`LinkNeighborloader` fails due to `Could not run 'pyg::index_sort' with arguments from the 'CPU' backend`

Question

`LinkNeighborloader` fails due to `Could not run 'pyg::index_sort' with arguments from the 'CPU' backend`

Closed this issue 4 months ago · 19 comments

🐛 Describe the bug

edge_label_index = train_data["user", "rates", "movie"].edge_label_index
edge_label = train_data["user", "rates", "movie"].edge_label

train_loader = LinkNeighborLoader(
    data=train_data,
    num_neighbors=[20, 10],
    neg_sampling_ratio=2.0,
    edge_label_index=(("user", "rates", "movie"), edge_label_index),
    edge_label=edge_label,
    batch_size=128,
    shuffle=True,
)

# Inspect a sample:
sampled_data = next(iter(train_loader))

print("Sampled mini-batch:")
print("===================")
print(sampled_data)

assert sampled_data["user", "rates", "movie"].edge_label_index.size(1) == 3 * 128
assert sampled_data["user", "rates", "movie"].edge_label.min() == 0
assert sampled_data["user", "rates", "movie"].edge_label.max() == 1`

Environment

pyg-lib version:
PyTorch version:
OS:
Python version:
CUDA/cuDNN version:
How you installed PyTorch and pyg-lib (conda, pip, source):
Any other relevant information:

Answer 1 · 2024-05-23T16:27:58.000Z

@rusty1s please resolve my issue as soon as possible, i have my presentation this monday!

Answer 2 · 2024-05-23T16:29:37.000Z

How did you install pyg-lib and which version?

Answer 3 · 2024-05-23T16:35:55.000Z

I have installed using this command:
!pip install --upgrade pyg-lib -f https://data.pyg.org/whl/nightly/torch-2.2.0+cpu.html

Answer 4 · 2024-05-23T16:36:27.000Z

actually since 2months eveyrthing was working fine
just today itself I encountered this error

Answer 5 · 2024-05-23T16:36:52.000Z

I was using the same code ...and everything was working fine but idk how it happened

Answer 6 · 2024-05-23T16:39:29.000Z

@rusty1s I was using the same installation since so many days without facing any errors!!
Using this below command:
!pip install --upgrade pyg-lib -f https://data.pyg.org/whl/nightly/torch-2.2.0+cpu.html

Answer 7 · 2024-05-23T16:49:57.000Z

What's the output of

import pyg_lib
import torch

inputs = torch.randperm(100_000)
pyg_lib.ops.index_sort(inputs)

Does that run through?

Answer 8 · 2024-05-23T16:52:31.000Z

showing the same error with your code snippet too!!!

Answer 9 · 2024-05-23T17:07:33.000Z

@rusty1s please see the above!!!

Answer 10 · 2024-05-23T18:02:26.000Z

For now, you can unblock yourself with

import torch_geometric
torch_geometric.typing.WITH_INDEX_SORT = False

Answer 11 · 2024-05-23T18:07:43.000Z

Still showing the same error!! @rusty1s

Answer 12 · 2024-05-23T19:35:15.000Z

Yes, but the error will disappear inside LinkNeighborLoader (the original issue).

Answer 13 · 2024-05-24T12:45:55.000Z

This is the error I am getting now!!!

Please try to tell the solution asap as I have my conference presentation tomorrow for my accepted research paper!!

Answer 14 · 2024-05-24T17:51:18.000Z

@rusty1s please see the above

Answer 15 · 2024-05-25T19:39:29.000Z

Can you post a reproducible example here (no screenshot)? It's hard to tell what might have gone wrong.

Answer 16 · 2024-05-25T19:39:42.000Z

You can also share the colab with me if that is easier.

Answer 17 · 2024-05-26T13:21:05.000Z

https://colab.research.google.com/drive/1uR19Ym_q4SUpSCR623IF345zTngyaXqW?usp=sharing

I am sharing the link for colab notebook @rusty1s

Answer 18 · 2024-05-27T07:50:36.000Z

The issue is that you are re-installing torch. Colab comes with PyTorch pre-installed, so simply do

!pip install torch_geometric torch_scatter torch_sparse torch_cluster torch_spline_conv -f https://data.pyg.org/whl/torch-2.3.0+cu121.html
!pip install --upgrade pyg-lib -f https://data.pyg.org/whl/nightly/torch-2.3.0+cu121.html

import pyg_lib
inputs = torch.randperm(100_000)
pyg_lib.ops.index_sort(inputs)

then works

Answer 19 · 2024-05-27T13:46:31.000Z

okayy thank you so much!!! its working now