# 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.
# ============================================================================
"""Bernoulli Distribution"""
from mindspore.common import dtype as mstype
from mindspore.ops import operations as P
from mindspore.ops import composite as C
from mindspore._checkparam import Validator
from .distribution import Distribution
from ._utils.utils import check_prob, check_distribution_name, clamp_probs
from ._utils.custom_ops import exp_generic, log_generic
[docs]class Bernoulli(Distribution):
"""
Bernoulli Distribution.
Args:
probs (float, list, numpy.ndarray, Tensor): The probability of that the outcome is 1.
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.int32.
name (str): The name of the distribution. Default: 'Bernoulli'.
Supported Platforms:
``Ascend`` ``GPU``
Note:
`probs` must be a proper probability (0 < p < 1).
`dist_spec_args` is `probs`.
Examples:
>>> import mindspore
>>> import mindspore.nn as nn
>>> import mindspore.nn.probability.distribution as msd
>>> from mindspore import Tensor
>>> # To initialize a Bernoulli distribution of the probability 0.5.
>>> b1 = msd.Bernoulli(0.5, dtype=mindspore.int32)
>>> # A Bernoulli distribution can be initialized without arguments.
>>> # In this case, `probs` must be passed in through arguments during function calls.
>>> b2 = msd.Bernoulli(dtype=mindspore.int32)
>>> # Here are some tensors used below for testing
>>> value = Tensor([1, 0, 1], dtype=mindspore.int32)
>>> probs_a = Tensor([0.6], dtype=mindspore.float32)
>>> probs_b = Tensor([0.2, 0.3, 0.4], dtype=mindspore.float32)
>>> # Private interfaces of probability functions corresponding to public interfaces, including
>>> # `prob`, `log_prob`, `cdf`, `log_cdf`, `survival_function`, and `log_survival`, are the same as follows.
>>> # Args:
>>> # value (Tensor): the value to be evaluated.
>>> # probs1 (Tensor): the probability of success. Default: self.probs.
>>> # Examples of `prob`.
>>> # Similar calls can be made to other probability functions
>>> # by replacing `prob` by the name of the function.
>>> ans = b1.prob(value)
>>> print(ans.shape)
(3,)
>>> # Evaluate `prob` with respect to distribution b.
>>> ans = b1.prob(value, probs_b)
>>> print(ans.shape)
(3,)
>>> # `probs` must be passed in during function calls.
>>> ans = b2.prob(value, probs_a)
>>> print(ans.shape)
(3,)
>>> # Functions `mean`, `sd`, `var`, and `entropy` have the same arguments.
>>> # Args:
>>> # probs1 (Tensor): the probability of success. Default: self.probs.
>>> # Examples of `mean`. `sd`, `var`, and `entropy` are similar.
>>> ans = b1.mean() # return 0.5
>>> print(ans.shape)
()
>>> ans = b1.mean(probs_b) # return probs_b
>>> print(ans.shape)
(3,)
>>> # `probs` must be passed in during function calls.
>>> ans = b2.mean(probs_a)
>>> print(ans.shape)
(1,)
>>> # Interfaces of `kl_loss` and `cross_entropy` are the same as follows:
>>> # Args:
>>> # dist (str): the name of the distribution. Only 'Bernoulli' is supported.
>>> # probs1_b (Tensor): the probability of success of distribution b.
>>> # probs1_a (Tensor): the probability of success of distribution a. Default: self.probs.
>>> # Examples of kl_loss. `cross_entropy` is similar.
>>> ans = b1.kl_loss('Bernoulli', probs_b)
>>> print(ans.shape)
(3,)
>>> ans = b1.kl_loss('Bernoulli', probs_b, probs_a)
>>> print(ans.shape)
(3,)
>>> # An additional `probs_a` must be passed in.
>>> ans = b2.kl_loss('Bernoulli', probs_b, probs_a)
>>> print(ans.shape)
(3,)
>>> # Examples of `sample`.
>>> # Args:
>>> # shape (tuple): the shape of the sample. Default: ().
>>> # probs1 (Tensor): the probability of success. Default: self.probs.
>>> ans = b1.sample()
>>> print(ans.shape)
()
>>> ans = b1.sample((2,3))
>>> print(ans.shape)
(2, 3)
>>> ans = b1.sample((2,3), probs_b)
>>> print(ans.shape)
(2, 3, 3)
>>> ans = b2.sample((2,3), probs_a)
>>> print(ans.shape)
(2, 3, 1)
"""
def __init__(self,
probs=None,
seed=None,
dtype=mstype.int32,
name="Bernoulli"):
"""
Constructor of Bernoulli.
"""
param = dict(locals())
param['param_dict'] = {'probs': probs}
valid_dtype = mstype.int_type + mstype.uint_type + mstype.float_type
Validator.check_type_name(
"dtype", dtype, valid_dtype, type(self).__name__)
super(Bernoulli, self).__init__(seed, dtype, name, param)
self._probs = self._add_parameter(probs, 'probs')
if self._probs is not None:
check_prob(self.probs)
# ops needed for the class
self.exp = exp_generic
self.log = log_generic
self.squeeze = P.Squeeze(0)
self.cast = P.Cast()
self.const = P.ScalarToArray()
self.floor = P.Floor()
self.fill = P.Fill()
self.less = P.Less()
self.shape = P.Shape()
self.select = P.Select()
self.uniform = C.uniform
def extend_repr(self):
if self.is_scalar_batch:
s = f'probs = {self.probs}'
else:
s = f'batch_shape = {self._broadcast_shape}'
return s
@property
def probs(self):
"""
Return the probability of that the outcome is 1
after casting to dtype.
