually convert an animation. This can be displayed using IPython’s HTML display class:

from IPython.display import HTML

The HTMLWriter class can also be used to generate an HTML file by asking for the html writer.

Enhancements to polar plot

The polar axes transforms have been greatly re-factored to allow for more customization of view limits and tick labelling. Additional options for view limits allow for creating an annulus, a sector, or some combination of the two.

The set_rorigin() method may be used to provide an offset to the minimum plotting radius, producing an annulus.

The set_theta_zero_location() method now has an optional offset argument. This argument may be used to further specify the zero location based on the given anchor point.


Polar Offset Demo

The set_thetamin() and set_thetamax() methods may be used to limit the range of angles plotted, producing sectors of a circle.


Polar Sector Demo

Previous releases allowed plots containing negative radii for which the negative values are simply used as labels, and the real radius is shifted by the configured minimum. This release also allows negative radii to be used for grids and ticks, which were previously silently ignored.

Radial ticks have been modified to be parallel to the circular grid line, and angular ticks have been modified to be parallel to the grid line. It may also be useful to rotate tick labels to match the boundary. Calling ax.tick_params(rotation='auto') will enable the new behavior: radial tick labels will be parallel to the circular grid line, and angular tick labels will be perpendicular to the grid line (i.e., parallel to the outer boundary). Additionally, tick labels now obey the padding settings that previously only worked on Cartesian plots. Consequently, the frac argument to PolarAxes.set_thetagrids is no longer applied. Tick padding can be modified with the pad argument to Axes.tick_params or Axis.set_tick_params.

Figure class now has subplots method

The Figure class now has a subplots() method which behaves the same as pyplot.subplots() but on an existing figure.

Metadata savefig keyword argument

savefig() now accepts metadata as a keyword argument. It can be used to store key/value pairs in the image metadata.

  • ‘png’ with Agg backend
  • ‘pdf’ with PDF backend (see writeInfoDict() for a list of supported keywords)
  • ‘eps’ and ‘ps’ with PS backend (only ‘Creator’ key is accepted)
plt.savefig('test.png', metadata={'Software': 'My awesome software'})

Busy Cursor

The interactive GUI backends will now change the cursor to busy when Matplotlib is rendering the canvas.


A PolygonSelector class has been added to matplotlib.widgets. See Polygon Selector Demo for details.

Added matplotlib.ticker.PercentFormatter

The new PercentFormatter formatter has some nice features like being able to convert from arbitrary data scales to percents, a customizable percent symbol and either automatic or manual control over the decimal points.

Reproducible PS, PDF and SVG output

The SOURCE_DATE_EPOCH environment variable can now be used to set the timestamp value in the PS and PDF outputs. See source date epoch.

Alternatively, calling savefig with metadata={'creationDate': None} will omit the timestamp altogether for the PDF backend.

The reproducibility of the output from the PS and PDF backends has so far been tested using various plot elements but only default values of options such as {ps,pdf}.fonttype that can affect the output at a low level, and not with the mathtext or usetex features. When Matplotlib calls external tools (such as PS distillers or LaTeX) their versions need to be kept constant for reproducibility, and they may add sources of nondeterminism outside the control of Matplotlib.

For SVG output, the svg.hashsalt rc parameter has been added in an earlier release. This parameter changes some random identifiers in the SVG file to be deterministic. The downside of this setting is that if more than one file is generated using deterministic identifiers and they end up as parts of one larger document, the identifiers can collide and cause the different parts to affect each other.

These features are now enabled in the tests for the PDF and SVG backends, so most test output files (but not all of them) are now deterministic.

Orthographic projection for mplot3d

Axes3D now accepts proj_type keyword argument and has a method set_proj_type(). The default option is 'persp' as before, and supplying 'ortho' enables orthographic view.

Compare the z-axis which is vertical in orthographic view, but slightly skewed in the perspective view.

import numpy as np
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D

fig = plt.figure(figsize=(4, 6))
ax1 = fig.add_subplot(2, 1, 1, projection='3d')
ax1.set_title('Perspective (default)')

ax2 = fig.add_subplot(2, 1, 2, projection='3d')

(Source code, png, pdf)


voxels function for mplot3d

Axes3D now has a voxels method, for visualizing boolean 3D data. Uses could include plotting a sparse 3D heat map, or visualizing a volumetric model.


Voxel Demo


CheckButtons widget get_status function

A get_status() method has been added to the matplotlib.widgets.CheckButtons class. This get_status method allows user to query the status (True/False) of all of the buttons in the CheckButtons object.

