mirror of
https://github.com/pytorch/pytorch.git
synced 2025-10-20 21:14:14 +08:00
40576bceaf
This PR fixes #69466 and introduces some other minor changes. Tests are somewhat more involved because a reference implementation in `scipy` is not available; tests proceed differently for discrete and continuous distributions. For continuous distributions, we evaluate the gradient of the `log_prob` at the mode. Tests pass if the gradient is zero OR (the mode is at the boundary of the support of the distribution AND the `log_prob` decreases as we move away from the boundary to the interior of the support). For discrete distributions, the notion of a gradient is not well defined. We thus "look" ahead and behind one step (e.g. if the mode of a Poisson distribution is 9, we consider 8 and 10). If the step ahead/behind is still within the support of the distribution, we assert that the `log_prob` is smaller than at the mode. For one-hot encoded distributions (currently just `OneHotCategorical`), we evaluate the underlying mode (i.e. encoded as an integral tensor), "advance" by one label to get another sample that should have lower probability using `other = (mode + 1) % event_size` and re-encode as one-hot. The resultant `other` sample should have lower probability than the mode. Furthermore, Gamma, half Cauchy, and half normal distributions have their support changed from positive to nonnegative. This change is necessary because the mode of the "half" distributions is zero, and the mode of the gamma distribution is zero for `concentration <= 1`. cc @fritzo Pull Request resolved: https://github.com/pytorch/pytorch/pull/76690 Approved by: https://github.com/neerajprad
59 lines
1.8 KiB
Python
59 lines
1.8 KiB
Python
from torch.distributions import constraints
|
|
from torch.distributions.transforms import ExpTransform
|
|
from torch.distributions.normal import Normal
|
|
from torch.distributions.transformed_distribution import TransformedDistribution
|
|
|
|
|
|
class LogNormal(TransformedDistribution):
|
|
r"""
|
|
Creates a log-normal distribution parameterized by
|
|
:attr:`loc` and :attr:`scale` where::
|
|
|
|
X ~ Normal(loc, scale)
|
|
Y = exp(X) ~ LogNormal(loc, scale)
|
|
|
|
Example::
|
|
|
|
>>> m = LogNormal(torch.tensor([0.0]), torch.tensor([1.0]))
|
|
>>> m.sample() # log-normal distributed with mean=0 and stddev=1
|
|
tensor([ 0.1046])
|
|
|
|
Args:
|
|
loc (float or Tensor): mean of log of distribution
|
|
scale (float or Tensor): standard deviation of log of the distribution
|
|
"""
|
|
arg_constraints = {'loc': constraints.real, 'scale': constraints.positive}
|
|
support = constraints.positive
|
|
has_rsample = True
|
|
|
|
def __init__(self, loc, scale, validate_args=None):
|
|
base_dist = Normal(loc, scale, validate_args=validate_args)
|
|
super(LogNormal, self).__init__(base_dist, ExpTransform(), validate_args=validate_args)
|
|
|
|
def expand(self, batch_shape, _instance=None):
|
|
new = self._get_checked_instance(LogNormal, _instance)
|
|
return super(LogNormal, self).expand(batch_shape, _instance=new)
|
|
|
|
@property
|
|
def loc(self):
|
|
return self.base_dist.loc
|
|
|
|
@property
|
|
def scale(self):
|
|
return self.base_dist.scale
|
|
|
|
@property
|
|
def mean(self):
|
|
return (self.loc + self.scale.pow(2) / 2).exp()
|
|
|
|
@property
|
|
def mode(self):
|
|
return (self.loc - self.scale.square()).exp()
|
|
|
|
@property
|
|
def variance(self):
|
|
return (self.scale.pow(2).exp() - 1) * (2 * self.loc + self.scale.pow(2)).exp()
|
|
|
|
def entropy(self):
|
|
return self.base_dist.entropy() + self.loc
|