【Hackathon 5th No.18】Add Binomial and Poisson API -part

python/paddle/distribution/__init__.py

@@ -33,6 +33,8 @@
 from paddle.distribution.uniform import Uniform
 from paddle.distribution.laplace import Laplace
 from paddle.distribution.geometric import Geometric
+from paddle.distribution.binomial import Binomial
+from paddle.distribution.poisson import Poisson
 
 __all__ = [
     'Bernoulli',
@@ -55,6 +57,8 @@
     'LogNormal',
     'Gumbel',
     'Geometric',
+    'Binomial',
+    'Poisson',
 ]
 
 __all__.extend(transform.__all__)

python/paddle/distribution/binomial.py

@@ -0,0 +1,268 @@
+# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
+#
+# 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.
+
+from collections.abc import Sequence
+
+import paddle
+from paddle.distribution import distribution
+
+
+class Binomial(distribution.Distribution):
+    r"""
+    The Binomial distribution with size `total_count` and `probs` parameters.
+
+    In probability theory and statistics, the binomial distribution is the most basic discrete probability distribution defined on :math:`[0, n] \cap \mathbb{N}`,
+    which can be viewed as the number of times a potentially unfair coin is tossed to get heads, and the result
+    of its random variable can be viewed as the sum of a series of independent Bernoulli experiments.
+
+    The probability mass function (pmf) is
+
+    .. math::
+
+        pmf(x; n, p) = \frac{n!}{x!(n-x)!}p^{x}(1-p)^{n-x}
+
+    In the above equation:
+
+    * :math:`total\_count = n`: is the size, meaning the total number of Bernoulli experiments.
+    * :math:`probs = p`: is the probability of the event happening in one Bernoulli experiments.
+
+    Args:
+        total_count(int|Tensor): The size of Binomial distribution which should be greater than 0, meaning the number of independent bernoulli
+            trials with probability parameter :math:`p`. The data type will be converted to 1-D Tensor with paddle global default dtype if the input
+            :attr:`probs` is not Tensor, otherwise will be converted to the same as :attr:`probs`.
+        probs(float|Tensor): The probability of Binomial distribution which should reside in [0, 1], meaning the probability of success
+            for each individual bernoulli trial. If the input data type is float, it will be converted to a 1-D Tensor with paddle global default dtype.
+
+    Examples:
+        .. code-block:: python
+
+            >>> import paddle
+            >>> from paddle.distribution import Binomial
+            >>> paddle.set_device('cpu')
+            >>> paddle.seed(100)
+            >>> rv = Binomial(100, paddle.to_tensor([0.3, 0.6, 0.9]))
+
+            >>> print(rv.sample([2]))
+            Tensor(shape=[2, 3], dtype=float32, place=Place(cpu), stop_gradient=True,
+            [[31., 62., 93.],
+             [29., 54., 91.]])
+
+            >>> print(rv.mean)
+            Tensor(shape=[3], dtype=float32, place=Place(cpu), stop_gradient=True,
+            [30.00000191, 60.00000381, 90.        ])
+
+            >>> print(rv.entropy())
+            Tensor(shape=[3], dtype=float32, place=Place(cpu), stop_gradient=True,
+            [2.94053698, 3.00781751, 2.51124287])
+    """
+
+    def __init__(self, total_count, probs):
+        self.dtype = paddle.get_default_dtype()
+        self.total_count, self.probs = self._to_tensor(total_count, probs)
+
+        if not self._check_constraint(self.total_count, self.probs):
+            raise ValueError(
+                'Every element of input parameter `total_count` should be grater than or equal to one, and `probs` should be grater than or equal to zero and less than or equal to one.'
+            )
+        if self.total_count.shape == []:
+            batch_shape = (1,)
+        else:
+            batch_shape = self.total_count.shape
+        super().__init__(batch_shape)
+
+    def _to_tensor(self, total_count, probs):
+        """Convert the input parameters into Tensors if they were not and broadcast them
+
+        Returns:
+            Tuple[Tensor, Tensor]: converted total_count and probs.
+        """
+        # convert type
+        if isinstance(probs, float):
+            probs = paddle.to_tensor(probs, dtype=self.dtype)
+        else:
+            self.dtype = probs.dtype
+        if isinstance(total_count, int):
+            total_count = paddle.to_tensor(total_count, dtype=self.dtype)
+        else:
+            total_count = paddle.cast(total_count, dtype=self.dtype)
+
+        # broadcast tensor
+        return paddle.broadcast_tensors([total_count, probs])
+
+    def _check_constraint(self, total_count, probs):
+        """Check the constraints for input parameters
