- Behavior changes
pgisupport dropped- rcParam changes
- Exception changes
- Removals
- :mod:`matplotlib.mlab` removals
- pylab removals
- mplot3d changes
- Testing
- Dependency changes
- Mathtext changes
- Signature deprecations
- Changes in parameter names
- Class/method/attribute deprecations
- Undeprecations
- New features
- Invalid inputs
Switching backends via matplotlib.use is now allowed by default, regardless of whether matplotlib.pyplot has been imported. If the user tries to switch from an already-started interactive backend to a different interactive backend, an ImportError will be raised.
PathCollections created with ~.Axes.scatter now keep track of invalid points. Previously, points with nonfinite (infinite or nan) coordinates would not be included in the offsets (as returned by .PathCollection.get_offsets) of a .PathCollection created by ~.Axes.scatter, and points with nonfinite values (as specified by the c kwarg) would not be included in the array (as returned by .PathCollection.get_array)
Such points are now included, but masked out by returning a masked array.
If the plotnonfinite kwarg to ~.Axes.scatter is set, then points with nonfinite values are plotted using the bad color of the .collections.PathCollection's colormap (as set by :meth:`.colors.Colormap.set_bad`).
The alpha-channel of RBGA images is now re-sampled independently of RGB channels. While this is a bug fix, it does change the output and may result in some down-stream image comparison tests to fail.
On log-axes where a single value is plotted at a "full" decade (1, 10, 100, etc.), the autoscaling now expands the axis symmetrically around that point, instead of adding a decade only to the right.
When the default .LogLocator would generate no ticks for an axis (e.g., an axis with limits from 0.31 to 0.39) or only a single tick, it now instead falls back on the linear .AutoLocator to pick reasonable tick positions.
Calling .Figure.add_subplot() with no positional arguments used to do
nothing; this now is equivalent to calling add_subplot(111) instead.
~.Axes.bxp now respects :rc:`boxplot.boxprops.linewidth` even when patch_artist is set. Previously, when the patch_artist parameter was set, ~.Axes.bxp would ignore :rc:`boxplot.boxprops.linewidth`. This was an oversight -- in particular, ~.Axes.boxplot did not ignore it.
Minor ticks that collide with major ticks are now hidden by default. Previously, certain locator classes (~.ticker.LogLocator, ~.ticker.AutoMinorLocator) contained custom logic to avoid emitting tick locations that collided with major ticks when they were used as minor locators. This logic has now moved to the ~.axis.Axis class, and is used regardless of the locator class. You can control this behavior via the ~.Axis.remove_overlapping_locs attribute on ~.axis.Axis.
If you were relying on both the major and minor tick labels to appear on the same tick, you may need to update your code. For example, the following snippet
import numpy as np
import matplotlib.dates as mdates
import matplotlib.pyplot as plt
t = np.arange("2018-11-03", "2018-11-06", dtype="datetime64")
x = np.random.rand(len(t))
fig, ax = plt.subplots()
ax.plot(t, x)
ax.xaxis.set(
major_locator=mdates.DayLocator(),
major_formatter=mdates.DateFormatter("\n%a"),
minor_locator=mdates.HourLocator((0, 6, 12, 18)),
minor_formatter=mdates.DateFormatter("%H:%M"),
)
# disable removing overlapping locations
ax.xaxis.remove_overlapping_locs = False
plt.show()
labeled days using major ticks, and hours and minutes using minor ticks and added a newline to the major ticks labels to avoid them crashing into the minor tick labels. Setting the ~.Axis.remove_overlapping_locs property (also accessible via ~.Axis.set_remove_overlapping_locs / ~.Axis.get_remove_overlapping_locs and ~.pyplot.setp) disables removing overlapping tick locations.
The major tick labels could also be adjusted include hours and
minutes, as the minor ticks are gone, so the major_formatter
would be:
mdates.DateFormatter("%H:%M\n%a")
Previously, if :rc:`text.usetex` was True, then constructing a .TextPath on
a non-mathtext string with usetex=False would rely on the mathtext parser
(but not on usetex support!) to parse the string. The mathtext parser is not
invoked anymore, which may cause slight changes in glyph positioning.
.matplotlib.axes.Axes.get_window_extent used to return a bounding box that was slightly larger than the axes, presumably to take into account the ticks that may be on a spine. However, it was not scaling the tick sizes according to the dpi of the canvas, and it did not check if the ticks were visible, or on the spine.
