API changes:

• A refactor of the internals based on the pandas ExtensionArray interface (#1000). The main user visible changes are:
  • The .dtype of a GeoSeries is now a 'geometry' dtype (and no longer a numpy object dtype).
  • The .values of a GeoSeries now returns a custom GeometryArray, and no longer a numpy array. To get back a numpy array of Shapely scalars, you can convert explicitly using np.asarray(..).
• The GeoSeries constructor now raises a warning when passed non-geometry data. Currently the constructor falls back to return a pandas Series, but in the future this will raise an error (#1085).
• The missing value handling has been changed to now separate the concepts of missing geometries and empty geometries (#601, 1062). In practice this means that (see the docs for more details):
  • GeoSeries.isna now considers only missing values, and if you want to check for empty geometries, you can use GeoSeries.is_empty (GeoDataFrame.isna already only looked at missing values).
  • GeoSeries.dropna now actually drops missing values (before it didn't drop either missing or empty geometries)
  • GeoSeries.fillna only fills missing values (behaviour unchanged).
  • GeoSeries.align uses missing values instead of empty geometries by default to fill non-matching index entries.

New features and improvements:

• Addition of a GeoSeries.affine_transform method, equivalent of Shapely's function (#1008).
• Addition of a GeoDataFrame.rename_geometry method to easily rename the active geometry column (#1053).
• Addition of geopandas.show_versions() function, which can be used to give an overview of the installed libraries in bug reports (#899).
• The legend_kwds keyword of the plot() method can now also be used to specify keywords for the color bar (#1102).
• Performance improvement in the sjoin() operation by re-using existing spatial index of the input dataframes, if available (#789).
• Updated documentation to work with latest version of geoplot and contextily (#1044, #1088).
• A new geopandas.options configuration, with currently a single option to control the display precision of the coordinates (options.display_precision). The default is now to show less coordinates (3 for projected and 5 for geographic coordinates), but the default can be overridden with the option.

Bug fixes:

• Also try to use pysal instead of mapclassify if available (#1082).
• The GeoDataFrame.astype() method now correctly returns a GeoDataFrame if the geometry column is preserved (#1009).
• The to_crs method now uses always_xy=True to ensure correct lon/lat order handling for pyproj>=2.2.0 (#1122).
• Fixed passing list-like colors in the plot() method in case of "multi" geometries (#1119).
• Fixed the coloring of shapes and colorbar when passing a custom norm in the plot() method (#1091, #1089).
• Fixed GeoDataFrame.to_file to preserve VFS file paths (e.g. when a "s3://" path is specified) (#1124).
• Fixed failing case in geopandas.sjoin with empty geometries (#1138).

In addition, the minimum required versions of some dependencies have been increased: GeoPandas now requires pandas >=0.23.4 and matplotlib >=2.0.1 (#1002).

Acknowledgments

Thanks to everyone who contributed to this release!
A total of 20 people contributed patches to this release. People with a "+" by their names contributed a patch for the first time.

• Alan D. Snow +
• Aleksey Bilogur
• Archana Alva +
• François Leblanc
• Geir Arne Hjelle
• Ian Rose +
• James Gaboardi +
• James McBride
• Joris Van den Bossche
• Joshua Wilson
• Kushal Borkar +
• Leah Wasser
• Martin Fleischmann
• Mike Taves +
• René Buffat +
• Sergio Rey +
• Thomas Pinder +
• awa5114 +
• donlo +
• jbrockmendel +