AttributeError: 'torch.dtype' object has no attribute 'kind'

[Coluding]

Hi!

I am getting with torch=2.5.1, torch_geometric=2.4.0 and torch_geometric_temporal=0.54.0 and the following simple script:

from torch_geometric_temporal.signal import DynamicGraphTemporalSignal
import torch
import torch_geometric_temporal
import torch_geometric

print(torch_geometric_temporal.__version__)
print(torch_geometric.__version__)

# Node features for each time step (varying number of nodes)
node_features = [
    torch.tensor([[1, 2], [3, 4], [5, 6]], dtype=torch.float), 
    torch.tensor([[2, 3], [4, 5]], dtype=torch.float),         
    torch.tensor([[3, 4], [5, 6], [7, 8], [9, 10]], dtype=torch.float) 
]

# Edge indices for each time step
edge_indices = [
    torch.tensor([[0, 1], [1, 2]], dtype=torch.long).t(),  
    torch.tensor([[0, 1]], dtype=torch.long).t(),          
    torch.tensor([[0, 1], [1, 2], [2, 3]], dtype=torch.long).t()  
]

# Define edge attributes (optional)
edge_attributes = [
    torch.tensor([[0.1], [0.2]], dtype=torch.float), 
    torch.tensor([[0.3]], dtype=torch.float),        
    torch.tensor([[0.5], [0.6], [0.7]], dtype=torch.float)  
]

# Define targets for each time step (optional)
targets = [
    torch.tensor([0.5, 0.6, 0.7], dtype=torch.float), 
    torch.tensor([0.6, 0.7], dtype=torch.float),     
    torch.tensor([0.7, 0.8, 0.9, 1.0], dtype=torch.float)  
]

# Create a DynamicGraphTemporalSignal object
temporal_graph = DynamicGraphTemporalSignal(edge_indices, node_features, edge_attributes, targets)

print(temporal_graph[0])

The following error: AttributeError: 'torch.dtype' object has no attribute 'kind' stemming from the getitem call in temporal_graph[0].

It comes from this part in DynamicGraphTemporalSignal:

    def _get_target(self, time_index: int):
        if self.targets[time_index] is None:
            return self.targets[time_index]
        else:
            if not self.targets[time_index].dtype.kind == "i":
                return torch.LongTensor(self.targets[time_index])
            elif not self.targets[time_index].dtype.kind == "f":
                return torch.FloatTensor(self.targets[time_index])

Switching it to:

    def _get_target(self, time_index: int):
        if self.targets[time_index] is None:
            return self.targets[time_index]
        else:
            if not self.targets[time_index].dtype.is_floating_point:
                return torch.FloatTensor(self.targets[time_index])
            else:
                return torch.LongTensor(self.targets[time_index])

fixed it.

This is a probably a torch version mismatch. Can you validate my versions? Otherwise I will just keep the workaround.

Best regards and thank you!