# Copyright 2020 Huawei Technologies Co., Ltd
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ============================================================================
"""Gamma Distribution"""
import numpy as np
from mindspore.ops import operations as P
from mindspore.ops import composite as C
import mindspore.nn as nn
from mindspore._checkparam import Validator
from mindspore.common import dtype as mstype
from .distribution import Distribution
from ._utils.utils import check_greater_zero, check_distribution_name
from ._utils.custom_ops import log_generic
[文档]class Gamma(Distribution):
r"""
Gamma distribution.
A Gamma distributio is a continuous distribution with the range :math:`[0, 1]`
and the probability density function:
.. math::
f(x, \alpha, \beta) = \beta^\alpha / \Gamma(\alpha) x^{\alpha - 1} \exp(-\beta x).
where :math:`G` is the Gamma function,
and :math:`\alpha, \beta` are the concentration and the rate of the distribution respectively.
Args:
concentration (int, float, list, numpy.ndarray, Tensor): The concentration,
also know as alpha of the Gamma distribution. Default: None.
rate (int, float, list, numpy.ndarray, Tensor): The rate, also know as
beta of the Gamma distribution. Default: None.
seed (int): The seed used in sampling. The global seed is used if it is None. Default: None.
dtype (mindspore.dtype): The type of the event samples. Default: mstype.float32.
name (str): The name of the distribution. Default: 'Gamma'.
Supported Platforms:
``Ascend``
Note:
`concentration` and `rate` must be greater than zero.
`dist_spec_args` are `concentration` and `rate`.
`dtype` must be a float type because Gamma distributions are continuous.
Raises:
ValueError: When concentration <= 0 or rate <= 0.
TypeError: When the input `dtype` is not a subclass of float.
Examples:
>>> import mindspore
>>> import mindspore.nn as nn
>>> import mindspore.nn.probability.distribution as msd
>>> from mindspore import Tensor
>>> # To initialize a Gamma distribution of the concentration 3.0 and the rate 4.0.
>>> g1 = msd.Gamma([3.0], [4.0], dtype=mindspore.float32)
>>> # A Gamma distribution can be initialized without arguments.
>>> # In this case, `concentration` and `rate` must be passed in through arguments.
>>> g2 = msd.Gamma(dtype=mindspore.float32)
>>> # Here are some tensors used below for testing
>>> value = Tensor([1.0, 2.0, 3.0], dtype=mindspore.float32)
>>> concentration_a = Tensor([2.0], dtype=mindspore.float32)
>>> rate_a = Tensor([2.0, 2.0, 2.0], dtype=mindspore.float32)
>>> concentration_b = Tensor([1.0], dtype=mindspore.float32)
>>> rate_b = Tensor([1.0, 1.5, 2.0], dtype=mindspore.float32)
>>>
>>> # Private interfaces of probability functions corresponding to public interfaces, including
>>> # `prob`, `log_prob`, `cdf`, `log_cdf`, `survival_function`, and `log_survival`,
>>> # have the same arguments as follows.
>>> # Args:
>>> # value (Tensor): the value to be evaluated.
>>> # concentration (Tensor): the concentration of the distribution. Default: self._concentration.
>>> # rate (Tensor): the rate of the distribution. Default: self._rate.
>>> # Examples of `prob`.
>>> # Similar calls can be made to other probability functions
>>> # by replacing 'prob' by the name of the function
>>> ans = g1.prob(value)
>>> print(ans.shape)
(3,)
>>> # Evaluate with respect to the distribution b.
>>> ans = g1.prob(value, concentration_b, rate_b)
>>> print(ans.shape)
(3,)
>>> # `concentration` and `rate` must be passed in during function calls for g2.
>>> ans = g2.prob(value, concentration_a, rate_a)
>>> print(ans.shape)
(3,)
>>> # Functions `mean`, `sd`, `mode`, `var`, and `entropy` have the same arguments.
>>> # Args:
>>> # concentration (Tensor): the concentration of the distribution. Default: self._concentration.
>>> # rate (Tensor): the rate of the distribution. Default: self._rate.
>>> # Example of `mean`, `sd`, `mode`, `var`, and `entropy` are similar.
>>> ans = g1.mean()
>>> print(ans.shape)
(1,)
>>> ans = g1.mean(concentration_b, rate_b)
>>> print(ans.shape)
(3,)
>>> # `concentration` and `rate` must be passed in during function calls.
>>> ans = g2.mean(concentration_a, rate_a)
>>> print(ans.shape)
(3,)
>>> # Interfaces of 'kl_loss' and 'cross_entropy' are the same:
>>> # Args:
>>> # dist (str): the type of the distributions. Only "Gamma" is supported.
>>> # concentration_b (Tensor): the concentration of distribution b.
>>> # rate_b (Tensor): the rate of distribution b.
