Merge pull request #6 from cms-pepr/gravnet_jit

Minor changes so that gravnet will jit out of the box

torch_cmspepr/gravnet_model.py

@@ -16,18 +16,18 @@ def global_exchange(x: Tensor, batch: Tensor) -> Tensor:
     """
     n_hits_per_event = scatter_count(batch)
     n_hits, n_features = x.size()
-    batch_size = batch.max()+1
+    batch_size = int(batch.max()) + 1
 
     # minmeanmax: (batch_size x 3*n_features)
     meanminmax = torch.cat((
         scatter_mean(x, batch, dim=0),
         scatter_min(x, batch, dim=0)[0],
         scatter_max(x, batch, dim=0)[0]
         ), dim=1)
-    assert meanminmax.size() == (batch_size, 3*n_features)
+    assert list(meanminmax.size()) == [batch_size, 3*n_features]
 
     meanminmax = torch.repeat_interleave(meanminmax, n_hits_per_event, dim=0)
-    assert meanminmax.size() == (n_hits, 3*n_features)
+    assert list(meanminmax.size()) == [n_hits, 3*n_features]
 
     out = torch.cat((meanminmax, x), dim=1)
     assert out.size() == (n_hits, 4*n_features)
@@ -68,7 +68,7 @@ def __init__(
         self.gravnet_layer = GravNetConv(
                 in_channels, out_channels,
                 space_dimensions, propagate_dimensions, k
-                )
+                ).jittable()
         self.post_gravnet = nn.Sequential(
             nn.BatchNorm1d(out_channels),
             nn.Linear(out_channels, 128),