add additional interpolation modes for torch.quantile (#48711)
Summary:
Fix https://github.com/pytorch/pytorch/issues/48523
Related https://github.com/pytorch/pytorch/issues/38349
**BC-breaking Note:**
This PR updates PyTorch's quantile function to add additional interpolation methods `lower`, `higher`, `nearest`, and `midpoint`, and these interpolation methods are currently supported by NumPy.
New parameter `interpolation` is added to the signature for both `torch.quantile` and `torch.nanquantile` functions.
- `quantile(input, q, dim=None, interpolation='linear', keepdim=False, *, out=None) -> Tensor`
- `nanquantile(input, q, dim=None, interpolation='linear', keepdim=False, *, out=None) -> Tensor`
Function signatures followed the NumPy-like style for the moment, keeping `out` at the end to be consistent with PyTorch.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/48711
Reviewed By: H-Huang
Differential Revision: D25428587
Pulled By: heitorschueroff
fbshipit-source-id: e98d24f6a651d302eb94f4ff4da18e38bdbf0124
