pyg-nightly 2.7.0.dev20250604__py3-none-any.whl → 2.7.0.dev20250606__py3-none-any.whl

{pyg_nightly-2.7.0.dev20250604.dist-info → pyg_nightly-2.7.0.dev20250606.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: pyg-nightly
-Version: 2.7.0.dev20250604
+Version: 2.7.0.dev20250606
 Summary: Graph Neural Network Library for PyTorch
 Keywords: deep-learning,pytorch,geometric-deep-learning,graph-neural-networks,graph-convolutional-networks
 Author-email: Matthias Fey <matthias@pyg.org>

{pyg_nightly-2.7.0.dev20250604.dist-info → pyg_nightly-2.7.0.dev20250606.dist-info}/RECORD

@@ -1,4 +1,4 @@
-torch_geometric/__init__.py,sha256=di4kZd7rQFDpBlATSXKDIp6hVI7yFbo5CFhhBq2M5zs,2255
+torch_geometric/__init__.py,sha256=BD3Th7gs3RFBES0GFQ42zZ7rQwA2D8n5zYL1L0hvKFs,2255
 torch_geometric/_compile.py,sha256=f-WQeH4VLi5Hn9lrgztFUCSrN_FImjhQa6BxFzcYC38,1338
 torch_geometric/_onnx.py,sha256=V9ffrIKSqhDw6xUZ12lkuSfNs48cQp2EeJ6Z19GfnVw,349
 torch_geometric/backend.py,sha256=lVaf7aLoVaB3M-UcByUJ1G4T4FOK6LXAg0CF4W3E8jo,1575
@@ -442,7 +442,7 @@ torch_geometric/nn/models/dimenet.py,sha256=O2rqEx5HWs_lMwRD8eq6WMkbqJaCLL5zgWUJ
 torch_geometric/nn/models/dimenet_utils.py,sha256=Eyn_EiJqwKvuYj6BtRpSxrzMG3v4Gk98X9MxZ7uvwm4,5069
 torch_geometric/nn/models/g_retriever.py,sha256=tVibbqM_r-1LnA3R3oVyzp0bpuN3qPoYqcU6LZ8dYEk,8260
 torch_geometric/nn/models/git_mol.py,sha256=Wc6Hx6RDDR7sDWRWHfA5eK9e9gFsrTZ9OLmpMfoj3pE,12676
-torch_geometric/nn/models/glem.py,sha256=sT0XM4klVlci9wduvUoXupATUw9p25uXtaJBrmv3yvs,16431
+torch_geometric/nn/models/glem.py,sha256=1Tl9lb5i8-GwtYMiqFtRmgAC7esSr606ekd2RXXBX2Y,16449
 torch_geometric/nn/models/gnnff.py,sha256=15dkiLgy0LmH1hnUrpeoHioIp4BPTfjpVATpnGRt9E0,7860
 torch_geometric/nn/models/gpse.py,sha256=3tKzVPkvntAtj13qSPJO1rittr8X1Q_7OqVTOaSpp2s,41982
 torch_geometric/nn/models/graph_mixer.py,sha256=mthMeCOikR8gseEsu4oJ3Cd9C35zHSv1p32ROwnG-6s,9246
@@ -620,7 +620,7 @@ torch_geometric/utils/embedding.py,sha256=b-CQ-aapEgahxSS7fuL4aNQX6GJROboV0xclZ_
 torch_geometric/utils/functions.py,sha256=orQdS_6EpzWSmBHSok3WhxCzLy9neB-cin1aTnlXY-8,703
 torch_geometric/utils/geodesic.py,sha256=-xsqE3FZU7Y9gMbucIlGJ4FM-3nk8o0AQBxIdN-QfEw,4770
 torch_geometric/utils/hetero.py,sha256=ok4uAAOyMiaeEPmvyS4DNoDwdKnLS2gmgs5WVVklxOo,5539
-torch_geometric/utils/influence.py,sha256=b2v4G2jA-FKWWSohIc4iNkyFgUvS6j_g97qWT6xfrzI,9920
+torch_geometric/utils/influence.py,sha256=7R-NW4myJMJPkbNiwcHTmO_m_B3gPB2IlBbQkB446xc,10348
 torch_geometric/utils/isolated.py,sha256=nUxCfMY3q9IIFjelr4eyAJH4sYG9W3lGdpWidnp3dm4,3588
 torch_geometric/utils/laplacian.py,sha256=ludDil4yS1A27PEuYOjZtCtE3o-t0lnucJKfiqENhvM,3695
 torch_geometric/utils/loop.py,sha256=MUWUS7a5GxuxLKlCtRq95U1hc3MndybAhqKD5IAe2RY,23051
@@ -640,7 +640,7 @@ torch_geometric/utils/undirected.py,sha256=H_nfpI0_WluOG6VfjPyldvcjL4w5USAKWu2x5
 torch_geometric/visualization/__init__.py,sha256=b-HnVesXjyJ_L1N-DnjiRiRVf7lhwKaBQF_2i5YMVSU,208
 torch_geometric/visualization/graph.py,sha256=mfZHXYfiU-CWMtfawYc80IxVwVmtK9hbIkSKhM_j7oI,14311
 torch_geometric/visualization/influence.py,sha256=CWMvuNA_Nf1sfbJmQgn58yS4OFpeKXeZPe7kEuvkUBw,477
-pyg_nightly-2.7.0.dev20250604.dist-info/licenses/LICENSE,sha256=ic-27cMJc1kWoMEYncz3Ya3Ur2Bi3bNLWib2DT763-o,1067
-pyg_nightly-2.7.0.dev20250604.dist-info/WHEEL,sha256=G2gURzTEtmeR8nrdXUJfNiB3VYVxigPQ-bEQujpNiNs,82
-pyg_nightly-2.7.0.dev20250604.dist-info/METADATA,sha256=pOkfyi_H6pwQ7MLQrQ2nbMuuzDu7C1CLo_GNz0l-NsM,62967
-pyg_nightly-2.7.0.dev20250604.dist-info/RECORD,,
+pyg_nightly-2.7.0.dev20250606.dist-info/licenses/LICENSE,sha256=ic-27cMJc1kWoMEYncz3Ya3Ur2Bi3bNLWib2DT763-o,1067
+pyg_nightly-2.7.0.dev20250606.dist-info/WHEEL,sha256=G2gURzTEtmeR8nrdXUJfNiB3VYVxigPQ-bEQujpNiNs,82
+pyg_nightly-2.7.0.dev20250606.dist-info/METADATA,sha256=DZmmGP4X5ZCOarKh3B4kkaqENUf31OB0kSOV_zeizTI,62967
+pyg_nightly-2.7.0.dev20250606.dist-info/RECORD,,

