torch.distributions.utils.clamp_probs

torch.distributions.utils.clamp_probs(probs)

Clamps the probabilities to be in the open interval (0, 1).

The probabilities would be clamped between eps and 1 - eps, and eps would be the smallest representable positive number for the input data type.

Parameters:

probs (Tensor) – A tensor of probabilities.

Returns:

The clamped probabilities.

Return type:

Tensor

Examples

>>> probs = torch.tensor([0.0, 0.5, 1.0])
>>> clamp_probs(probs)
tensor([1.1921e-07, 5.0000e-01, 1.0000e+00])

>>> probs = torch.tensor([0.0, 0.5, 1.0], dtype=torch.float64)
>>> clamp_probs(probs)
tensor([2.2204e-16, 5.0000e-01, 1.0000e+00], dtype=torch.float64)