Merge pull request #676 from martinfleis/apply

ENH: add Graph.apply, Graph.aggregate and allow callable as transformation in transform

libpysal/graph/base.py

@@ -27,7 +27,7 @@
     _sparse_to_arrays,
 )
 
-ALLOWED_TRANSFORMATIONS = ("O", "B", "R", "D", "V")
+ALLOWED_TRANSFORMATIONS = ("O", "B", "R", "D", "V", "C")
 
 # listed alphabetically
 __author__ = """"
@@ -68,6 +68,7 @@ def __init__(self, adjacency, transformation="O", is_sorted=False):
             - **R** -- Row-standardization (global sum :math:`=n`)
             - **D** -- Double-standardization (global sum :math:`=1`)
             - **V** -- Variance stabilizing
+            - **C** -- Custom
         is_sorted : bool, default False
             ``adjacency`` capturing the graph needs to be canonically sorted to
             initialize the class. The MultiIndex needs to be ordered i-->j
@@ -885,7 +886,7 @@ def transform(self, transformation):
 
         Parameters
         ----------
-        transformation : str
+        transformation : str | callable
             Transformation method. The following are
             valid transformations.
 
@@ -894,6 +895,9 @@ def transform(self, transformation):
             - **D** -- Double-standardization (global sum :math:`=1`)
             - **V** -- Variance stabilizing
 
+            Alternatively, you can pass your own callable passed to
+            ``self.adjacency.groupby(level=0).transform()``.
+
         Returns
         -------
         Graph
@@ -904,7 +908,8 @@ def transform(self, transformation):
         ValueError
             Value error for unsupported transformation
         """
-        transformation = transformation.upper()
+        if isinstance(transformation, str):
+            transformation = transformation.upper()
 
         if self.transformation == transformation:
             return self.copy()
@@ -929,10 +934,14 @@ def transform(self, transformation):
             n_q = self.n / s.sum()
             standardized = (s * n_q).fillna(0).values  # isolate comes as NaN -> 0
 
+        elif callable(transformation):
+            standardized = self._adjacency.groupby(level=0).transform(transformation)
+            transformation = "C"
+
         else:
             raise ValueError(
                 f"Transformation '{transformation}' is not supported. "
-                f"Use one of {ALLOWED_TRANSFORMATIONS[1:]}"
+                f"Use one of {ALLOWED_TRANSFORMATIONS[1:]} or pass a callable."
             )
 
         standardized_adjacency = pd.Series(
@@ -1455,6 +1464,53 @@ def assign_self_weight(self, weight=1):
         )
         return Graph(adj, is_sorted=True)
 
+    def apply(self, y, func, **kwargs):
+        """Apply a reduction across the neighbor sets
+
+        Applies ``func`` over groups of ``y`` defined by neighbors for each focal.
+
+        Parameters
+        ----------
+        y : array_like
+            array of values to be grouped. Can be 1-D or 2-D and will be coerced to a
+            pandas object
+        func : function, str, list, dict or None
+            Function to use for aggregating the data passed to pandas ``GroupBy.apply``.
+
+        Returns
+        -------
+        Series | DataFrame
+            pandas object indexed by unique_ids
+        """
+        if not isinstance(y, pd.Series | pd.DataFrame):
+            y = pd.DataFrame(y) if hasattr(y, "ndim") and y.ndim == 2 else pd.Series(y)
+        grouper = y.take(self._adjacency.index.codes[1]).groupby(
+            self._adjacency.index.codes[0]
+        )
+        result = grouper.apply(func, **kwargs)
+        result.index = self.unique_ids
+        if isinstance(result, pd.Series):
+            result.name = None
+        return result
+
+    def aggregate(self, func):
+        """Aggregate weights within a neighbor set
+
+        Apply a custom aggregation function to a group of weights of the same focal
+        geometry.
+
+        Parameters
+        ----------
+        func : callable
+            A callable accepted by pandas ``groupby.agg`` method
+
+        Returns
+        -------
+        pd.Series
+            Aggregated weights
+        """
+        return self._adjacency.groupby(level=0).agg(func)
+
 
 def _arrange_arrays(heads, tails, weights, ids=None):
     """

libpysal/graph/tests/test_base.py

@@ -761,6 +761,12 @@ def test_transform(self):
         with pytest.raises(ValueError, match="Transformation 'X' is not"):
             self.g_int.transform("x")
 
