added seed handling for trees

main.py

@@ -148,6 +148,8 @@ def create_tree_graphs(
     min_nnodes: int,
     max_nnodes: int,
     motif: Literal["cycle", "grid", "grid_v2"],
+    *,
+    seed: int | None = None,
 ) -> tuple[ExplainerDataset, NegativeGraphs]:
     """Create tree datasets with and without motifs.
 
@@ -159,6 +161,10 @@ def create_tree_graphs(
         Number of negative graphs (without motifs) to generate.
     motif : {"cycle", "grid", "grid_v2"}
         Motif type to insert into positive graphs.
+    seed : int or None, optional
+        Base seed for the internal RNG used to sample node counts. The negative
+        graph generator uses ``seed + 1``. If ``None``, a random base seed is
+        sampled to keep positive/negative distributions independent.
 
     Returns
     -------
@@ -193,10 +199,20 @@ def create_tree_graphs(
     else:
         motif_generator = motif
 
+    # Set up different seeds for positive and negative datasets
+    if seed is None:
+        base_seed = int(torch.randint(0, 2**31 - 1, (1,)).item())
+    else:
+        base_seed = seed
+    pos_seed = base_seed
+    neg_seed = base_seed + 1
+
+    # Generate graph datasets
     pos_ds = ExplainerDataset(
         graph_generator=UniformPruferTreeGraph(
             range(min_nnodes - motif_nnodes[motif], max_nnodes - motif_nnodes[motif] + 1),
             undirected=True,
+            seed=pos_seed,
         ),
         motif_generator=motif_generator,
         motif_generator_kwargs=motif_gen_args[motif],
@@ -205,7 +221,11 @@ def create_tree_graphs(
         transform=ShuffleNodes(),
     )
     neg_ds = NegativeGraphs(
-        graph_generator=UniformPruferTreeGraph(range(min_nnodes, max_nnodes + 1), undirected=True),
+        graph_generator=UniformPruferTreeGraph(
+            range(min_nnodes, max_nnodes + 1),
+            undirected=True,
+            seed=neg_seed,
+        ),
         num_graphs=num_neg_samples,
         transform=ShuffleNodes(),
     )