I got an error stating "Expected all tensors to be on the same device".

[arymandeshwal]

I am training a GNN model and while explaining the model via GNNExplainer, I got the following error. Could someone please help me out with it:

RuntimeError Traceback (most recent call last)
Cell In[11], line 19
1 explainer = Explainer(
2 model=model,
3 algorithm=CaptumExplainer('IntegratedGradients'),
(...)
15 ),
16 )
18 index = torch.tensor([500,1000, 5, 100, 10000, 80, 0]) # Explain edge labels with index 2 and 10.
---> 19 explanation = explainer(
20 train_data.x_dict,
21 train_data.edge_index_dict,
22 index=index,
23 edge_label_index=train_data['user',"rates" ,'movie'].edge_label_index,
24 )
25 print(f'Generated explanations in {explanation.available_explanations}')
27 path = 'feature_importance.png'

File c:\users\admin'\desktop\course exercises\xai miniproject.env\lib\site-packages\torch_geometric\explain\explainer.py:192, in Explainer.call(self, x, edge_index, target, index, **kwargs)
188 if target is not None:
189 warnings.warn(
190 f"The 'target' should not be provided for the explanation "
191 f"type '{self.explanation_type.value}'")
--> 192 prediction = self.get_prediction(x, edge_index, **kwargs)
193 target = self.get_target(prediction)
195 training = self.model.training

File c:\users\admin'\desktop\course exercises\xai miniproject.env\lib\site-packages\torch\utils_contextlib.py:115, in context_decorator..decorate_context(*args, **kwargs)
112 @functools.wraps(func)
113 def decorate_context(*args, **kwargs):
114 with ctx_factory():
--> 115 return func(*args, **kwargs)

File c:\users\admin'\desktop\course exercises\xai miniproject.env\lib\site-packages\torch_geometric\explain\explainer.py:115, in Explainer.get_prediction(self, *args, **kwargs)
112 self.model.eval()
114 with torch.no_grad():
--> 115 out = self.model(*args, **kwargs)
117 self.model.train(training)
119 return out

File c:\users\admin'\desktop\course exercises\xai miniproject.env\lib\site-packages\torch\nn\modules\module.py:1501, in Module._call_impl(self, *args, **kwargs)
1496 # If we don't have any hooks, we want to skip the rest of the logic in
1497 # this function, and just call forward.
1498 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
1499 or _global_backward_pre_hooks or _global_backward_hooks
1500 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1501 return forward_call(*args, **kwargs)
1502 # Do not call functions when jit is used
1503 full_backward_hooks, non_full_backward_hooks = [], []

Cell In[4], line 52, in Model.forward(self, x_dict, edge_index_dict, edge_label_index)
51 def forward(self, x_dict, edge_index_dict, edge_label_index):
---> 52 z_dict = self.encoder(x_dict, edge_index_dict)
53 return self.decoder(z_dict, edge_label_index)

File c:\users\admin'\desktop\course exercises\xai miniproject.env\lib\site-packages\torch\fx\graph_module.py:662, in GraphModule.recompile..call_wrapped(self, *args, **kwargs)
661 def call_wrapped(self, *args, **kwargs):
--> 662 return self._wrapped_call(self, *args, **kwargs)

File c:\users\admin'\desktop\course exercises\xai miniproject.env\lib\site-packages\torch\fx\graph_module.py:281, in _WrappedCall.call(self, obj, *args, **kwargs)
279 raise e.with_traceback(None)
280 else:
--> 281 raise e

File c:\users\admin'\desktop\course exercises\xai miniproject.env\lib\site-packages\torch\fx\graph_module.py:271, in _WrappedCall.call(self, obj, *args, **kwargs)
269 return self.cls_call(obj, *args, **kwargs)
270 else:
--> 271 return super(self.cls, obj).call(*args, **kwargs) # type: ignore[misc]
272 except Exception as e:
273 assert e.traceback

File c:\users\admin'\desktop\course exercises\xai miniproject.env\lib\site-packages\torch\nn\modules\module.py:1501, in Module._call_impl(self, *args, **kwargs)
1496 # If we don't have any hooks, we want to skip the rest of the logic in
1497 # this function, and just call forward.
1498 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
1499 or _global_backward_pre_hooks or _global_backward_hooks
1500 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1501 return forward_call(*args, **kwargs)
1502 # Do not call functions when jit is used
1503 full_backward_hooks, non_full_backward_hooks = [], []

