mirror of
https://github.com/zebrajr/pytorch.git
synced 2025-12-07 12:21:27 +01:00
2431eac7c0
Summary: This adds tests in tests/test_distributions.py to ensure that all methods of `Distribution` objects are jittable. I've replaced a few samplers with jittable versions: - `.uniform_()` -> `torch.rand()` - `.exponential_()` -> `-(-torch.rand()).log1p()` - `.normal_()` -> `torch.normal(torch.zeros(...), torch.ones(...), ...)` Some jit failures remain, and are marked in test_distributions.py - `Cauchy` and `HalfCauchy` do not support sampling due to missing `.cauchy_()` - `Binomial` does not support `.enumerate_support()` due to `arange` ignoring its first arg. - `MultivariateNormal`, `LowRankMultivariateNormal` do not support `.mean`, `.entropy` - [x] Currently some tests fail (I've skipped those) due to unavailability of `aten::uniform` and `aten::cauchy` in the jit. Can someone suggest how to add these? I tried to add declarations to `torch/csrc/ir.cpp` and `torch/csrc/passes/shape_analysis.cpp`, but that resulted in "Couldn't find operator" errors. - [x] There are still lots of `TracerWarning`s that something doesn't match something. I'm not sure whether these are real. Pull Request resolved: https://github.com/pytorch/pytorch/pull/11560 Differential Revision: D9816327 Pulled By: apaszke fbshipit-source-id: 72ec998ea13fc4c76d1ed003d9502e0fbaf728b8
84 lines
2.5 KiB
Python
84 lines
2.5 KiB
Python
from numbers import Number
|
|
|
|
import torch
|
|
from torch.distributions import constraints
|
|
from torch.distributions.exp_family import ExponentialFamily
|
|
from torch.distributions.utils import broadcast_all
|
|
|
|
|
|
class Exponential(ExponentialFamily):
|
|
r"""
|
|
Creates a Exponential distribution parameterized by :attr:`rate`.
|
|
|
|
Example::
|
|
|
|
>>> m = Exponential(torch.tensor([1.0]))
|
|
>>> m.sample() # Exponential distributed with rate=1
|
|
tensor([ 0.1046])
|
|
|
|
Args:
|
|
rate (float or Tensor): rate = 1 / scale of the distribution
|
|
"""
|
|
arg_constraints = {'rate': constraints.positive}
|
|
support = constraints.positive
|
|
has_rsample = True
|
|
_mean_carrier_measure = 0
|
|
|
|
@property
|
|
def mean(self):
|
|
return self.rate.reciprocal()
|
|
|
|
@property
|
|
def stddev(self):
|
|
return self.rate.reciprocal()
|
|
|
|
@property
|
|
def variance(self):
|
|
return self.rate.pow(-2)
|
|
|
|
def __init__(self, rate, validate_args=None):
|
|
self.rate, = broadcast_all(rate)
|
|
batch_shape = torch.Size() if isinstance(rate, Number) else self.rate.size()
|
|
super(Exponential, self).__init__(batch_shape, validate_args=validate_args)
|
|
|
|
def expand(self, batch_shape, _instance=None):
|
|
new = self._get_checked_instance(Exponential, _instance)
|
|
batch_shape = torch.Size(batch_shape)
|
|
new.rate = self.rate.expand(batch_shape)
|
|
super(Exponential, new).__init__(batch_shape, validate_args=False)
|
|
new._validate_args = self._validate_args
|
|
return new
|
|
|
|
def rsample(self, sample_shape=torch.Size()):
|
|
shape = self._extended_shape(sample_shape)
|
|
if torch._C._get_tracing_state():
|
|
# [JIT WORKAROUND] lack of support for ._exponential()
|
|
u = torch.rand(shape, dtype=self.rate.dtype, device=self.rate.device)
|
|
return -(-u).log1p() / self.rate
|
|
return self.rate.new(shape).exponential_() / self.rate
|
|
|
|
def log_prob(self, value):
|
|
if self._validate_args:
|
|
self._validate_sample(value)
|
|
return self.rate.log() - self.rate * value
|
|
|
|
def cdf(self, value):
|
|
if self._validate_args:
|
|
self._validate_sample(value)
|
|
return 1 - torch.exp(-self.rate * value)
|
|
|
|
def icdf(self, value):
|
|
if self._validate_args:
|
|
self._validate_sample(value)
|
|
return -torch.log(1 - value) / self.rate
|
|
|
|
def entropy(self):
|
|
return 1.0 - torch.log(self.rate)
|
|
|
|
@property
|
|
def _natural_params(self):
|
|
return (-self.rate, )
|
|
|
|
def _log_normalizer(self, x):
|
|
return -torch.log(-x)
|