Merge pull request #55 from tomalrussell/fix/drop_duplicate_geometries

fix indexing error in drop_duplicate_geometries

src/snkit/network.py

@@ -484,11 +484,13 @@ def node_connectivity_degree(node, network):
 
 def drop_duplicate_geometries(gdf, keep="first"):
     """Drop duplicate geometries from a dataframe"""
+
     # convert to wkb so drop_duplicates will work
     # discussed in https://github.com/geopandas/geopandas/issues/521
-    mask = gdf.geometry.apply(lambda geom: geom.wkb)
-    # use dropped duplicates index to drop from actual dataframe
-    return gdf.iloc[mask.drop_duplicates(keep=keep).index]
+    mask = gdf.geometry.apply(lambda geom: geom.wkb).drop_duplicates(keep=keep).index
+
+    # use mask to drop from actual dataframe
+    return gdf.loc[mask]
 
 
 def nearest_point_on_edges(point, edges):

tests/test_init.py

@@ -7,6 +7,7 @@
     warnings.filterwarnings("ignore", category=DeprecationWarning)
     from geopandas import GeoDataFrame
 
+import pandas as pd
 from pandas.testing import assert_frame_equal
 from pytest import fixture, mark
 from shapely.geometry import Point, LineString, MultiPoint, MultiLineString
@@ -406,6 +407,24 @@ def test_passing_slice():
     assert_frame_equal(actual, expected)
 
 
+def test_drop_duplicate_geometries():
+    a = Point((0, 0))
+    b = Point((0, 2))
+    c = Point((0, 1))
+    ac = LineString([a, c])
+    cb = LineString([c, b])
+    # use an index that doesn't start from 0 to check our indexing hygiene
+    index = pd.Index([2, 3, 5, 7, 11, 13])
+    gdf_with_dupes = GeoDataFrame(
+        index=index,
+        data=[a, a, b, ac, ac, cb],
+        columns=["geometry"]
+    )
+    deduped = snkit.network.drop_duplicate_geometries(gdf_with_dupes)
+    # we should have just the first of each duplicate item
+    assert (deduped.index == pd.Index([2, 5, 7, 13])).all()
+
+
 @mark.skipif(not USE_NX, reason="networkx not available")
 def test_to_networkx(connected):
     """Test conversion to networkx"""