"""
return self._probs
def _get_dist_type(self):
return "Bernoulli"
def _get_dist_args(self, probs1=None):
if probs1 is not None:
self.checktensor(probs1, 'probs')
else:
probs1 = self.probs
return (probs1,)
def _mean(self, probs1=None):
r"""
.. math::
MEAN(B) = probs1
"""
probs1 = self._check_param_type(probs1)
return probs1
def _mode(self, probs1=None):
r"""
.. math::
MODE(B) = 1 if probs1 > 0.5 else = 0
"""
probs1 = self._check_param_type(probs1)
zeros = self.fill(self.dtype, self.shape(probs1), 0.0)
ones = self.fill(self.dtype, self.shape(probs1), 1.0)
comp = self.less(0.5, probs1)
return self.select(comp, ones, zeros)
def _var(self, probs1=None):
r"""
.. math::
VAR(B) = probs1 * probs0
"""
probs1 = self._check_param_type(probs1)
probs0 = 1.0 - probs1
return self.exp(self.log(probs0) + self.log(probs1))
def _entropy(self, probs1=None):
r"""
.. math::
H(B) = -probs0 * \log(probs0) - probs1 * \log(probs1)
"""
probs1 = self._check_param_type(probs1)
probs0 = 1.0 - probs1
return -(probs0 * self.log(probs0)) - (probs1 * self.log(probs1))
def _cross_entropy(self, dist, probs1_b, probs1=None):
"""
Evaluate cross entropy between Bernoulli distributions.
Args:
dist (str): The type of the distributions. Should be "Bernoulli" in this case.
probs1_b (Tensor): `probs1` of distribution b.
probs1_a (Tensor): `probs1` of distribution a. Default: self.probs.
"""
check_distribution_name(dist, 'Bernoulli')
return self._entropy(probs1) + self._kl_loss(dist, probs1_b, probs1)
def _log_prob(self, value, probs1=None):
r"""
Log probability mass function of Bernoulli distributions.
Args:
value (Tensor): A Tensor composed of only zeros and ones.
probs (Tensor): The probability of outcome is 1. Default: self.probs.
.. math::
pmf(k) = probs1 if k = 1;
pmf(k) = probs0 if k = 0;
"""
value = self._check_value(value, 'value')
value = self.cast(value, self.parameter_type)
probs1 = self._check_param_type(probs1)
# clamp value for numerical stability
probs1 = clamp_probs(probs1)
probs0 = 1.0 - probs1
return self.log(probs1) * value + self.log(probs0) * (1.0 - value)
def _cdf(self, value, probs1=None):
r"""
Cumulative distribution function (cdf) of Bernoulli distributions.
Args:
value (Tensor): The value to be evaluated.
probs (Tensor): The probability of that the outcome is 1. Default: self.probs.
.. math::
cdf(k) = 0 if k < 0;
cdf(k) = probs0 if 0 <= k <1;
cdf(k) = 1 if k >=1;
"""
value = self._check_value(value, 'value')
value = self.cast(value, self.parameter_type)
value = self.floor(value)
probs1 = self._check_param_type(probs1)
broadcast_shape_tensor = value * probs1
value = self.broadcast(value, broadcast_shape_tensor)
probs0 = self.broadcast((1.0 - probs1), broadcast_shape_tensor)
comp_zero = self.less(value, 0.0)
comp_one = self.less(value, 1.0)
zeros = self.fill(self.parameter_type, self.shape(
broadcast_shape_tensor), 0.0)
ones = self.fill(self.parameter_type, self.shape(
broadcast_shape_tensor), 1.0)
less_than_zero = self.select(comp_zero, zeros, probs0)
return self.select(comp_one, less_than_zero, ones)
def _kl_loss(self, dist, probs1_b, probs1=None):
r"""
Evaluate bernoulli-bernoulli kl divergence, i.e. KL(a||b).
Args:
dist (str): The type of the distributions. Should be "Bernoulli" in this case.
probs1_b (Tensor, Number): `probs1` of distribution b.
probs1_a (Tensor, Number): `probs1` of distribution a. Default: self.probs.
.. math::
KL(a||b) = probs1_a * \log(\frac{probs1_a}{probs1_b}) +
probs0_a * \log(\frac{probs0_a}{probs0_b})
"""
check_distribution_name(dist, 'Bernoulli')
probs1_b = self._check_value(probs1_b, 'probs1_b')
probs1_b = self.cast(probs1_b, self.parameter_type)
probs1_a = self._check_param_type(probs1)
probs0_a = 1.0 - probs1_a
probs0_b = 1.0 - probs1_b
return probs1_a * self.log(probs1_a / probs1_b) + probs0_a * self.log(probs0_a / probs0_b)
def _sample(self, shape=(), probs1=None):
"""
Sampling.
Args:
shape (tuple): The shape of the sample. Default: ().
probs1 (Tensor): `probs1` of the samples. Default: self.probs.
Returns:
Tensor, shape is shape + batch_shape.
"""
shape = self.checktuple(shape, 'shape')
probs1 = self._check_param_type(probs1)
origin_shape = shape + self.shape(probs1)
if origin_shape == ():
sample_shape = (1,)
else:
sample_shape = origin_shape
l_zero = self.const(0.0)
h_one = self.const(1.0)
sample_uniform = self.uniform(sample_shape, l_zero, h_one, self.seed)
sample = self.less(sample_uniform, probs1)
value = self.cast(sample, self.dtype)
if origin_shape == ():
value = self.squeeze(value)
return value