Path lifting (Graph to Hypergraph)

configs/transforms/liftings/graph2hypergraph/path_lifting.yaml

@@ -0,0 +1,3 @@
+transform_type: 'lifting'
+transform_name: "PathLifting"
+feature_lifting: ProjectionSum

modules/transforms/data_transform.py

@@ -12,6 +12,7 @@
 from modules.transforms.liftings.graph2hypergraph.knn_lifting import (
     HypergraphKNNLifting,
 )
+from modules.transforms.liftings.graph2hypergraph.path_lifting import PathLifting
 from modules.transforms.liftings.graph2simplicial.clique_lifting import (
     SimplicialCliqueLifting,
 )
@@ -31,6 +32,7 @@
     "OneHotDegreeFeatures": OneHotDegreeFeatures,
     "NodeFeaturesToFloat": NodeFeaturesToFloat,
     "KeepOnlyConnectedComponent": KeepOnlyConnectedComponent,
+    "PathLifting": PathLifting,
 }
 
 

modules/transforms/liftings/graph2hypergraph/path_lifting.py

@@ -0,0 +1,136 @@
+"""A module for the PathLifting class."""
+import networkx as nx
+import numpy as np
+import torch
+import torch_geometric
+
+from modules.transforms.liftings.graph2hypergraph.base import Graph2HypergraphLifting
+
+
+class PathLifting(Graph2HypergraphLifting):
+    """Lifts graphs to hypergraph domain by considering paths between nodes."""
+
+    def __init__(
+        self,
+        source_nodes: list[int],
+        target_nodes: list[int],
+        lengths: list[int],
+        include_smaller_paths=False,
+        **kwargs,
+    ):
+        # guard clauses
+        if len(source_nodes) != len(lengths):
+            raise ValueError("source_nodes and lengths must have the same length")
+        if target_nodes is not None and len(target_nodes) != len(source_nodes):
+            raise ValueError(
+                "When target_nodes is not None, it must have the same length"
+                "as source_nodes"
+            )
+        if len(source_nodes) == 0:
+            raise ValueError(
+                "source_nodes,target_nodes and lengths must have at least one element"
+            )
+
+        super().__init__(**kwargs)
+        self.source_nodes = source_nodes
+        self.target_nodes = target_nodes
+        self.lengths = lengths
+        self.include_smaller_paths = include_smaller_paths
+
+    def find_hyperedges(self, data: torch_geometric.data.Data):
+        """Finds hyperedges from paths between nodes in a graph."""
+        G = torch_geometric.utils.convert.to_networkx(data, to_undirected=True)
+        s_hyperedges = set()
+
+        if self.target_nodes is None:  # all paths stemming from source nodes only
+            for source, length in zip(self.source_nodes, self.lengths, strict=True):
+                D, d_id2label, l_leafs = self.build_stemmingTree(G, source, length)
+                s = self.extract_hyperedgesFromStemmingTree(D, d_id2label, l_leafs)
+                s_hyperedges = s_hyperedges.union(s)
+
+        else:  # paths from source_nodes to target_nodes or from source nodes only
+            for source, target, length in zip(
+                self.source_nodes, self.target_nodes, self.lengths, strict=True
+            ):
+                if target is None:
+                    D, d_id2label, l_leafs = self.build_stemmingTree(G, source, length)
+                    s = self.extract_hyperedgesFromStemmingTree(D, d_id2label, l_leafs)
+                    s_hyperedges = s_hyperedges.union(s)
+                else:
