v0.8.0

Experimental: optional use of PyGEOS to speed up spatial operations (#1155). PyGEOS is a faster alternative for Shapely (being contributed back to a future version of Shapely), and is used in element-wise spatial operations and for spatial index in e.g. sjoin (#1343, #1401, #1421, #1427, #1428). See the installation docs for more info and how to enable it.

New features and improvements:

• IO enhancements:
  • New GeoDataFrame.to_postgis() method to write to PostGIS database (#1248).
  • New Apache Parquet and Feather file format support (#1180, #1435)
  • Allow appending to files with GeoDataFrame.to_file (#1229).
  • Add support for the ignore_geometry keyword in read_file to only read the attribute data. If set to True, a pandas DataFrame without geometry is returned (#1383).
  • geopandas.read_file now supports reading from file-like objects (#1329).
  • GeoDataFrame.to_file now supports specifying the CRS to write to the file (#802). By default it still uses the CRS of the GeoDataFrame.
  • New chunksize keyword in geopandas.read_postgis to read a query in chunks (#1123).
• Improvements related to geometry columns and CRS:
  • Any column of the GeoDataFrame that has a "geometry" dtype is now returned as a GeoSeries. This means that when having multiple geometry columns, not only the "active" geometry column is returned as a GeoSeries, but also accessing another geometry column (gdf["other_geom_column"]) gives a GeoSeries (#1336).
  • Multiple geometry columns in a GeoDataFrame can now each have a different CRS. The global gdf.crs attribute continues to returns the CRS of the "active" geometry column. The CRS of other geometry columns can be accessed from the column itself (eg gdf["other_geom_column"].crs) (#1339).
  • New set_crs() method on GeoDataFrame/GeoSeries to set the CRS of naive geometries (#747).
• Improvements related to plotting:
  • The y-axis is now scaled depending on the center of the plot when using a geographic CRS, instead of using an equal aspect ratio (#1290).
  • When passing a column of categorical dtype to the column= keyword of the GeoDataFrame plot(), we now honor all categories and its order (#1483). In addition, a new categories keyword allows to specify all categories and their order otherwise (#1173).
  • For choropleths using a classification scheme (using scheme=), the legend_kwds accept two new keywords to control the formatting of the legend: fmt with a format string for the bin edges (#1253), and labels to pass fully custom class labels (#1302).
• New covers() and covered_by() methods on GeoSeries/GeoDataframe for the equivalent spatial predicates (#1460, #1462).
• GeoPandas now warns when using distance-based methods with data in a geographic projection (#1378).

Deprecations:

• When constructing a GeoSeries or GeoDataFrame from data that already has a CRS, a deprecation warning is raised when both CRS don't match, and in the future an error will be raised in such a case. You can use the new set_crs method to override an existing CRS. See the docs.
• The helper functions in the geopandas.plotting module are deprecated for public usage (#656).
• The geopandas.io functions are deprecated, use the top-level read_file and to_file instead (#1407).
• The set operators (&, |, ^, -) are deprecated, use the intersection(), union(), symmetric_difference(), difference() methods instead (#1255).
• The sindex for empty dataframe will in the future return an empty spatial index instead of None (#1438).
• The objects keyword in the intersection method of the spatial index returned by the sindex attribute is deprecated and will be removed in the future (#1440).

Bug fixes:

• Fix the total_bounds() method to ignore missing and empty geometries (#1312).
• Fix geopandas.clip when masking with non-overlapping area resulting in an empty GeoDataFrame (#1309, #1365).
• Fix error in geopandas.sjoin when joining on an empty geometry column (#1318).
• CRS related fixes: pandas.concat preserves CRS when concatenating GeoSeries objects (#1340), preserve the CRS in geopandas.clip (#1362) and in GeoDataFrame.astype (#1366).
• Fix bug in GeoDataFrame.explode() when 'level_1' is one of the column names (#1445).
• Better error message when rtree is not installed (#1425).
• Fix bug in GeoSeries.equals() (#1451).
• Fix plotting of multi-part geometries with additional style keywords (#1385).

And we now have a Code of Conduct!

GeoPandas 0.8.0 is the last release to support Python 3.5. The next release
will require Python 3.6, pandas 0.24, numpy 1.15 and shapely 1.6 or higher.