Add fill_bar argument to AnchoredSizeBar

The mpl_toolkits class AnchoredSizeBar now has an additional fill_bar argument, which makes the size bar a solid rectangle instead of just drawing the border of the rectangle. The default is None, and whether or not the bar will be filled by default depends on the value of size_vertical. If size_vertical is nonzero, fill_bar will be set to True. If size_vertical is zero then fill_bar will be set to False. If you wish to override this default behavior, set fill_bar to True or False to unconditionally always or never use a filled patch rectangle for the size bar.

import matplotlib.pyplot as plt
from mpl_toolkits.axes_grid1.anchored_artists import AnchoredSizeBar

fig, ax = plt.subplots(figsize=(3, 3))

bar0 = AnchoredSizeBar(ax.transData, 0.3, 'unfilled', loc=3, frameon=False,
                       size_vertical=0.05, fill_bar=False)
bar1 = AnchoredSizeBar(ax.transData, 0.3, 'filled', loc=4, frameon=False,
                       size_vertical=0.05, fill_bar=True)

(Source code, png, pdf)


Annotation can use a default arrow style

Annotations now use the default arrow style when setting arrowprops={}, rather than no arrow (the new behavior actually matches the documentation).

Barbs and Quiver Support Dates

When using the quiver() and barbs() plotting methods, it is now possible to pass dates, just like for other methods like plot(). This also allows these functions to handle values that need unit-conversion applied.

Hexbin default line color

The default linecolor keyword argument for hexbin() is now 'face', and supplying 'none' now prevents lines from being drawn around the hexagons.

Figure.legend() can be called without arguments

Calling Figure.legend() can now be done with no arguments. In this case a legend will be created that contains all the artists on all the axes contained within the figure.

Multiple legend keys for legend entries

A legend entry can now contain more than one legend key. The extended HandlerTuple class now accepts two parameters: ndivide divides the legend area in the specified number of sections; pad changes the padding between the legend keys.


Multiple Legend Keys

New parameter clear for figure()

When the pyplot’s function figure() is called with a num parameter, a new window is only created if no existing window with the same value exists. A new bool parameter clear was added for explicitly clearing its existing contents. This is particularly useful when utilized in interactive sessions. Since subplots() also accepts keyword arguments from figure(), it can also be used there:

import matplotlib.pyplot as plt

fig0 = plt.figure(num=1)
fig0.suptitle("A fancy plot")
print("fig0.texts: ", [t.get_text() for t in fig0.texts])

fig1 = plt.figure(num=1, clear=False)  # do not clear contents of window
fig1.text(0.5, 0.5, "Really fancy!")
print("fig0 is fig1: ",  fig0 is fig1)
print("fig1.texts: ", [t.get_text() for t in fig1.texts])

fig2, ax2 = plt.subplots(2, 1, num=1, clear=True)  # clear contents
print("fig0 is fig2: ",  fig0 is fig2)
print("fig2.texts: ", [t.get_text() for t in fig2.texts])

# The output:
# fig0.texts:  ['A fancy plot']
# fig0 is fig1:  True
# fig1.texts:  ['A fancy plot', 'Really fancy!']
# fig0 is fig2:  True
# fig2.texts:  []

Specify minimum value to format as scalar for LogFormatterMathtext

LogFormatterMathtext now includes the option to specify a minimum value exponent to format as a scalar (i.e., 0.001 instead of 10-3).

New quiverkey angle keyword argument

Plotting a quiverkey() now admits the angle keyword argument, which sets the angle at which to draw the key arrow.

Colormap reversed method

The methods matplotlib.colors.LinearSegmentedColormap.reversed() and matplotlib.colors.ListedColormap.reversed() return a reversed instance of the Colormap. This implements a way for any Colormap to be reversed.

Artist.setp (and pyplot.setp) accept a file argument

The argument is keyword-only. It allows an output file other than sys.stdout to be specified. It works exactly like the file argument to print.

streamplot streamline generation more configurable

The starting point, direction, and length of the stream lines can now be configured. This allows to follow the vector field for a longer time and can enhance the visibility of the flow pattern in some use cases.

