v0.14.4

What's Changed

• Fall back to pandas to recognise null types by @has2k1 in #69

Full Changelog: v0.14.3...v0.14.4

v0.14.3

• Fixed bug in hue_pal where drawing 24 or more colors broke the palette and you got colors from a very narrow hue band close to the first color.

Full Changelog: v0.14.2...v0.14.3

v0.14.2

Nothing new, only fixed bad test cases that were failing for the upcoming python 3.14 release.

v0.14.1

Bug Fixes

• Fixed but in to_rgba() used by plotnine to recognise numpy integers and floats as alpha values. And also to recognise None and and an empty string to also mean "none", which is recognised by Matplotlib.

v0.14.0

API Changes

• Changed hue_pal to use HCL color space from HSL (or HSLuv) space. The previous functionality is now available with hls_pal.
• breaks_date now only accepts parameter n, the desired number of breaks. To generate breaks by the width, use breaks_date_width.

New

• Added tableau color in the "tab:*" colorname space. e.g. "tab:red".
• breaks_width to generate numeric breaks by specifying the width (interval between the breaks).
• breaks_timedelta_width to generate timedelta breaks by specifying the width (interval between the breaks).

Enhancements

• Added name "gray" aliases to all XKCD colors partially named "grey".
• New algorithms to calculate datetime breaks by either n the desired number of breaks or width the interval between the breaks. When using the new breaks_date_width, you can align the dates by using an offset. (#58)

v0.13.5

Bug Fixes

• Fixed using a discrete palette and requesting less colours than than the smallest swatch in the palette. Among the affected palettes are category10, category20, category20b, category20c.

Full Changelog: v0.13.4...v0.13.5

v0.13.4

Enhancements

Added aliases to the vega colormaps for the names used in matplotlib. (plotnine issue #938)

v0.13.3

Enhancements

• Make it possible to train discrete data that is a sequence of tuples. (plotnine issue #933)

v0.13.2

Enhancements

• For python >=3.13 mizani.bounds.squish_infinite and mizani.bounds.squish do not warning about invalid values when passed arrays with dtype=uint64and a range with negative values.

v0.13.1

Enhancements

• Type checking passes with numpy 2.2.