>>> # concentration_a (Tensor): the concentration of distribution a. Default: self._concentration.
>>> # rate_a (Tensor): the rate of distribution a. Default: self._rate.
>>> # Examples of `kl_loss`. `cross_entropy` is similar.
>>> ans = g1.kl_loss('Gamma', concentration_b, rate_b)
>>> print(ans.shape)
(3,)
>>> ans = g1.kl_loss('Gamma', concentration_b, rate_b, concentration_a, rate_a)
>>> print(ans.shape)
(3,)
>>> # Additional `concentration` and `rate` must be passed in.
>>> ans = g2.kl_loss('Gamma', concentration_b, rate_b, concentration_a, rate_a)
>>> print(ans.shape)
(3,)
>>> # Examples of `sample`.
>>> # Args:
>>> # shape (tuple): the shape of the sample. Default: ()
>>> # concentration (Tensor): the concentration of the distribution. Default: self._concentration.
>>> # rate (Tensor): the rate of the distribution. Default: self._rate.
>>> ans = g1.sample()
>>> print(ans.shape)
(1,)
>>> ans = g1.sample((2,3))
>>> print(ans.shape)
(2, 3, 1)
>>> ans = g1.sample((2,3), concentration_b, rate_b)
>>> print(ans.shape)
(2, 3, 3)
>>> ans = g2.sample((2,3), concentration_a, rate_a)
>>> print(ans.shape)
(2, 3, 3)
"""
def __init__(self,
concentration=None,
rate=None,
seed=None,
dtype=mstype.float32,
name="Gamma"):
"""
Constructor of Gamma.
"""
param = dict(locals())
param['param_dict'] = {'concentration': concentration, 'rate': rate}
valid_dtype = mstype.float_type
Validator.check_type_name(
"dtype", dtype, valid_dtype, type(self).__name__)
# As some operators can't accept scalar input, check the type here
if isinstance(concentration, (int, float)):
raise TypeError("Input concentration can't be scalar")
if isinstance(rate, (int, float)):
raise TypeError("Input rate can't be scalar")
super(Gamma, self).__init__(seed, dtype, name, param)
self._concentration = self._add_parameter(
concentration, 'concentration')
self._rate = self._add_parameter(rate, 'rate')
if self._concentration is not None:
check_greater_zero(self._concentration, "concentration")
if self._rate is not None:
check_greater_zero(self._rate, "rate")
# ops needed for the class
self.log = log_generic
self.square = P.Square()
self.sqrt = P.Sqrt()
self.squeeze = P.Squeeze(0)
self.cast = P.Cast()
self.dtypeop = P.DType()
self.fill = P.Fill()
self.shape = P.Shape()
self.select = P.Select()
self.greater = P.Greater()
self.lgamma = nn.LGamma()
self.digamma = nn.DiGamma()
self.igamma = nn.IGamma()
def extend_repr(self):
"""Display instance object as string."""
if self.is_scalar_batch:
s = 'concentration = {}, rate = {}'.format(
self._concentration, self._rate)
else:
s = 'batch_shape = {}'.format(self._broadcast_shape)
return s
@property
def concentration(self):
"""
Return the concentration, also know as the alpha of the Gamma distribution,
after casting to dtype.
Output:
Tensor, the concentration parameter of the distribution.
"""
return self._concentration
@property
def rate(self):
"""
Return the rate, also know as the beta of the Gamma distribution,
after casting to dtype.
Output:
Tensor, the rate parameter of the distribution.
"""
return self._rate
def _get_dist_type(self):
return "Gamma"
def _get_dist_args(self, concentration=None, rate=None):
if concentration is not None:
self.checktensor(concentration, 'concentration')
else:
concentration = self._concentration
if rate is not None:
self.checktensor(rate, 'rate')
else:
rate = self._rate
return concentration, rate
def _mean(self, concentration=None, rate=None):
"""
The mean of the distribution.
"""
concentration, rate = self._check_param_type(concentration, rate)
return concentration / rate
def _var(self, concentration=None, rate=None):
"""
The variance of the distribution.
"""
concentration, rate = self._check_param_type(concentration, rate)
return concentration / self.square(rate)
def _sd(self, concentration=None, rate=None):
"""
The standard deviation of the distribution.
"""
concentration, rate = self._check_param_type(concentration, rate)
return self.sqrt(concentration) / rate
def _mode(self, concentration=None, rate=None):
"""
The mode of the distribution.
"""
concentration, rate = self._check_param_type(concentration, rate)
mode = (concentration - 1.) / rate
nan = self.fill(self.dtypeop(concentration),
self.shape(concentration), np.nan)
comp = self.greater(concentration, 1.)
return self.select(comp, mode, nan)
def _entropy(self, concentration=None, rate=None):
r"""
Evaluate entropy.
