torch.histogramdd Inconsistent/Incorrect Documentation

[daohanlu]

📚 The doc issue

The doc for torch.histogramdd says that "every bin is exclusive of its left edge. Only the rightmost bin is inclusive of its right edge." However, running some code suggests that every bin is inclusive of its left edge.

-> torch.histogramdd(torch.tensor([[0.0, 0.0], [0.5, 0.5], [1.0, 1.0]]), range=(0, 1.0, 0, 1.0), bins=2)
torch.return_types.histogramdd(
hist=tensor([[1., 0.],
        [0., 2.]]),
bin_edges=(tensor([0.0000, 0.5000, 1.0000]), tensor([0.0000, 0.5000, 1.0000])))

That is, [0,0] gets sent to the [0,0]th bin and both [0.5, 0.5] and [1.0, 1.0] get sent to the [1,1]th bin.

Additionally, I believe what the documentation says is not logically sound. If "every bin is exclusive of its left edge" and all but the rightmost bin is exclusive of its right edge, then there'd be holes in between the bins.

Suggest a potential alternative/fix

Change the documentation to instead say "every bin is inclusive of its left edge."

cc @svekars @brycebortree @albanD

[jbschlosser]

Thanks for the report! We'd accept a PR fixing the docs to match the code behavior.