Now .matplotlib.axes.Axes.get_window_extent just returns the axes extent with no padding for ticks.
This affects .matplotlib.axes.Axes.get_tightbbox in cases where there are outward ticks with no tick labels, and it also removes the (small) pad around axes in that case.
.spines.Spine.get_window_extent now takes into account ticks that are on the spine.
Previously, .Sankey.add would only accept a single string as the labels argument if its length is equal to the number of flows, in which case it would use one character of the string for each flow.
The behavior has been changed to match the documented one: when a single string is passed, it is used to label all the flows.
.font_manager.FontManager.score_weight is now more strict with its inputs. Previously, when a weight string was passed to .font_manager.FontManager.score_weight,
- if the weight was the string representation of an integer, it would be converted to that integer,
- otherwise, if the weight was not a standard weight name, it would be silently replaced by a value of 500 ("normal" weight).
.font_manager.FontManager.score_weight now raises an exception on such inputs.
Text alignment was previously incorrect, in particular for multiline text objects with large descenders (i.e. subscripts) and rotated text. These have been fixed and made more consistent, but could make old code that has compensated for this no longer have the correct alignment.
Support for passing single-letter colors (one of "rgbcmykw") as UPPERCASE characters is deprecated; these colors will become case-sensitive (lowercase) after the deprecation period has passed.
The goal is to decrease the number of ambiguous cases when using the data
keyword to plotting methods; e.g. plot("X", "Y", data={"X": ..., "Y": ...})
will not warn about "Y" possibly being a color anymore after the deprecation
period has passed.
When bounds passed to ~.axes.Axes.set_xlim are degenerate (i.e. the lower and upper value are equal), the method used to "expand" the bounds now matches the expansion behavior of autoscaling when the plot contains a single x-value, and should in particular produce nicer limits for non-linear scales.
In certain cases, ~.Axes.plot would previously accept format strings
specifying more than one linestyle (e.g. "---." which specifies both
"--" and "-."); only use one of them would be used. This now raises a
ValueError instead.
The HTMLWriter constructor is more strict: it no longer normalizes unknown values of default_mode to 'loop', but errors out instead.
In accordance with the AFM spec, the AFM parser no longer truncates the
UnderlinePosition and UnderlineThickness fields to integers.
The Notice field (which can only be publicly accessed by the deprecated
afm.parse_afm API) is no longer decoded to a str, but instead kept as
bytes, to support non-conformant AFM files that use non-ASCII characters in
that field.
.Artist.set now normalizes keywords before sorting them. Previously it sorted
its keyword arguments in reverse alphabetical order (with a special-case to
put color at the end) before applying them.
It now normalizes aliases (and, as above, emits a warning on duplicate
properties) before doing the sorting (so c goes to the end too).
Previously .Axes.tick_params silently did nothing when an invalid axis parameter was supplied. This behavior has been changed to raise a ValueError instead.
Input that consists of multiple empty lists will now return a list of histogram
values for each one of the lists. For example, an input of [[],[]] will
return 2 lists of histogram values. Previously, a single list was returned.
Currently, backend_bases.TimerBase.remove_callback(func, *args,
**kwargs) removes a callback previously added by
backend_bases.Timer.add_callback(func, *args, **kwargs), but if
*args, **kwargs is not passed in (i.e.,
TimerBase.remove_callback(func)), then the first callback with a
matching func is removed, regardless of whether it was added with
or without *args, **kwargs.
In a future version, .TimerBase.remove_callback will always use the latter
behavior (not consider *args, **kwargs); to specifically consider them, add
the callback as a functools.partial object
cb = timer.add_callback(functools.partial(func, *args, **kwargs)) # ... # later timer.remove_callback(cb)
.TimerBase.add_callback was modified to return func to simplify the above usage (previously it returned None); this also allows using it as a decorator.
The new API is modelled after atexit.register / atexit.unregister.
~.container.StemContainer objects can now store a ~.collections.LineCollection object instead of a list of ~.lines.Line2D objects for stem lines plotted using ~.Axes.stem. This gives a very large performance boost to displaying and moving ~.Axes.stem plots.
This will become the default behaviour in Matplotlib 3.3. To use it now, the use_line_collection keyword argument to ~.Axes.stem can be set to True
ax.stem(..., use_line_collection=True)
Individual line segments can be extracted from the ~.collections.LineCollection using ~.collections.LineCollection.get_segments(). See the ~.collections.LineCollection documentation for other methods to retrieve the collection properties.
matplotlib.colorbar.ColorbarBase is no longer a subclass of
.cm.ScalarMappable. This inheritance lead to a confusing situation
where the .cm.ScalarMappable passed to matplotlib.colorbar.Colorbar
(~.Figure.colorbar) had a set_norm method, as did the colorbar.