.. math::
H(X) = \alpha - \log(\beta) + \log(\Gamma(\alpha)) + (1 - \alpha) * \digamma(\alpha)
"""
concentration, rate = self._check_param_type(concentration, rate)
return concentration - self.log(rate) + self.lgamma(concentration) \
+ (1. - concentration) * self.digamma(concentration)
def _cross_entropy(self, dist, concentration_b, rate_b, concentration_a=None, rate_a=None):
r"""
Evaluate cross entropy between Gamma distributions.
Args:
dist (str): Type of the distributions. Should be "Gamma" in this case.
concentration_b (Tensor): concentration of distribution b.
rate_b (Tensor): rate of distribution b.
concentration_a (Tensor): concentration of distribution a. Default: self._concentration.
rate_a (Tensor): rate of distribution a. Default: self._rate.
"""
check_distribution_name(dist, 'Gamma')
return self._entropy(concentration_a, rate_a) +\
self._kl_loss(dist, concentration_b, rate_b,
concentration_a, rate_a)
def _log_prob(self, value, concentration=None, rate=None):
r"""
Evaluate log probability.
Args:
value (Tensor): The value to be evaluated.
concentration (Tensor): The concentration of the distribution. Default: self._concentration.
rate (Tensor): The rate the distribution. Default: self._rate.
.. math::
L(x) = (\alpha - 1) * \log(x) - \beta * x - \log(\gamma(\alpha)) - \alpha * \log(\beta)
"""
value = self._check_value(value, 'value')
value = self.cast(value, self.dtype)
concentration, rate = self._check_param_type(concentration, rate)
unnormalized_log_prob = (concentration - 1.) * \
self.log(value) - rate * value
log_normalization = self.lgamma(
concentration) - concentration * self.log(rate)
return unnormalized_log_prob - log_normalization
def _cdf(self, value, concentration=None, rate=None):
r"""
Evaluate the cumulative distribution function on the given value. Note that igamma returns
the regularized incomplete gamma function, which is what we want for the CDF.
Args:
value (Tensor): The value to be evaluated.
concentration (Tensor): The concentration of the distribution. Default: self._concentration.
rate (Tensor): The rate the distribution. Default: self._rate.
.. math::
cdf(x) = \igamma(\alpha, \beta * x)
"""
value = self._check_value(value, 'value')
value = self.cast(value, self.dtype)
concentration, rate = self._check_param_type(concentration, rate)
return self.igamma(concentration, rate * value)
def _kl_loss(self, dist, concentration_b, rate_b, concentration_a=None, rate_a=None):
r"""
Evaluate Gamma-Gamma KL divergence, i.e. KL(a||b).
Args:
dist (str): The type of the distributions. Should be "Gamma" in this case.
concentration_b (Tensor): The concentration of distribution b.
rate_b (Tensor): The rate distribution b.
concentration_a (Tensor): The concentration of distribution a. Default: self._concentration.
rate_a (Tensor): The rate distribution a. Default: self._rate.
.. math::
KL(a||b) = (\alpha_{a} - \alpha_{b}) * \digamma(\alpha_{a}) + \log(\gamma(\alpha_{b}))
- \log(\gamma(\alpha_{a})) + \alpha_{b} * \log(\beta{a}) - \alpha_{b} * \log(\beta{b})
+ \alpha_{a} * \frac{\beta{b}}{\beta{a} - 1}
"""
check_distribution_name(dist, 'Gamma')
concentration_b = self._check_value(concentration_b, 'concentration_b')
rate_b = self._check_value(rate_b, 'rate_b')
concentration_b = self.cast(concentration_b, self.parameter_type)
rate_b = self.cast(rate_b, self.parameter_type)
concentration_a, rate_a = self._check_param_type(
concentration_a, rate_a)
return (concentration_a - concentration_b) * self.digamma(concentration_a) \
+ self.lgamma(concentration_b) - self.lgamma(concentration_a) \
+ concentration_b * self.log(rate_a) - concentration_b * self.log(rate_b) \
+ concentration_a * (rate_b / rate_a - 1.)
def _sample(self, shape=(), concentration=None, rate=None):
"""
Sampling.
Args:
shape (tuple): The shape of the sample. Default: ().
concentration (Tensor): The concentration of the samples. Default: self._concentration.
rate (Tensor): The rate of the samples. Default: self._rate.
Returns:
Tensor, with the shape being shape + batch_shape.
"""
shape = self.checktuple(shape, 'shape')
concentration, rate = self._check_param_type(concentration, rate)
batch_shape = self.shape(concentration + rate)
origin_shape = shape + batch_shape
if origin_shape == ():
sample_shape = (1,)
else:
sample_shape = origin_shape
sample_gamma = C.gamma(sample_shape, concentration, rate, self.seed)
value = self.cast(sample_gamma, self.dtype)
if origin_shape == ():
value = self.squeeze(value)
return value