GH-3474: add random seed parameter to dataset splitting and downsampl…

[MattGPT-ai]

#3474

Adds random seed parameters to dataset splitting and downsampling functions

[helpmefindaname]

Hi @MattGPT-ai
thank you for your contribution.
The changes look good to me, I just requests a small change. Besides that, I think this is ready to merge

[MattGPT-ai]

Thanks for your review! Pushed the changes you requested

[MattGPT-ai]

Thanks for the approval! How can we trigger the checks to run so that it can be merged?

[alanakbik]

Thanks for the PR! Looks good, but there is a small mypy error (I added a suggestion to fix).

[MattGPT-ai]

Applied the change. I had a question, I saw that step is necessary because Dataset doesn't have __len__ implemented. Is there a way to implement __len__, even if it just calls _len_dataset? Does constructing the DatasetLoader class take any non-negligible time? I actually see the same warning for return len(loader) in my PyCharm that it "Expected type Sized, got DataLoader instead", and a similar warning for _iter_dataset

[alanakbik]

Thanks! I agree that it would be better to implement __len__ rather than having to call _len_dataset.

The problem is that Dataset is a PyTorch class that does not have __len__ or __iter__ implemented, so we can't add them here. Alternatively, we might change the code that instead of Dataset, everything is based on either PyTorch's IterDataset or our own FlairDataset.

[alanakbik]

@MattGPT-ai thanks a lot for adding this!