The colorbar is now purely a follower to the .ScalarMappable norm and
colormap, and the old inherited methods
~matplotlib.colorbar.ColorbarBase.set_norm,
~matplotlib.colorbar.ColorbarBase.set_cmap,
~matplotlib.colorbar.ColorbarBase.set_clim are deprecated, as are
the getter versions of those calls. To set the norm associated with a
colorbar do colorbar.mappable.set_norm() etc.
The MPLBASEDIRLIST environment variables and basedirlist entry in
setup.cfg have no effect anymore. Instead, if building in situations where
FreeType or libpng are not in the compiler or linker's default path, set the
standard environment variables CFLAGS/LDFLAGS on Linux or OSX, or
CL/LINK on Windows, to indicate the relevant paths.
See details in :doc:`/users/installing`.
Setting the same artist property multiple time via aliases is deprecated. Previously, code such as
plt.plot([0, 1], c="red", color="blue")
would emit a warning indicating that c and color are aliases
of one another, and only keep the color kwarg. This behavior has
been deprecated; in a future version, this will raise a TypeError,
similar to Python's behavior when a keyword argument is passed twice
plt.plot([0, 1], c="red", c="blue")
This warning is raised by ~.cbook.normalize_kwargs.
Path code types like Path.MOVETO are now np.uint8 instead of int
Path.STOP, Path.MOVETO, Path.LINETO, Path.CURVE3,
Path.CURVE4 and Path.CLOSEPOLY are now of the type Path.code_type
(np.uint8 by default) instead of plain int. This makes their type
match the array value type of the Path.codes array.
Previously, the rc file keys pgf.preamble and text.latex.preamble were
parsed using commmas as separators. This would break valid LaTeX code, such as:
\usepackage[protrusion=true, expansion=false]{microtype}
The parsing has been modified to pass the complete line to the LaTeX system, keeping all commas. Passing a list of strings from within a Python script still works as it used to. Passing a list containing non-strings now fails, instead of coercing the results to strings.
The method .Axes.spy now raises a TypeError for the keyword arguments interpolation and linestyle instead of silently ignoring them.
Furthermore, .Axes.spy spy does now allow for an extent argument (was silently ignored so far).
A bug with Axes.spy(..., origin='lower') is fixed. Previously this
flipped the data but not the y-axis resulting in a mismatch between
axes labels and actual data indices. Now, origin='lower' flips both
the data and the y-axis labels.
The manage_xticks parameter of ~.Axes.boxplot and ~.Axes.bxp has been renamed (with a deprecation period) to manage_ticks, to take into account the fact that it manages either x or y ticks depending on the vert parameter.
When manage_ticks=True (the default), these methods now attempt to
take previously drawn boxplots into account when setting the axis
limits, ticks, and tick labels.
MouseEvents now include the event name in their str().
Previously they contained the prefix "MPL MouseEvent".
.FigureCanvasAgg.buffer_rgba and .RendererAgg.buffer_rgba now
return a memoryview The buffer_rgba method now allows direct
access to the renderer's underlying buffer (as a (m, n, 4)-shape
memoryview) rather than copying the data to a new bytestring. This is
consistent with the behavior on Py2, where a buffer object was
returned.
matplotlib.font_manager.win32InstalledFonts returns an empty list instead of None if no fonts are found.
Previously, if the user provided a .Axes.fmt_xdata or .Axes.fmt_ydata function that raised a TypeError (or set them to a non-callable), the exception would be silently ignored and the default formatter be used instead. This is no longer the case; the exception is now propagated out.
The gridOn, tick1On, tick2On, label1On, and label2On
~.Tick attributes have been deprecated. Directly get and set the visibility
on the underlying artists, available as the gridline, tick1line,
tick2line, label1, and label2 attributes.
The label attribute, which was an alias for label1, has been
deprecated.
Subclasses that relied on setting the above visibility attributes needs to be updated; see e.g. :file:`examples/api/skewt.py`.
Instead of plt.plot(..., linestyle="steps--"), use plt.plot(...,
linestyle="--", drawstyle="steps"). ds is now an alias for drawstyle.
Support for pgi in the GTK3 backends has been dropped. pgi is
an alternative implementation to PyGObject. PyGObject should
be used instead.