torch_geometric/__init__.py

@@ -31,7 +31,7 @@ from .lazy_loader import LazyLoader
 contrib = LazyLoader('contrib', globals(), 'torch_geometric.contrib')
 graphgym = LazyLoader('graphgym', globals(), 'torch_geometric.graphgym')
 
-__version__ = '2.7.0.dev20250604'
+__version__ = '2.7.0.dev20250606'
 
 __all__ = [
     'Index',

torch_geometric/nn/models/glem.py

@@ -197,7 +197,7 @@ class GLEM(torch.nn.Module):
             optimizer.zero_grad()
             all_out.append(out)
             total_correct += int(out.argmax(dim=-1).eq(labels).sum())
-            total_loss += float(loss)
+            total_loss += float(loss.detach())
             if verbose:
                 pbar.update(batch['n_id'].size(0))
 
@@ -251,7 +251,7 @@ class GLEM(torch.nn.Module):
             loss.backward()
             optimizer.step()
             optimizer.zero_grad()
-            total_loss += float(loss)
+            total_loss += float(loss.detach())
             total_correct += int(out.argmax(dim=-1).eq(labels).sum())
             if verbose:
                 pbar.update(batch.batch_size)

torch_geometric/utils/influence.py

@@ -200,16 +200,24 @@ def total_influence(
     device: Union[torch.device, str] = "cpu",
     vectorize: bool = True,
 ) -> Tuple[Tensor, float]:
-    r"""Compute Jacobian‑based influence aggregates for *multiple* seed nodes.
+    r"""Compute Jacobian‑based influence aggregates for *multiple* seed nodes,
+    as introduced in the
+    `"Towards Quantifying Long-Range Interactions in Graph Machine Learning:
+    a Large Graph Dataset and a Measurement"
+    <https://arxiv.org/abs/2503.09008>`_ paper.
+    This measurement quantifies how a GNN model's output at a node is
+    influenced by features of other nodes at increasing hop distances.
 
-    For every sampled node :math:`v`, this method
+    Specifically, for every sampled node :math:`v`, this method
 
     1. evaluates the **L1‑norm** of the Jacobian of the model output at
        :math:`v` w.r.t. the node features of its *k*-hop induced sub‑graph;
     2. sums these scores **per hop** to obtain the influence vector
        :math:`(I_{0}, I_{1}, \dots, I_{k})`;
     3. optionally averages those vectors over all sampled nodes and
-       (optionally) normalises them by :math:`I_{0}`.
+       optionally normalises them by :math:`I_{0}`.
+
+    Please refer to Section 4 of the paper for a more detailed definition.
 
     Args:
         model (torch.nn.Module): A PyTorch Geometric‑compatible model with
@@ -217,10 +225,10 @@ def total_influence(
         data (torch_geometric.data.Data): Graph data object providing at least
             :obj:`x` (node features) and :obj:`edge_index` (connectivity).
         max_hops (int): Maximum hop distance :math:`k`.
-        num_samples (int, optional): Number of seed nodes to evaluate.
+        num_samples (int, optional): Number of random seed nodes to evaluate.
             If :obj:`None`, all nodes are used. (default: :obj:`None`)
-        normalize (bool, optional): If :obj:`True`, divide each hop‑wise
-            average by the influence of hop 0. (default: :obj:`True`)
+        normalize (bool, optional): If :obj:`True`, normalize each hop‑wise
+            influence by the influence of hop 0. (default: :obj:`True`)
         average (bool, optional): If :obj:`True`, return the hop‑wise **mean**
             over all seed nodes (shape ``[k+1]``).
             If :obj:`False`, return the full influence matrix of shape
@@ -235,8 +243,8 @@ def total_influence(
     Returns:
         Tuple[Tensor, float]:
             * **avg_influence** (*Tensor*):
-              • shape ``[k+1]`` if :obj:`average=True`;
-              • shape ``[N, k+1]`` otherwise.
+              shape ``[k+1]`` if :obj:`average=True`;
+              shape ``[N, k+1]`` otherwise.
             * **R** (*float*): Influence‑weighted receptive‑field breadth
               returned by :func:`influence_weighted_receptive_field`.