IndexError: tensors used as indices must be long, byte or bool tensors

[atale480]

I wrote a small GCN code and got above error. My code is as follows

import torch
import torch.nn as nn
from torch.nn import ReLU
from torch_geometric.nn import Sequential, GCNConv, BatchNorm

class GCNConvBNRelu(nn.Module):
    def __init__(self, channel_in : 'Number of diamensions of input',
    channel_out: 'Number of out channels'):
        super(GCNConvBNRelu, self).__init__()

        self.layers = Sequential('x, edge_index',[
            (GCNConv(channel_in, channel_out), 'x, edge_index -> x'),
            (BatchNorm(channel_out)),
            ReLU(inplace=True),
        ])
    
    def forward(self,x,edge_index):
        x = self.layers(x,edge_index)

        return x

a = GCNConvBNRelu(3, 64)
a(dataset[0].x, dataset[0].edge_index)

print(dataset[0]) gives Data(x=[1343, 3], edge_index=[2, 8046], y=[1])

Error is given for x = self.layers(x, edge_index) line as tensors used as indices must be long, byte or bool tensors
I'm not sure what's wrong in this program.

System Info:

Windows 10 Pro
cuda version :   11.6
PyTorch version :   1.12.1
PyTorch Geometric :  2.2.0
Python :   3.8

[wsad1]

Could you share the full error message. Also is your edge_index of type float, if so try converting it to torch.long.

[atale480]

Here is the full error message
for more information I've added 3 print statements in program which prints dataset[0], dataset[0].x and dataset[0].edge_index. You can see those tensors before error along with respective label.

(pytorch_geometric) F:\Rohan\MDCGCN-HiDDeN>python tmp.py

dataset[0] => Data(x=[4706, 3], edge_index=[2, 28224], y=[1])

dataset[0].x => tensor([[ 0.1445, -0.0143,  0.3818],
        [-0.1538, -0.0457,  0.3610],
        [-0.1539, -0.0573,  0.3713],
        ...,
        [ 0.0581, -0.0246,  0.7790],
        [ 0.0685, -0.0210,  0.7644],
        [ 0.0610, -0.0379,  0.7644]])

dataset[0].edge_index => tensor([[   0,    0,    0,  ..., 4705, 4705, 4705],
        [ 455,  892, 3511,  ..., 4695, 4703, 4704]], dtype=torch.int32)

Traceback (most recent call last):
  File "tmp.py", line 30, in <module>
    a(dataset[0].x, dataset[0].edge_index)
  File "C:\Users\DELL\anaconda3\envs\pytorch_geometric\lib\site-packages\torch\nn\modules\module.py", line 1130, in _call_impl
    return forward_call(*input, **kwargs)
  File "tmp.py", line 19, in forward
    x = self.layers(x,edge_index)
  File "C:\Users\DELL\anaconda3\envs\pytorch_geometric\lib\site-packages\torch\nn\modules\module.py", line 1130, in _call_impl
    return forward_call(*input, **kwargs)
  File "C:\Users\DELL\AppData\Local\Temp\DELL_pyg\tmpvhlh2w_d.py", line 18, in forward
    x = self.module_0(x, edge_index)
  File "C:\Users\DELL\anaconda3\envs\pytorch_geometric\lib\site-packages\torch\nn\modules\module.py", line 1130, in _call_impl
    return forward_call(*input, **kwargs)
  File "C:\Users\DELL\anaconda3\envs\pytorch_geometric\lib\site-packages\torch_geometric\nn\conv\gcn_conv.py", line 175, in forward
    edge_index, edge_weight = gcn_norm(  # yapf: disable
  File "C:\Users\DELL\anaconda3\envs\pytorch_geometric\lib\site-packages\torch_geometric\nn\conv\gcn_conv.py", line 60, in gcn_norm
    edge_index, tmp_edge_weight = add_remaining_self_loops(
  File "C:\Users\DELL\anaconda3\envs\pytorch_geometric\lib\site-packages\torch_geometric\utils\loop.py", line 224, in add_remaining_self_loops
    loop_attr[edge_index[0][inv_mask]] = edge_attr[inv_mask]
IndexError: tensors used as indices must be long, byte or bool tensors

[rusty1s]

Can you convert your edge_index to torch.int64`?

[atale480]

Yes that did the trick. Worked like charm. Thank You