The following deprecated rcParams have been removed:
text.dvipnghacknbagg.transparent(use :rc:`figure.facecolor` instead)plugins.directoryaxes.holdbackend.qt4andbackend.qt5(set the :envvar:`QT_API` environment variable instead)
The associated validator functions rcsetup.validate_qt4 and
validate_qt5 are deprecated.
The verbose.fileo and verbose.level rcParams have been deprecated.
These have had no effect since the switch from Matplotlib's old custom Verbose
logging to the stdlib's logging module. In addition the
rcsetup.validate_verbose function is deprecated.
The text.latex.unicode rcParam now defaults to True and is
deprecated (i.e., in future versions
of Matplotlib, unicode input will always be supported).
Moreover, the underlying implementation now uses \usepackage[utf8]{inputenc}
instead of \usepackage{ucs}\usepackage[utf8x]{inputenc}.
- mpl_toolkits.axes_grid1.axes_size.GetExtentHelper now raises ValueError for invalid directions instead of KeyError.
- Previously, subprocess failures in the animation framework would raise either in a RuntimeError or a ValueError depending on when the error occurred. They now raise a subprocess.CalledProcessError with attributes set as documented by the exception class.
- In certain cases, Axes methods (and pyplot functions) used to raise
a RuntimeError if they were called with a
datakwarg and otherwise mismatched arguments. They now raise a TypeError instead. - .Axes.streamplot does not support irregularly gridded
xandyvalues. So far, it used to silently plot an incorrect result. This has been changed to raise a ValueError instead. - The .streamplot.Grid class, which is internally used by streamplot code, also throws a ValueError when irregularly gridded values are passed in.
The following deprecated APIs have been removed:
Verbose(replaced by python logging library)artist.Artist.hitlist(no replacement)artist.Artist.is_figure_set(useartist.figure is not Noneinstead)axis.Axis.unit_data(useaxis.Axis.unitsinstead)backend_bases.FigureCanvasBase.onRemove(no replacement)backend_bases.FigureManagerBase.show_popup(this never did anything)backend_wx.SubplotToolWx(no replacement)backend_wx.Toolbar(usebackend_wx.NavigationToolbar2Wxinstead)cbook.align_iterators(no replacement)contour.ContourLabeler.get_real_label_width(no replacement)legend.Legend.draggable(use legend.Legend.set_draggable() instead)texmanager.TexManager.postscriptd,texmanager.TexManager.pscnt,texmanager.TexManager.make_ps,texmanager.TexManager.get_ps_bbox(no replacements)
- The fig kwarg to .GridSpec.get_subplot_params and .GridSpecFromSubplotSpec.get_subplot_params (use the argument figure instead)
- Passing 'box-forced' to .Axes.set_adjustable (use 'box' instead)
- Support for the strings 'on'/'true'/'off'/'false' to mean
True / False (directly use True / False instead).
The following functions are affected:
- .axes.Axes.grid
- .Axes3D.grid
- .Axis.set_tick_params
- .pyplot.box
- Using .pyplot.axes with an .axes.Axes type argument (use .pyplot.sca instead)
The following miscellaneous API elements have been removed
svgfont support (in :rc:`svg.fonttype`)
Logging is now done with the standard python
logginglibrary.matplotlib.verboseand the command line switches--verbose-LEVELhave been removed.To control the logging output use:
import logging logger = logging.getLogger('matplotlib') logger.setLevel(logging.INFO) # configure log handling: Either include it into your ``logging`` hierarchy, # e.g. by configuring a root looger using ``logging.basicConfig()``, # or add a standalone handler to the matplotlib logger: logger.addHandler(logging.StreamHandler())__version__numpy__collections.CIRCLE_AREA_FACTORfont_manager.USE_FONTCONFIGfont_manager.cachedir
:mod:`matplotlib.mlab` removals
Lots of code inside the :mod:`matplotlib.mlab` module which was deprecated in Matplotlib 2.2 has been removed. See below for a list:
mlab.exp_safe(use numpy.exp instead)mlab.amapmlab.logspace(use numpy.logspace instead)mlab.rms_flatmlab.l1norm(usenumpy.linalg.norm(a, ord=1)instead)mlab.l2norm(usenumpy.linalg.norm(a, ord=2)instead)mlab.norm_flat(usenumpy.linalg.norm(a.flat, ord=2)instead)mlab.frange(use numpy.arange instead)mlab.identity(use numpy.identity instead)mlab.base_reprmlab.binary_reprmlab.ispower2mlab.log2(use numpy.log2 instead)mlab.isvectormlab.movavgmlab.safe_isinf(use numpy.isinf instead)mlab.safe_isnan(use numpy.isnan instead)mlab.cohere_pairs(use scipy.signal.coherence instead)mlab.entropy(use scipy.stats.entropy instead)mlab.normpdf(use scipy.stats.norm.pdf instead)mlab.find(usenp.nonzero(np.ravel(condition))instead)mlab.longest_contiguous_onesmlab.longest_onesmlab.PCAmlab.prctile(use numpy.percentile instead)mlab.prctile_rankmlab.center_matrixmlab.rk4(use scipy.integrate.ode instead)mlab.bivariate_normalmlab.get_xyz_wheremlab.get_sparse_matrixmlab.dist(use numpy.hypot instead)mlab.dist_point_to_segmentmlab.griddata(use scipy.interpolate.griddata)mlab.less_simple_linear_interpolation(use numpy.interp)mlab.slopesmlab.stineman_interpmlab.segments_intersectmlab.fftsurrmlab.offset_linemlab.quad2cubicmlab.vector_lengthsmlab.distances_along_curvemlab.path_lengthmlab.cross_from_abovemlab.cross_from_belowmlab.contiguous_regions(use .cbook.contiguous_regions instead)mlab.is_closed_polygonmlab.poly_betweenmlab.poly_belowmlab.inside_polymlab.csv2recmlab.rec2csv(use numpy.recarray.tofile instead)mlab.rec2text(use numpy.recarray.tofile instead)mlab.rec_summarizemlab.rec_joinmlab.recs_joinmlab.rec_groupbymlab.rec_keep_fieldsmlab.rec_drop_fieldsmlab.rec_append_fieldsmlab.csvformat_factorymlab.get_formatdmlab.FormatDatetime(use datetime.datetime.strftime instead)mlab.FormatDate(use datetime.date.strftime instead)mlab.FormatMillions,mlab.FormatThousands,mlab.FormatPercent,mlab.FormatBool,mlab.FormatInt,mlab.FormatFloat,mlab.FormatFormatStr,mlab.FormatString,mlab.FormatObjmlab.donothing_callback
Lots of code inside the :mod:`matplotlib.mlab` module which was deprecated in Matplotlib 2.2 has been removed. This means the following functions are no longer available in the pylab module:
amapbase_reprbinary_reprbivariate_normalcenter_matrixcsv2rec(use numpy.recarray.tofile instead)dist(use numpy.hypot instead)dist_point_to_segmentdistances_along_curveentropy(use scipy.stats.entropy instead)exp_safe(use numpy.exp instead)fftsurrfind(usenp.nonzero(np.ravel(condition))instead)frange(use numpy.arange instead)get_sparse_matrixget_xyz_wheregriddata(use scipy.interpolate.griddata instead)identity(use numpy.identity instead)inside_polyis_closed_polygonispower2isvectorl1norm(usenumpy.linalg.norm(a, ord=1)instead)l2norm(usenumpy.linalg.norm(a, ord=2)instead)log2(use numpy.log2 instead)longest_contiguous_oneslongest_onesmovavgnorm_flat(usenumpy.linalg.norm(a.flat, ord=2)instead)normpdf(use scipy.stats.norm.pdf instead)path_lengthpoly_belowpoly_betweenprctile(use numpy.percentile instead)prctile_rankrec2csv(use numpy.recarray.tofile instead)rec_append_fieldsrec_drop_fieldsrec_joinrk4(use scipy.integrate.ode instead)rms_flatsegments_intersectslopesstineman_interpvector_lengths
.Axes3D.voxels now shades the resulting voxels; for more details see What's new. The previous behavior can be achieved by passing
ax.voxels(.., shade=False)
Setting the aspect on 3D axes previously returned non-sensical results
(e.g. see :ghissue:`1077`). Calling ax.set_aspect('equal') or
ax.set_aspect(num) on a 3D axes now raises a
NotImplementedError.
.Poly3DCollection.set_zsort no longer silently ignores invalid
inputs, or False (which was always broken). Passing True to mean
"average" is deprecated.
The --no-network flag to tests.py has been removed (no test requires
internet access anymore). If it is desired to disable internet access both for
old and new versions of Matplotlib, use tests.py -m 'not network' (which is
now a no-op).
The image comparison test decorators now skip (rather than xfail) the test for uncomparable formats. The affected decorators are ~.image_comparison and ~.check_figures_equal. The deprecated ~.ImageComparisonTest class is likewise changed.