Axis.set_tick_params now responds to rotation

Bulk setting of tick label rotation is now possible via set_tick_params() using the rotation keyword.

ax.xaxis.set_tick_params(which='both', rotation=90)

Shading in 3D bar plots

A new shade parameter has been added the 3D bar plotting method. The default behavior remains to shade the bars, but now users have the option of setting shade to False.

import numpy as np
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D

x = np.arange(2)
y = np.arange(3)
x2d, y2d = np.meshgrid(x, y)
x, y = x2d.ravel(), y2d.ravel()
z = np.zeros_like(x)
dz = x + y

fig = plt.figure(figsize=(4, 6))
ax1 = fig.add_subplot(2, 1, 1, projection='3d')
ax1.bar3d(x, y, z, 1, 1, dz, shade=True)
ax1.set_title('Shading On')

ax2 = fig.add_subplot(2, 1, 2, projection='3d')
ax2.bar3d(x, y, z, 1, 1, dz, shade=False)
ax2.set_title('Shading Off')

(Source code, png, pdf)


New which Parameter for autofmt_xdate

A which parameter now exists for the method autofmt_xdate(). This allows a user to format major, minor or both tick labels selectively. The default behavior will rotate and align the major tick labels.

fig.autofmt_xdate(bottom=0.2, rotation=30, ha='right', which='minor')

New Figure Parameter for subplot2grid

A fig parameter now exists for the function subplot2grid(). This allows a user to specify the figure where the subplots will be created. If fig is None (default) then the method will use the current figure retrieved by gcf().

subplot2grid(shape, loc, rowspan=1, colspan=1, fig=myfig)

Interpolation in fill_betweenx

The interpolate parameter now exists for the method fill_betweenx(). This allows a user to interpolate the data and fill the areas in the crossover points, similarly to fill_between().

New keyword argument sep for EngFormatter

A new sep keyword argument has been added to EngFormatter and provides a means to define the string that will be used between the value and its unit. The default string is " ", which preserves the former behavior. Additionally, the separator is now present between the value and its unit even in the absence of SI prefix. There was formerly a bug that was causing strings like "3.14V" to be returned instead of the expected "3.14 V" (with the default behavior).

Extend MATPLOTLIBRC behavior

The environmental variable can now specify the full file path or the path to a directory containing a matplotlibrc file.

density kwarg to hist

The hist() method now prefers density to normed to control if the histogram should be normalized, following a change upstream to NumPy. This will reduce confusion as the behavior has always been that the integral of the histogram is 1 (rather than sum or maximum value).


New TransformedPatchPath caching object

A newly added TransformedPatchPath provides a means to transform a Patch into a Path via a Transform while caching the resulting path. If neither the patch nor the transform have changed, a cached copy of the path is returned.

This class differs from the older TransformedPath in that it is able to refresh itself based on the underlying patch while the older class uses an immutable path.

Abstract base class for movie writers

The new AbstractMovieWriter class defines the API required by a class that is to be used as the writer in the method. The existing MovieWriter class now derives from the new abstract base class.

Stricter validation of line style rcParams

The validation of rcParams that are related to line styles (lines.linestyle, boxplot.*.linestyle, grid.linestyle and contour.negative_linestyle) now effectively checks that the values are valid line styles. Strings like 'dashed' or '--' are accepted, as well as even-length sequences of on-off ink like [1, 1.65]. In this latter case, the offset value is handled internally and should not be provided by the user.

The new validation scheme replaces the former one used for the contour.negative_linestyle rcParams, that was limited to 'solid' and 'dashed' line styles.

The validation is case-insensitive. The following are now valid:

grid.linestyle             : (1, 3)   # loosely dotted grid lines
contour.negative_linestyle : dashdot  # previously only solid or dashed


The automated tests have been switched from nose to pytest.


Path simplification updates

Line simplification controlled by the path.simplify and path.simplify_threshold parameters has been improved. You should notice better rendering performance when plotting large amounts of data (as long as the above parameters are set accordingly). Only the line segment portion of paths will be simplified – if you are also drawing markers and experiencing problems with rendering speed, you should consider using the markevery option to plot. See the Performance section in the usage tutorial for more information.

The simplification works by iteratively merging line segments into a single vector until the next line segment’s perpendicular distance to the vector (measured in display-coordinate space) is greater than the path.simplify_threshold parameter. Thus, higher values of path.simplify_threshold result in quicker rendering times. If you are plotting just to explore data and not for publication quality, pixel perfect plots, then a value of 1.0 can be safely used. If you want to make sure your plot reflects your data exactly, then you should set path.simplify to false and/or path.simplify_threshold to 0. Matplotlib currently defaults to a conservative value of 1/9, smaller values are unlikely to cause any visible differences in your plots.

Implement intersects_bbox in c++

intersects_bbox() has been implemented in c++ which improves the performance of automatically placing the legend.