[feature request] Convert "indices" variable in "torch.utils.data.dataset.random_split" to list

[floatn]

Issue description

random_split fills indices variable using torch.randperm function, which returns tensor. Through Subset class its values goes to Dataset.getitem. If I override getitem, tensor values are ok for indexing list, but with pandas DataFrame it raises TypeError. Maybe it would be better to convert indices to list variable at start.

Code example

df = pd.DataFrame([1, 2])
tn = torch.tensor([0, 1])
df.iloc[tn[0]]
...
...
TypeError: len() of unsized object

[fmassa]

I think we can just add a .tolist() after the code in here. Could you send a PR?

[floatn]

Done. #9237 .

[soumith]

closed via the discussion in #9237