Acknowledgments

Thanks to everyone who contributed to this release!
A total of 28 people contributed patches to this release. People with a "+" by their names contributed a patch for the first time.

• Adrian Garcia Badaracco +
• Alan D. Snow
• Bhavika Tekwani +
• Bo-Deng +
• Brendan Ward
• Christopher Yeh +
• Geir Arne Hjelle
• Henrikki Tenkanen +
• Ian Rose
• James McBride
• Joris Van den Bossche
• Julia Signell +
• Kyle Barron +
• Martin Fleischmann
• Martin Jul +
• Mateusz Konieczny +
• Mike Taves
• Oliver Schillinger +
• Rowan Molony +
• Sergio Rey
• SylvainLan +
• TimothyLucas +
• abonte +
• harryposner +
• pietro +
• raphacosta27 +
• rwijtvliet +
• sangarshanan

v0.7.0

Support for Python 2.7 has been dropped. GeoPandas now works with Python >= 3.5.

The important API change of this release is that GeoPandas now requires PROJ > 6 and pyproj > 2.2, and that the .crs attribute of a GeoSeries and GeoDataFrame no longer stores the CRS information as a proj4 string or dict, but as a pyproj.CRS object (#1101).

This gives a better user interface and integrates improvements from pyproj and PROJ 6, but might also require some changes in your code. Check the migration guide in the documentation.

Other API changes:

• The GeoDataFrame.to_file method will now also write the GeoDataFrame index to the file, if the index is named and/or non-integer. You can use the index=True/False keyword to overwrite this default inference (#1059).

New features and improvements:

• A new geopandas.clip function to clip a GeoDataFrame to the spatial extent of another shape (#1128).
• The geopandas.overlay function now works for all geometry types, including points and linestrings in addition to polygons (#1110).
• The plot() method gained support for missing values (in the column that determines the colors). By default it doesn't plot the corresponding geometries, but using the new missing_kwds argument you can specify how to style those geometries (#1156).
• The plot() method now also supports plotting GeometryCollection and LinearRing objects (#1225).
• Added support for filtering with a geometry or reading a subset of the rows in geopandas.read_file (#1160).
• Added support for the new nullable integer data type of pandas in GeoDataFrame.to_file (#1220).

Bug fixes:

• GeoSeries.reset_index() now correctly results in a GeoDataFrame instead of DataFrame (#1252).
• Fixed the geopandas.sjoin function to handle MultiIndex correctly (#1159).
• Fixed the geopandas.sjoin function to preserve the index name of the left GeoDataFrame (#1150).

Acknowledgments

Thanks to everyone who contributed to this release!
A total of 12 people contributed patches to this release. People with a "+" by their names contributed a patch for the first time.

• Alan D. Snow
• Aleksey Bilogur
• Ardie Orden +
• Brendan Ward +
• Brett Naul
• James McBride
• Joris Van den Bossche
• Leah Wasser
• Martin Fleischmann
• Mike Taves
• jess +
• sangarshanan +

v0.6.3

Small bug-fix release:

• Compatibility with Shapely 1.7 and pandas 1.0 (#1244).
• Fix GeoDataFrame.fillna to accept non-geometry values again when there are
  no missing values in the geometry column. This should make it easier to fill
  the numerical columns of the GeoDataFrame (#1279).

v0.6.2

Small bug-fix release fixing a few regressions:

• Fix a regression in passing an array of RRB(A) tuples to the .plot()
  method (#1178, #1211).
• Fix the bounds and total_bounds attributes for empty GeoSeries, which
  also fixes the repr of an empty or all-NA GeoSeries (#1184, #1195).
• Fix filtering of a GeoDataFrame to preserve the index type when ending up
  with an empty result (#1190).

v0.6.1

Small bug-fix release fixing a few regressions:

• Fix astype when converting to string with Multi geometries (#1145) or when converting a dataframe without geometries (#1144).
• Fix GeoSeries.fillna to accept np.nan again (#1149).

v0.6.0

GeoPandas 0.6.0 features a refactor of the internals based on the new pandas ExtensionArray interface, for better integration with pandas. Although this change should keep the user interface mostly stable, there are a few changes summarized below. Further, this release includes a nice set of other improvements and bug fixes.