+                    paths = list(
+                        nx.all_simple_paths(
+                            G, source=source, target=target, cutoff=length
+                        )
+                    )
+                    if not self.include_smaller_paths:
+                        paths = [path for path in paths if len(path) - 1 == length]
+                    s_hyperedges = s_hyperedges.union({frozenset(x) for x in paths})
+        return s_hyperedges
+
+    def lift_topology(self, data: torch_geometric.data.Data):
+        s_hyperedges = self.find_hyperedges(data)
+        indices = [[], []]
+        for edge_id, x in enumerate(s_hyperedges):
+            indices[1].extend([edge_id] * len(x))
+            indices[0].extend(list(x))
+        incidence = torch.sparse_coo_tensor(
+            indices, torch.ones(len(indices[0])), (len(data.x), len(s_hyperedges))
+        )
+        return {
+            "incidence_hyperedges": incidence,
+            "num_hyperedges": len(s_hyperedges),
+            "x_0": data.x,
+        }
+
+    def build_stemmingTree(self, G, source_root, length, verbose=False):
+        """Creates a directed tree from a source node with paths of a given length."""
+        d_id2label = {}
+        stack = []
+        D = nx.DiGraph()
+        n_id = 0
+        D.add_node(n_id)
+        d_id2label[n_id] = source_root
+        stack.append(n_id)
+        n_id += 1
+        l_leafs = []
+        while len(stack) > 0:
+            node = stack.pop()
+            neighbors = list(G.neighbors(d_id2label[node]))
+            visited_id = nx.shortest_path(D, source=0, target=node)
+            visited_labels = [d_id2label[i] for i in visited_id]
+            for neighbor in neighbors:
+                if neighbor not in visited_labels:
+                    D.add_node(n_id)
+                    d_id2label[n_id] = neighbor
+                    if len(visited_labels) < length:
+                        stack.append(n_id)
+                    elif len(visited_labels) == length:
+                        l_leafs.append(n_id)
+                    else:
+                        raise ValueError("Visited labels length is greater than length")
+                    D.add_edge(node, n_id)
+                    n_id += 1
+            if verbose:
+                print("\nLoop Variables Summary:")
+                print("nodes:", node)
+                print("neighbors:", neighbors)
+                print("visited_id:", visited_id)
+                print("visited_labels:", visited_labels)
+                print("stack:", stack)
+                print("id2label:", d_id2label)
+        return D, d_id2label, l_leafs
+
+    def extract_hyperedgesFromStemmingTree(self, D, d_id2label, l_leafs):
+        """From the root of the directed tree D,
+        extract hyperedges from the paths to the leafs."""
+        a_paths = np.array(
+            [list(map(d_id2label.get, nx.shortest_path(D, 0, x))) for x in l_leafs]
+        )
+        s_hyperedges = {
+            (frozenset(x)) for x in a_paths
+        }  # set bc != paths can be same hpedge
+        if self.include_smaller_paths:
+            for i in range(a_paths.shape[1] - 1, 1, -1):
+                a_paths = np.unique(a_paths[:, :i], axis=0)
+                s_hyperedges = s_hyperedges.union({(frozenset(x)) for x in a_paths})
+        return s_hyperedges