+    def test_transform_callable(self):
+        contig = graph.Graph.build_contiguity(self.nybb)
+        trans = contig.transform(lambda x: x * 10)
+        assert trans.transformation == "C"
+        assert trans.adjacency.sum() == 100
+
     def test_asymmetry(self):
         neighbors = {
             "a": ["b", "c", "d"],
@@ -982,3 +988,112 @@ def test_assign_self_weight(self):
                 diag_array._adjacency[(contig.unique_ids[i], contig.unique_ids[i])]
                 == val
             )
+
+    def test_apply(self):
+        contig = graph.Graph.build_contiguity(self.nybb)
+
+        # pandas built-in
+        expected = pd.Series(
+            [1.62382200e09, 3.76087588e09, 3.68168493e09, 6.16961834e09, 3.68168493e09],
+            index=pd.Index(
+                ["Staten Island", "Queens", "Brooklyn", "Manhattan", "Bronx"],
+                name="focal",
+            ),
+        )
+        pd.testing.assert_series_equal(contig.apply(self.nybb.area, "sum"), expected)
+
+        # numpy
+        expected = pd.Series(
+            [1.62382200e09, 1.18692629e09, 1.84084247e09, 1.93747835e09, 1.84084247e09],
+            index=pd.Index(
+                ["Staten Island", "Queens", "Brooklyn", "Manhattan", "Bronx"],
+                name="focal",
+            ),
+        )
+        pd.testing.assert_series_equal(
+            contig.apply(self.nybb.area, np.median), expected
+        )
+
+        # lambda over geometry
+        expected = pd.Series(
+            [2.06271959e09, 6.68788190e09, 7.57087991e09, 8.78957337e09, 7.57087991e09],
+            index=pd.Index(
+                ["Staten Island", "Queens", "Brooklyn", "Manhattan", "Bronx"],
+                name="focal",
+            ),
+        )
+        pd.testing.assert_series_equal(
+            contig.apply(self.nybb.geometry, lambda x: x.unary_union.convex_hull.area),
+            expected,
+        )
+
+        # reduction over a dataframe
+        expected = pd.DataFrame(
+            [
+                [3.30470010e05, 1.62381982e09],
+                [1.56477261e06, 3.76087473e09],
+                [1.25564314e06, 3.68168433e09],
+                [2.10181756e06, 6.16961599e09],
+                [1.25564314e06, 3.68168433e09],
+            ],
+            columns=["Shape_Leng", "Shape_Area"],
+            index=pd.Index(
+                ["Staten Island", "Queens", "Brooklyn", "Manhattan", "Bronx"],
+                name="focal",
+            ),
+        )
+        pd.testing.assert_frame_equal(
+            contig.apply(
+                self.nybb, lambda x: x[["Shape_Leng", "Shape_Area"]].sum(axis=None)
+            ),
+            expected,
+        )
+
+        # 1D array input
+        expected = pd.Series(
+            [1.62382200e09, 3.76087588e09, 3.68168493e09, 6.16961834e09, 3.68168493e09],
+            index=pd.Index(
+                ["Staten Island", "Queens", "Brooklyn", "Manhattan", "Bronx"],
+                name="focal",
+            ),
+        )
+        pd.testing.assert_series_equal(
+            contig.apply(self.nybb.area.values, "sum"), expected
+        )
+
+        # 2D array input
+        expected = pd.DataFrame(
+            [
+                [3.30470010e05, 1.62381982e09],
+                [1.56477261e06, 3.76087473e09],
+                [1.25564314e06, 3.68168433e09],
+                [2.10181756e06, 6.16961599e09],
+                [1.25564314e06, 3.68168433e09],
+            ],
+            index=pd.Index(
+                ["Staten Island", "Queens", "Brooklyn", "Manhattan", "Bronx"],
+                name="focal",
+            ),
+        )
+        pd.testing.assert_frame_equal(
+            contig.apply(
+                self.nybb[["Shape_Leng", "Shape_Area"]].values,
+                lambda x: x.sum(axis=None),
+            ),
+            expected,
+        )
+
+    def test_aggregate(self):
+        contig = graph.Graph.build_contiguity(self.nybb)
+        expected = pd.Series(
+            [7.3890561, 7.3890561, 20.08553692, 20.08553692, 1.0],
+            index=pd.Index(
+                ["Bronx", "Brooklyn", "Manhattan", "Queens", "Staten Island"],
+                name="focal",
+            ),
+            name="weight",
+        )
+        pd.testing.assert_series_equal(
+            contig.aggregate(lambda x: np.exp(np.sum(x))),
+            expected,
+        )