torcheval.metrics.functional.reciprocal_rank

torcheval.metrics.functional.reciprocal_rank(input: Tensor, target: Tensor, *, k: int | None = None) → Tensor

Compute the reciprocal rank of the correct class among the top predicted classes. Its class version is torcheval.metrics.ReciprocalRank.

Parameters:

• input (Tensor) – Predicted unnormalized scores (often referred to as logits) or class probabilities of shape (num_samples, num_classes).

• target (Tensor) – Ground truth class indices of shape (num_samples,).

• k (int, optional) – Number of top class probabilities to be considered.

Examples:

>>> import torch
>>> from torcheval.metrics.functional import reciprocal_rank
>>> input = torch.tensor([[0.3, 0.1, 0.6], [0.5, 0.2, 0.3], [0.2, 0.1, 0.7], [0.3, 0.3, 0.4]])
>>> target = torch.tensor([2, 1, 1, 0])
>>> reciprocal_rank(input, target)
tensor([1.0000, 0.3333, 0.3333, 0.5000])
>>> reciprocal_rank(input, target, k=2)
tensor([1.0000, 0.0000, 0.0000, 0.5000])