Matplotlib 3.1 now requires NumPy>=1.11.
Support for ghostscript 8.60 (released in 2007) has been removed. The oldest supported version of ghostscript is now 9.0 (released in 2010).
- In constructs such as
"$1~2$", mathtext now interprets the tilde as a space, consistently with TeX (this was previously a parse error).
- The
\stackrelmathtext command has been deprecated (it behaved differently from LaTeX's\stackrel. To stack two mathtext expressions, use\genfrac{left-delim}{right-delim}{fraction-bar-thickness}{}{top}{bottom}. - The
\mathcircledmathtext command (which is not a real TeX command) is deprecated. Directly use unicode characters (e.g."\N{CIRCLED LATIN CAPITAL LETTER A}"or"\u24b6") instead. - Support for setting :rc:`mathtext.default` to circled is deprecated.
The following signature related behaviours are deprecated:
- The withdash keyword argument to .Axes.text(). Consider using .Axes.annotate() instead.
- Passing (n, 1)-shaped error arrays to .Axes.errorbar(), which was not
documented and did not work for
n = 2. Pass a 1D array instead. - The frameon kwarg to ~.Figure.savefig and the :rc:`savefig.frameon` rcParam.
To emulate
frameon = False, set facecolor to fully transparent ("none", or(0, 0, 0, 0)). - Passing a non-1D (typically, (n, 1)-shaped) input to .Axes.pie. Pass a 1D array instead.
- The .TextPath constructor used to silently drop ignored arguments; this behavior is deprecated.
- The usetex parameter of .TextToPath.get_text_path is deprecated and
folded into the ismath parameter, which can now take the values
False, True, and
"TeX", consistently with other low-level text processing functions. - Passing
'normal'to .axes.Axes.axis() is deprecated, useax.axis('auto')instead. - Passing the block argument of .pyplot.show positionally is deprecated; it should be passed by keyword.
- When using the nbagg backend, .pyplot.show used to silently accept and ignore all combinations of positional and keyword arguments. This behavior is deprecated.
- The unused shape and imlim parameters to .Axes.imshow are deprecated. To avoid triggering the deprecation warning, the filternorm, filterrad, resample, and url arguments should be passed by keyword.
- The interp_at_native parameter to .BboxImage, which has had no effect since Matplotlib 2.0, is deprecated.
- All arguments to the
matplotlib.cbook.deprecation.deprecateddecorator andmatplotlib.cbook.deprecation.warn_deprecatedfunction, except the first one (the version where the deprecation occurred), are now keyword-only. The goal is to avoid accidentally setting the "message" argument when the "name" (or "alternative") argument was intended, as this has repeatedly occurred in the past. - The arguments of matplotlib.testing.compare.calculate_rms have been renamed
from
expectedImage, actualImage, toexpected_image, actual_image. - Passing positional arguments to .Axis.set_ticklabels beyond ticklabels itself has no effect, and support for them is deprecated.
- Passing
shade=Noneto ~.axes3d.Axes3D.plot_surface is deprecated. This was an unintended implementation detail with the same semantics asshade=False. Please use the latter code instead. - matplotlib.ticker.MaxNLocator and its set_params method will issue a warning on unknown keyword arguments instead of silently ignoring them. Future versions will raise an error.
The arg parameter to matplotlib.use has been renamed to backend.
This will only affect cases where that parameter has been set as a keyword argument. The common usage pattern as a positional argument
matplotlib.use('Qt5Agg')is not affected.The normed parameter to .Axes.hist2d has been renamed to density.
The s parameter to .Annotation (and indirectly .Axes.annotate) has been renamed to text.
The tolerence parameter to .bezier.find_bezier_t_intersecting_with_closedpath, .bezier.split_bezier_intersecting_with_closedpath,
bezier.find_r_to_boundary_of_closedpath, .bezier.split_path_inout and .bezier.check_if_parallel has been renamed to tolerance.
In each case, the old parameter name remains supported (it cannot be used simultaneously with the new name), but support for it will be dropped in Matplotlib 3.3.
Support for custom backends that do not provide a
.GraphicsContextBase.set_hatch_color method is deprecated. We
suggest that custom backends let their GraphicsContext class
inherit from .GraphicsContextBase, to at least provide stubs for all
required methods.
spine.Spine.is_frame_like
This has not been used in the codebase since its addition in 2009.
axis3d.Axis.get_tick_positionsThis has never been used internally, there is no equivalent method exists on the 2D Axis classes, and despite the similar name, it has a completely different behavior from the 2D Axis' axis.Axis.get_ticks_position method.