Important note! This will be the last release to support Python 2.7 (#1031)

API changes:

• A refactor of the internals based on the pandas ExtensionArray interface (#1000). Read more on that in this blogpost, and 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 +

v0.6.0rc1 (release candidate)

Version 0.6.0rc1

v0.5.1

• Compatibility with latest mapclassify version 2.1.0 (#1025).

v0.5.0

GeoPandas 0.5.0 includes some improvements for writing files with fiona (better performance, better support for data types and mixed geometry types), along with many other new features and bug fixes, see the full list below.

Improvements:

• Significant performance improvement (around 10x) for GeoDataFrame.iterfeatures,
  which also improves GeoDataFrame.to_file (#864).
• File IO enhancements based on Fiona 1.8:
  • Support for writing bool dtype (#855) and datetime dtype, if the file format supports it (#728).
  • Support for writing dataframes with multiple geometry types, if the file format allows it (e.g. GeoJSON for all types, or ESRI Shapefile for Polygon+MultiPolygon) (#827, #867, #870).
• Compatibility with pyproj >= 2 (#962).
• A new geopandas.points_from_xy() helper function to convert x and y coordinates to Point objects (#896).
• The buffer and interpolate methods now accept an array-like to specify a variable distance for each geometry (#781).
• Addition of a relate method, corresponding to the shapely method that returns the DE-9IM matrix (#853).
• Plotting improvements:
  • Performance improvement in plotting by only flattening the geometries if there are actually 'Multi' geometries (#785).
  • Choropleths: access to all mapclassify classification schemes and addition of the classification_kwds keyword in the plot method to specify options for the scheme (#876).
  • Ability to specify a matplotlib axes object on which to plot the color bar with the cax keyword, in order to have more control over the color bar placement (#894).
• Changed the default provider in geopandas.tools.geocode from Google (now requires an API key) to Geocode.Farm (#907, #975).

Bug fixes:

• Remove the edge in the legend marker (#807).
• Fix the align method to preserve the CRS (#829).
• Fix geopandas.testing.assert_geodataframe_equal to correctly compare left and right dataframes (#810).
• Fix in choropleth mapping when the values contain missing values (#877).
• Better error message in sjoin if the input is not a GeoDataFrame (#842).
• Fix in read_postgis to handle nullable (missing) geometries (#856).
• Correctly passing through the parse_dates keyword in read_postgis to the underlying pandas method (#860).
• Fixed the shape of Antarctica in the included demo dataset 'naturalearth_lowres'
  (by updating to the latest version) (#804).

Acknowledgments

Thanks to everyone who contributed to this release!
A total of 33 people contributed to this release. People with a "+" by their
names contributed a patch for the first time.

• Andy Garfield +
• Benjamin Goldenberg +
• Brett Naul +
• Brian Lewis +
• Dmitry Nikolaev +
• Dr Martin Black +
• Filipe
• Geir Arne Hjelle
• Géraud +
• Henry Walshaw +
• James McBride
• Jesse Pisel +
• Joris Van den Bossche
• Joshua Wilson
• Justin Shenk +
• Kris Vanhoof +
• Leah Wasser +
• Levi John Wolf
• Martin Fleischmann +
• Matthieu Viry +
• Philipp Kats +
• Pratap Vardhan +
• Pulkit Maloo +
• Raphael Delhome +
• Sean Gillies
• Simon Andersson +
• TimoRoth +
• Yohann Rebattu +
• YuichiNotoya +
• byrman +
• lmmarsano +
• Émile Nadeau +
• Ömer Özak

v0.4.1

Small bug-fix release for compatibility with the latest Fiona and PySAL
releases:

• Compatibility with Fiona 1.8: fix deprecation warning (#854 and #916).
• Compatibility with PySAL 2.0: switched to mapclassify instead of PySAL as
  dependency for choropleth mapping with the scheme keyword (#872).
• Fix for new overlay implementation in case the intersection is empty (#800).

Acknowledgments

A total of 7 people contributed to this release. People with a "+" by their
names contributed a patch for the first time.

• Filipe
• Joris Van den Bossche
• Kris Vanhoof +
• Martin Fleischmann +
• Simon Andersson +
• TimoRoth +
• Ömer Özak