Stabilize node mapping encoder order

Author: FraLotito

hypergraphx/core/base.py

@@ -78,9 +78,9 @@ def expose_attributes_for_hashing(self):
         }
 
     def get_mapping(self):
-        from hypergraphx.utils.labeling import LabelEncoder
+        from hypergraphx.utils.labeling import fit_node_encoder
 
-        return LabelEncoder().fit(self.get_nodes())
+        return fit_node_encoder(self.get_nodes())
 
 
 class BaseHypergraph(SerializationMixin):

hypergraphx/utils/labeling.py

@@ -43,6 +43,20 @@ def inverse_transform(self, y: Iterable[int]) -> np.ndarray:
         return np.array([classes[int(i)] for i in y], dtype=object)
 
 
+def _stable_label_sort_key(value: Any) -> tuple[str, str]:
+    return (type(value).__name__, repr(value))
+
+
+def fit_node_encoder(nodes: Iterable[Any]) -> LabelEncoder:
+    """
+    Fit a LabelEncoder on a deterministic node order.
+
+    Nodes are sorted by type name and repr() so the mapping is stable even when
+    the underlying node iteration order is not.
+    """
+    return LabelEncoder().fit(sorted(nodes, key=_stable_label_sort_key))
+
+
 def relabel_edge(mapping: LabelEncoder, edge: Tuple):
     """
     Relabel an edge using a mapping.

tests/core/hypergraphs/test_hypergraphs.py

@@ -746,6 +746,13 @@ def test_get_mapping_roundtrip():
     assert set(encoder.inverse_transform(mapped)) == set(hg.get_nodes())
 
 
+def test_get_mapping_uses_deterministic_sorted_node_order():
+    hg = Hypergraph(edge_list=[("b", "a"), ("c", "b")])
+    encoder = hg.get_mapping()
+
+    assert list(encoder.classes_) == ["a", "b", "c"]
+
+
 def test_remove_node_keep_edges_updates_edges():
     hg = Hypergraph(weighted=True)
     hg.add_edge((1, 2, 3), weight=2.0, metadata={"kind": "tri"})

tests/utils/test_labeling.py

@@ -5,6 +5,7 @@
     relabel_edges,
     inverse_relabel_edge,
     inverse_relabel_edges,
+    fit_node_encoder,
     map_node,
     map_nodes,
     inverse_map_nodes,
@@ -58,3 +59,9 @@ def test_get_inverse_mapping():
 
     assert inv[0] == "a"
     assert inv[2] == "c"
+
+
+def test_fit_node_encoder_sorts_deterministically():
+    enc = fit_node_encoder(["b", "a", "c"])
+
+    assert list(enc.classes_) == ["a", "b", "c"]