.backend_pgf.LatexManagerFactorympl_toolkits.axisartist.axislines.SimpleChainedObjectsmpl_toolkits.Axes.AxisDict
checkdep_dvipngcheckdep_ghostscriptcheckdep_pdftopscheckdep_inkscapeticker.decade_upticker.decade_downcbook.dedentdocstring.Appenderdocstring.dedentdocstring.copy_dedent
Use the standard library's docstring manipulation tools instead, such as inspect.cleandoc and inspect.getdoc.
matplotlib.scale.get_scale_docs()matplotlib.pyplot.get_scale_docs()
These are considered internal and will be removed from the public API in a future version.
projections.process_projection_requirementsbackend_ps.PsBackendHelperbackend_ps.ps_backend_helper,cbook.iterablecbook.get_labelcbook.safezipManually check the lengths of the inputs instead, or rely on NumPy to do it.cbook.is_hashableUseisinstance(..., collections.abc.Hashable)instead.- The
.backend_bases.RendererBase.strip_math. Use .cbook.strip_math instead.
Multiple internal functions that were exposed as part of the public API of .mpl_toolkits.mplot3d are deprecated,
mpl_toolkits.mplot3d.art3d
mpl_toolkits.mplot3d.art3d.norm_anglempl_toolkits.mplot3d.art3d.norm_text_anglempl_toolkits.mplot3d.art3d.path_to_3d_segmentmpl_toolkits.mplot3d.art3d.paths_to_3d_segmentsmpl_toolkits.mplot3d.art3d.path_to_3d_segment_with_codesmpl_toolkits.mplot3d.art3d.paths_to_3d_segments_with_codesmpl_toolkits.mplot3d.art3d.get_patch_vertsmpl_toolkits.mplot3d.art3d.get_colorsmpl_toolkits.mplot3d.art3d.zalpha
mpl_toolkits.mplot3d.proj3d
mpl_toolkits.mplot3d.proj3d.line2dmpl_toolkits.mplot3d.proj3d.line2d_distmpl_toolkits.mplot3d.proj3d.line2d_seg_distmpl_toolkits.mplot3d.proj3d.modmpl_toolkits.mplot3d.proj3d.proj_transform_vecmpl_toolkits.mplot3d.proj3d.proj_transform_vec_clipmpl_toolkits.mplot3d.proj3d.vec_pad_onesmpl_toolkits.mplot3d.proj3d.proj_trans_clip_points
If your project relies on these functions, consider vendoring them.
backend_pdf.RendererPdf.afm_font_cachebackend_ps.RendererPS.afmfontdfont_manager.OSXInstalledFonts.TextToPath.glyph_to_path(Instead callfont.get_path()and manually transform the path.)
dates.seconds()dates.minutes()dates.hours()dates.weeks()dates.strpdate2numdates.bytespdate2num
These are brittle in the presence of locale changes. Use standard datetime
parsers such as time.strptime or dateutil.parser.parse, and additionally
call matplotlib.dates.date2num if you need to convert to Matplotlib's
internal datetime representation; or use dates.datestr2num.
- .axes3d.Axes3D.w_xaxis
- .axes3d.Axes3D.w_yaxis
- .axes3d.Axes3D.w_zaxis
Use .axes3d.Axes3D.xaxis, .axes3d.Axes3D.yaxis and .axes3d.Axes3D.zaxis instead.
matplotlib.testing.decorators.switch_backenddecorator
Test functions should use pytest.mark.backend, and the mark will be
picked up by the matplotlib.testing.conftest.mpl_test_settings fixture.
.colorattribute of .Quiver objects
Instead, use (as for any .Collection) the get_facecolor method.
Note that setting to the .color attribute did not update the quiver artist,
whereas calling set_facecolor does.
.get_py2exe_datafiles.tk_window_focus.backend_gtk3.FileChooserDialog.backend_gtk3.NavigationToolbar2GTK3.get_filechooser.backend_gtk3.SaveFigureGTK3.get_filechooser.NavigationToolbar2QT.adj_windowattribute. This is unused and alwaysNone..backend_wx.IDLE_DELAYglobal variable This is unused and only relevant to the now removed wx "idling" code (note that as it is a module-level global, no deprecation warning is emitted when accessing it).mlab.demeanbackend_gtk3cairo.FigureCanvasGTK3Cairo,backend_wx.debug_on_error,backend_wx.fake_stderr,backend_wx.raise_msg_to_str,backend_wx.MenuButtonWx,backend_wx.PrintoutWx,matplotlib.backends.qt_editor.formlayoutmodule
This module is a vendored, modified version of the official formlayout module available on PyPI. Install that module separately if you need it.
