Inconsistent behavior for dataset.num_classes and dataset.num_features

[EdisonLeeeee]

Hi, I am using PyG datasets with my custom transform function, but I found there is an inconsistent behavior between dataset.num_classes and dataset.num_features. The minimal code to reproduce my scenario is as follows:

import torch
from torch_geometric.datasets import Planetoid

def my_transform(data):
    data.x = torch.randn(100, 10) # num_features changed here
    data.y = torch.arange(100) # num_classes changed here
    return data

dataset = Planetoid('~/data/Cora', 'Cora')
print(f"num_features: {dataset.num_features}, num_classes: {dataset.num_classes}")
# num_features: 1433, num_classes: 7

dataset = Planetoid('~/data/Cora', 'Cora', transform=my_transform)
print(f"num_features: {dataset.num_features}, num_classes: {dataset.num_classes}")
# num_features: 10, num_classes: 7

That is, dataset.num_features returns the results from the transformed dataset while dataset.num_classes does not. I am not sure if this is intended behavior or a potential bug.

[rusty1s]

Thanks for reporting. I just fixed that here: #5274

[EdisonLeeeee]

Great, thanks!