xref: /aosp_15_r20/external/pytorch/torch/distributions/weibull.py (revision da0073e96a02ea20f0ac840b70461e3646d07c45) 
1# mypy: allow-untyped-defs
2import torch
3from torch.distributions import constraints
4from torch.distributions.exponential import Exponential
5from torch.distributions.gumbel import euler_constant
6from torch.distributions.transformed_distribution import TransformedDistribution
7from torch.distributions.transforms import AffineTransform, PowerTransform
8from torch.distributions.utils import broadcast_all
9
10
11__all__ = ["Weibull"]
12
13
14class Weibull(TransformedDistribution):
15    r"""
16    Samples from a two-parameter Weibull distribution.
17
18    Example:
19
20        >>> # xdoctest: +IGNORE_WANT("non-deterministic")
21        >>> m = Weibull(torch.tensor([1.0]), torch.tensor([1.0]))
22        >>> m.sample()  # sample from a Weibull distribution with scale=1, concentration=1
23        tensor([ 0.4784])
24
25    Args:
26        scale (float or Tensor): Scale parameter of distribution (lambda).
27        concentration (float or Tensor): Concentration parameter of distribution (k/shape).
28    """
29    arg_constraints = {
30        "scale": constraints.positive,
31        "concentration": constraints.positive,
32    }
33    support = constraints.positive
34
35    def __init__(self, scale, concentration, validate_args=None):
36        self.scale, self.concentration = broadcast_all(scale, concentration)
37        self.concentration_reciprocal = self.concentration.reciprocal()
38        base_dist = Exponential(
39            torch.ones_like(self.scale), validate_args=validate_args
40        )
41        transforms = [
42            PowerTransform(exponent=self.concentration_reciprocal),
43            AffineTransform(loc=0, scale=self.scale),
44        ]
45        super().__init__(base_dist, transforms, validate_args=validate_args)
46
47    def expand(self, batch_shape, _instance=None):
48        new = self._get_checked_instance(Weibull, _instance)
49        new.scale = self.scale.expand(batch_shape)
50        new.concentration = self.concentration.expand(batch_shape)
51        new.concentration_reciprocal = new.concentration.reciprocal()
52        base_dist = self.base_dist.expand(batch_shape)
53        transforms = [
54            PowerTransform(exponent=new.concentration_reciprocal),
55            AffineTransform(loc=0, scale=new.scale),
56        ]
57        super(Weibull, new).__init__(base_dist, transforms, validate_args=False)
58        new._validate_args = self._validate_args
59        return new
60
61    @property
62    def mean(self):
63        return self.scale * torch.exp(torch.lgamma(1 + self.concentration_reciprocal))
64
65    @property
66    def mode(self):
67        return (
68            self.scale
69            * ((self.concentration - 1) / self.concentration)
70            ** self.concentration.reciprocal()
71        )
72
73    @property
74    def variance(self):
75        return self.scale.pow(2) * (
76            torch.exp(torch.lgamma(1 + 2 * self.concentration_reciprocal))
77            - torch.exp(2 * torch.lgamma(1 + self.concentration_reciprocal))
78        )
79
80    def entropy(self):
81        return (
82            euler_constant * (1 - self.concentration_reciprocal)
83            + torch.log(self.scale * self.concentration_reciprocal)
84            + 1
85        )
86