GraphicsContextPS.shouldstroke
LogTransformBaseLog10TransformLog2Transform,NaturalLogTransformLogInvertedLogTransformBaseInvertedLog10TransformInvertedLog2TransformInvertedNaturalLogTransform
These classes defined in :mod:`matplotlib.scale` are deprecated. As a replacement, use the general ~.scale.LogTransform and ~.scale.InvertedLogTransform classes, whose constructors take a base argument.
OldScalarFormatter.pprint_valScalarFormatter.pprint_valLogFormatter.pprint_val
These are helper methods that do not have a consistent signature across formatter classes.
path.get_paths_extents
Use ~.path.get_path_collection_extents instead.
.Path.has_nonfiniteattribute
Use not np.isfinite(path.vertices).all() instead.
.bezier.find_r_to_boundary_of_closedpathfunction is deprecated
This has always returned None instead of the requested radius.
text.TextWithDashText.is_math_textTextPath.is_math_textTextPath.text_get_vertices_codes(As an alternative, construct a newTextPathobject.)
NavigationToolbar2QT.buttonsLine2D.verticalOffsetQuiver.keytextQuiver.keyvecSpanSelector.buttonDown
These are unused and never updated.
matplotlib.sphinxext.mathmpl.math_directivematplotlib.sphinxext.plot_directive.plot_directive
This is because the matplotlib.sphinxext.mathmpl and
matplotlib.sphinxext.plot_directive interfaces have changed from the
(Sphinx-)deprecated function-based interface to a class-based interface; this
should not affect end users.
mpl_toolkits.axisartist.axis_artist.UnimplementedException
- The
MATPLOTLIBDATAenvironment variable
Axis.iter_ticks
This only served as a helper to the private .Axis._update_ticks
The following API elements have been un-deprecated:
- The obj_type keyword argument to the
matplotlib.cbook.deprecation.deprecateddecorator. - xmin, xmax keyword arguments to .Axes.set_xlim and ymin, ymax keyword arguments to .Axes.set_ylim
For consistency with .Line2D, the ~.text.Text class has gained the c
alias for the color property. For example, one can now write
ax.text(.5, .5, "foo", c="red")
It is now possible to go beyond the tenth color in the property cycle using
Cn syntax, e.g.
plt.plot([1, 2], color="C11")
now uses the 12th color in the cycle.
Note that previously, a construct such as:
plt.plot([1, 2], "C11")
would be interpreted as a request to use color C1 and marker 1
(an "inverted Y"). To obtain such a plot, one should now use
plt.plot([1, 2], "1C1")
(so that the first "1" gets correctly interpreted as a marker specification), or, more explicitly:
plt.plot([1, 2], marker="1", color="C1")
The .Formatter class gained a new ~.Formatter.format_ticks method, which takes the list of all tick locations as a single argument and returns the list of all formatted values. It is called by the axis tick handling code and, by default, first calls ~.Formatter.set_locs with all locations, then repeatedly calls ~.Formatter.__call__ for each location.
Tick-handling code in the codebase that previously performed this sequence (~.Formatter.set_locs followed by repeated ~.Formatter.__call__) have been updated to use ~.Formatter.format_ticks.
~.Formatter.format_ticks is intended to be overridden by .Formatter subclasses for which the formatting of a tick value depends on other tick values, such as .ConciseDateFormatter.
pcolorfast now accepts 3D images (RGB or RGBA) arrays if the X and Y specifications allow image or pcolorimage rendering; they remain unsupported by the more general quadmesh rendering
Passing invalid locations to ~.Axes.legend and ~.Axes.table used to fallback on a default location. This behavior is deprecated and will throw an exception in a future version.
.offsetbox.AnchoredText is unable to handle the horizontalalignment or verticalalignment kwargs, and used to ignore them with a warning. This behavior is deprecated and will throw an exception in a future version.
Passing steps less than 1 or greater than 10 to ~.ticker.MaxNLocator used to result in undefined behavior. It now throws a ValueError.
The signature of the (private) Axis._update_ticks has been changed to not
take the renderer as argument anymore (that argument is unused).