File <eval_with_key>.1:11, in forward(self, x, edge_index)
9 edge_index__user__rates__movie = edge_index_dict.get(('user', 'rates', 'movie'), None)
10 edge_index__movie__rev_rates__user = edge_index_dict.get(('movie', 'rev_rates', 'user'), None); edge_index_dict = None
---> 11 conv1__movie = self.conv1.user__rates__movie((x__user, x__movie), edge_index__user__rates__movie)
12 conv1__user = self.conv1.movie__rev_rates__user((x__movie, x__user), edge_index__movie__rev_rates__user); x__movie = x__user = None
13 relu__user = conv1__user.relu(); conv1__user = None

File c:\users\admin'\desktop\course exercises\xai miniproject.env\lib\site-packages\torch\nn\modules\module.py:1501, in Module._call_impl(self, *args, **kwargs)
1496 # If we don't have any hooks, we want to skip the rest of the logic in
1497 # this function, and just call forward.
1498 if not (self._backward_hooks or self._backward_pre_hooks or self._forward_hooks or self._forward_pre_hooks
1499 or _global_backward_pre_hooks or _global_backward_hooks
1500 or _global_forward_hooks or _global_forward_pre_hooks):
-> 1501 return forward_call(*args, **kwargs)
1502 # Do not call functions when jit is used
1503 full_backward_hooks, non_full_backward_hooks = [], []

File c:\users\admin'\desktop\course exercises\xai miniproject.env\lib\site-packages\torch_geometric\nn\conv\sage_conv.py:130, in SAGEConv.forward(self, x, edge_index, size)
127 x = (self.lin(x[0]).relu(), x[1])
129 # propagate_type: (x: OptPairTensor)
--> 130 out = self.propagate(edge_index, x=x, size=size)
131 out = self.lin_l(out)
133 x_r = x[1]

File c:\users\admin'\desktop\course exercises\xai miniproject.env\lib\site-packages\torch_geometric\nn\conv\message_passing.py:455, in MessagePassing.propagate(self, edge_index, size, **kwargs)
452 for arg in decomp_args:
453 kwargs[arg] = decomp_kwargs[arg][i]
--> 455 coll_dict = self._collect(self._user_args, edge_index, size,
456 kwargs)
458 msg_kwargs = self.inspector.distribute('message', coll_dict)
459 for hook in self._message_forward_pre_hooks.values():

File c:\users\admin'\desktop\course exercises\xai miniproject.env\lib\site-packages\torch_geometric\nn\conv\message_passing.py:329, in MessagePassing._collect(self, args, edge_index, size, kwargs)
327 if isinstance(data, Tensor):
328 self._set_size(size, dim, data)
--> 329 data = self._lift(data, edge_index, dim)
331 out[arg] = data
333 if is_torch_sparse_tensor(edge_index):

File c:\users\admin'\desktop\course exercises\xai miniproject.env\lib\site-packages\torch_geometric\nn\conv\message_passing.py:276, in MessagePassing._lift(self, src, edge_index, dim)
269 raise IndexError(
270 f"Encountered an index error. Please ensure that all "
271 f"indices in 'edge_index' point to valid indices in "
272 f"the interval [0, {src.size(self.node_dim) - 1}] "
273 f"(got interval "
274 f"[{int(index.min())}, {int(index.max())}])")
275 else:
--> 276 raise e
278 if index.numel() > 0 and index.min() < 0:
279 raise ValueError(
280 f"Found negative indices in 'edge_index' (got "
281 f"{index.min().item()}). Please ensure that all "
282 f"indices in 'edge_index' point to valid indices "
283 f"in the interval [0, {src.size(self.node_dim)}) in "
284 f"your node feature matrix and try again.")

File c:\users\admin'\desktop\course exercises\xai miniproject.env\lib\site-packages\torch_geometric\nn\conv\message_passing.py:266, in MessagePassing._lift(self, src, edge_index, dim)
264 try:
265 index = edge_index[dim]
--> 266 return src.index_select(self.node_dim, index)
267 except (IndexError, RuntimeError) as e:
268 if index.min() < 0 or index.max() >= src.size(self.node_dim):

RuntimeError: Expected all tensors to be on the same device, but found at least two devices, cuda:0 and cpu! (when checking argument for argument index in method wrapper_CUDA__index_select)

​

[rusty1s]

Can you confirm that

train_data.x_dict
train_data.edge_index_dict
index
train_data['user',"rates" ,'movie'].edge_label_index

data is all on the same device?