PERF: eliminate redundant sorting (#793)

* PERF: eliminate redundant sorting

* remove redundant conversions in build_travel_cost

Author: martinfleis

libpysal/graph/_network.py

@@ -1,5 +1,6 @@
 import numpy as np
 
+from ._kernel import _kernel_functions
 from ._utils import _induce_cliques, _validate_geometry_input
 
 
@@ -61,7 +62,7 @@ def pdna_to_adj(origins, network, node_ids, threshold):
     return adj
 
 
-def build_travel_graph(df, network, threshold, mapping_distance):
+def build_travel_graph(df, network, threshold, mapping_distance, kernel=None):
     """Compute the shortest path between gdf centroids via a pandana.Network
     and return an adjacency list with weight=cost. Note unlike distance_band,
     :math:`G_{ij}` and :math:`G_{ji}` are often different because travel networks
@@ -81,6 +82,11 @@ def build_travel_graph(df, network, threshold, mapping_distance):
         snapping tolerance passed to ``pandana.Network.get_node_ids`` that defines
         the maximum range at which observations are snapped to nearest nodes in the
         network. Default is None
+    kernel : str
+        kernel transformation applied to the weights. See
+        libpysal.graph.Graph.build_kernel for more information on kernel
+        transformation options. Default is None, in which case the Graph weight
+        is pure distance between focal and neighbor
 
     Returns
     -------
@@ -118,4 +124,11 @@ def build_travel_graph(df, network, threshold, mapping_distance):
         .reset_index()
     )
 
+    if callable(kernel):
+        adj_cliques["weight"] = kernel(adj_cliques["weight"], threshold)
+    else:
+        adj_cliques["weight"] = _kernel_functions[kernel](
+            adj_cliques["weight"], threshold
+        )
+
     return adj_cliques

libpysal/graph/_utils.py

@@ -32,7 +32,6 @@ def _sparse_to_arrays(sparray, ids=None, resolve_isolates=True, return_adjacency
     When we know we are dealing with cliques, we don't want to resolve
     isolates here but will do that later once cliques are induced.
     """
-    argsort_kwds = {"stable": True} if NUMPY_GE_2 else {}
     sparray = sparray.tocoo(copy=False)
     if ids is not None:
         ids = np.asarray(ids)
@@ -42,15 +41,13 @@ def _sparse_to_arrays(sparray, ids=None, resolve_isolates=True, return_adjacency
                 f"the shape of sparse {sparray.shape}."
             )
 
-        sorter = sparray.row.argsort(**argsort_kwds)
-        head = ids[sparray.row][sorter]
-        tail = ids[sparray.col][sorter]
-        data = sparray.data[sorter]
+        head = ids[sparray.row]
+        tail = ids[sparray.col]
+        data = sparray.data
     else:
-        sorter = sparray.row.argsort(**argsort_kwds)
-        head = sparray.row[sorter]
-        tail = sparray.col[sorter]
-        data = sparray.data[sorter]
+        head = sparray.row
+        tail = sparray.col
+        data = sparray.data
         ids = np.arange(sparray.shape[0], dtype=int)
 
     if resolve_isolates:

libpysal/graph/base.py

@@ -1085,7 +1085,7 @@ def build_kernel(
             coplanar=coplanar,
         )
 
-        return cls.from_arrays(head, tail, weight)
+        return cls.from_arrays(head, tail, weight, is_sorted=True)
 
     @classmethod
     def build_knn(cls, data, k, metric="euclidean", p=2, coplanar="raise"):
@@ -1529,19 +1529,11 @@ def build_travel_cost(
                 117         333.639008
         Name: weight, dtype: float64
         """
-        adj = _build_travel_graph(df, network, threshold, mapping_distance)
-        g = cls.from_adjacency(adj)
-        if kernel is not None:
-            arrays = _kernel(
-                g.sparse,
-                metric="precomputed",
-                kernel=kernel,
-                bandwidth=threshold,
-                resolve_isolates=False,
-                ids=df.index.values,
-            )
-            return cls.from_arrays(*arrays)
-        return g
+        adj = _build_travel_graph(
+            df, network, threshold, mapping_distance, kernel=kernel
+        )
+
+        return cls.from_adjacency(adj)
 
     @cached_property
     def neighbors(self):