Make bootstrapping reproducible & add unit test

Set random seed to make sampling of border units and subsequent
estimation of the spatial separation index (venables) reproducible.

Add unit test for calculating UCI with bootstrapping_border=True. The
goal of bootstrapping is to improve the estimate of the maximum spatial
separation index (venables) compared to the heuristic approach (of using
an equal distribution along the border of the area of interest). Test if
the bootstrapping approach succeeds in this regard.

Note that for small areas, e.g. the 3x3 polygon grid from the unit test,
this may not always be the case. For the given random seed it is though.

tests/test_uci.py

@@ -131,3 +131,11 @@ def test_uci_spatial_link_isolated(self):
         np.testing.assert_equal(result['UCI'], np.nan)
         np.testing.assert_equal(result['proximity_index'], np.nan)
         self.assertAlmostEqual(result['location_coef'], 0.3809523, places=5)
+
+
+    def test_bootstrap_border(self):
+
+        result = uci(self.gdf, 'activities')
+        result_bootstrap = uci(self.gdf, 'activities', bootstrap_border=True)
+
+        self.assertGreater(result_bootstrap['spatial_separation_max'], result['spatial_separation_max'])  # Bootstrapping should in most cases increase the maximum estimate of the spatial separation index

uci/uci.py

@@ -186,6 +186,7 @@ def _simulate_border_values(length, n):
     """
     Simulate equal value distributions along random subset of the border.
     """
+    np.random.seed(0)
     positions = np.random.choice(range(length), size=n)
     values = _create_weight_vector(length, positions)
     return values