modules/utils/utils.py

@@ -233,7 +233,7 @@ def sort_vertices_ccw(vertices):
         n_hyperedges = incidence.shape[1]
         vertices += [i + n_vertices for i in range(n_hyperedges)]
         indices = incidence.indices()
-        edges = np.array([indices[1].numpy(), indices[0].numpy() + n_vertices]).T
+        edges = np.array([indices[0].numpy(), indices[1].numpy() + n_vertices]).T
         pos_n = [[i, 0] for i in range(n_vertices)]
         pos_he = [[i, 1] for i in range(n_hyperedges)]
         pos = pos_n + pos_he

test/transforms/liftings/graph2hypergraph/test_path_lifting.py

@@ -0,0 +1,156 @@
+"""Test the path lifting module."""
+
+import numpy as np
+
+from modules.data.load.loaders import GraphLoader
+from modules.transforms.liftings.graph2hypergraph.path_lifting import PathLifting
+from modules.utils.utils import load_dataset_config
+
+
+class TestHypergraphPathLifting:
+    """Test the PathLifting class."""
+
+    def setup_method(self):
+        """Initialise the PathLifting class."""
+        dataset_config = load_dataset_config("manual_dataset")
+        loader = GraphLoader(dataset_config)
+        self.dataset = loader.load()
+        self.data = self.dataset._data
+
+    def test_true(self):
+        """Naive test to check if the test is running."""
+        assert True
+
+    # def test_false(self):
+    #     """Naive test to check if the test is running."""
+    #     assert False
+
+    def test_1(self):
+        """Verifies setup_method is working."""
+        assert self.dataset is not None
+
+    def test_2(self):
+        """test: no target node for one source node returns something"""
+        source_nodes = [0, 2]
+        target_nodes = [1, None]
+        lengths = [2, 2]
+        include_smaller_paths = True
+        path_lifting = PathLifting(
+            source_nodes,
+            target_nodes,
+            lengths,
+            include_smaller_paths=include_smaller_paths,
+        )
+        res = path_lifting.find_hyperedges(self.data)
+        res_expected = [
+            [0, 1],
+            [0, 1, 2],
+            [0, 4, 1],
+            [2, 4],
+            [2, 1],
+            [2, 0],
+            [2, 7],
+            [2, 5],
+            [2, 3],
+            [2, 1, 4],
+            [2, 4, 0],
+            [2, 1, 0],
+            [2, 0, 7],
+            [2, 5, 7],
+            [2, 3, 6],
+            [2, 5, 6],
+            # [],
+        ]
+        assert {frozenset(x) for x in res_expected} == res
+
+    def test_3(self):
+        """test: include_smaller_paths=False"""
+        source_nodes = [0]
+        target_nodes = [1]
+        lengths = [2]
+        include_smaller_paths = False
+        res = PathLifting(
+            source_nodes,
+            target_nodes,
+            lengths,
+            include_smaller_paths=include_smaller_paths,
+        ).find_hyperedges(self.data)
+        assert frozenset({0, 1}) not in res
+
+    def test_4(self):
+        """test: include_smaller_paths=True"""
+        source_nodes = [0]
+        target_nodes = [1]
+        lengths = [2]
+        include_smaller_paths = True
+        res = PathLifting(
+            source_nodes,
+            target_nodes,
+            lengths,
+            include_smaller_paths=include_smaller_paths,
+        ).find_hyperedges(self.data)
+        assert frozenset({0, 1}) in res
+
+    def test_5(self):
+        """test: when include_smaller_paths=False all paths have the length specified"""
+        source_nodes = [0]
+        target_nodes = [1]
+        include_smaller_paths = False
+        for k in range(1, 5):
+            lengths = [k]
+            res = PathLifting(
+                source_nodes,
+                target_nodes,
+                lengths,
+                include_smaller_paths=include_smaller_paths,
+            ).find_hyperedges(self.data)
+            assert np.array([len(x) - 1 == k for x in res]).all()
+
+    def test_6(self):
+        """test: no target node global returns something"""
+        source_nodes = [0, 1]
+        target_nodes = None
+        lengths = [2, 2]
+        include_smaller_paths = False
+        res = PathLifting(
+            source_nodes,
+            target_nodes,
+            lengths,
+            include_smaller_paths=include_smaller_paths,
+        ).find_hyperedges(self.data)
+        assert len(res) > 0
+
+    def test_7(self):
+        """test: every hyperedge contains the source and target nodes when specified"""
+        a = np.random.default_rng().choice(
+            np.arange(len(self.data.x)), 2, replace=False
+        )
+        source_nodes = [a[0]]
+        target_nodes = [a[1]]
+        lengths = [np.random.default_rng().integers(1, 5)]
+        include_smaller_paths = False
+        res = PathLifting(
+            source_nodes,
+            target_nodes,
+            lengths,
+            include_smaller_paths=include_smaller_paths,
+        ).find_hyperedges(self.data)
+        if len(res) > 0:
+            assert (
+                np.array([source_nodes[0] in x for x in res]).all()
+                and np.array([target_nodes[0] in x for x in res]).all()
+            )
+
+    def test_8(self):
+        """test: no target node for one source node returns something"""
+        source_nodes = [0, 2]
+        target_nodes = [1, None]
+        lengths = [2, 2]
+        include_smaller_paths = False
+        res = PathLifting(
+            source_nodes,
+            target_nodes,
+            lengths,
+            include_smaller_paths=include_smaller_paths,
+        ).find_hyperedges(self.data)
+        assert len(res) > 0