+
+        Args:
+            total_count (Tensor)
+            probs (Tensor)
+
+        Returns:
+            bool: pass or not.
+        """
+        total_count_check = (total_count >= 1).all()
+        probability_check = (probs >= 0).all() * (probs <= 1).all()
+        return total_count_check and probability_check
+
+    @property
+    def mean(self):
+        """Mean of binomial distribuion.
+
+        Returns:
+            Tensor: mean value.
+        """
+        return self.total_count * self.probs
+
+    @property
+    def variance(self):
+        """Variance of binomial distribution.
+
+        Returns:
+            Tensor: variance value.
+        """
+        return self.total_count * self.probs * (1 - self.probs)
+
+    def sample(self, shape=()):
+        """Generate binomial samples of the specified shape. The final shape would be ``shape+batch_shape`` .
+
+        Args:
+            shape (Sequence[int], optional): Prepended shape of the generated samples.
+
+        Returns:
+            Tensor: Sampled data with shape `sample_shape` + `batch_shape`. The returned data type is the same as `probs`.
+        """
+        if not isinstance(shape, Sequence):
+            raise TypeError('sample shape must be Sequence object.')
+
+        with paddle.set_grad_enabled(False):
+            shape = tuple(shape)
+            batch_shape = tuple(self.batch_shape)
+            output_shape = tuple(shape + batch_shape)
+            output_size = paddle.broadcast_to(
+                self.total_count, shape=output_shape
+            )
+            output_prob = paddle.broadcast_to(self.probs, shape=output_shape)
+            sample = paddle.binomial(
+                paddle.cast(output_size, dtype="int32"), output_prob
+            )
+            return paddle.cast(sample, self.dtype)
+
+    def entropy(self):
+        r"""Shannon entropy in nats.
+
+        The entropy is
+
+        .. math::
+
+            \mathcal{H}(X) = - \sum_{x \in \Omega} p(x) \log{p(x)}
+
+        In the above equation:
+
+        * :math:`\Omega`: is the support of the distribution.
+
+        Returns:
+            Tensor: Shannon entropy of binomial distribution. The data type is the same as `probs`.
+        """
+        values = self._enumerate_support()
+        log_prob = self.log_prob(values)
+        return -(paddle.exp(log_prob) * log_prob).sum(0)
+
+    def _enumerate_support(self):
+        """Return the support of binomial distribution [0, 1, ... ,n]
+
+        Returns:
+            Tensor: the support of binomial distribution
+        """
+        values = paddle.arange(
+            1 + paddle.max(self.total_count), dtype=self.dtype
+        )
+        values = values.reshape((-1,) + (1,) * len(self.batch_shape))
+        return values
+
+    def log_prob(self, value):
+        """Log probability density/mass function.
+
+        Args:
+          value (Tensor): The input tensor.
+
+        Returns:
+          Tensor: log probability. The data type is the same as `probs`.
+        """
+        value = paddle.cast(value, dtype=self.dtype)
+
+        # combination
+        log_comb = (
+            paddle.lgamma(self.total_count + 1.0)
+            - paddle.lgamma(self.total_count - value + 1.0)
+            - paddle.lgamma(value + 1.0)
+        )
+        eps = paddle.finfo(self.probs.dtype).eps
+        probs = paddle.clip(self.probs, min=eps, max=1 - eps)
+        # log_p
+        return paddle.nan_to_num(
+            (
+                log_comb
+                + value * paddle.log(probs)
+                + (self.total_count - value) * paddle.log(1 - probs)
+            ),
+            neginf=-eps,
+        )
+
+    def prob(self, value):
+        """Probability density/mass function.
+
+        Args:
+            value (Tensor): The input tensor.
+
+        Returns:
+            Tensor: probability. The data type is the same as `probs`.
+        """
+        return paddle.exp(self.log_prob(value))
+
+    def kl_divergence(self, other):
+        r"""The KL-divergence between two binomial distributions with the same :attr:`total_count`.
+
+        The probability density function (pdf) is
+
+        .. math::
+
+            KL\_divergence(n_1, p_1, n_2, p_2) = \sum_x p_1(x) \log{\frac{p_1(x)}{p_2(x)}}
+
+        .. math::
+
+            p_1(x) = \frac{n_1!}{x!(n_1-x)!}p_1^{x}(1-p_1)^{n_1-x}
+
+        .. math::
+
+            p_2(x) = \frac{n_2!}{x!(n_2-x)!}p_2^{x}(1-p_2)^{n_2-x}
+
+        Args:
+            other (Binomial): instance of ``Binomial``.
+
+        Returns:
+            Tensor: kl-divergence between two binomial distributions. The data type is the same as `probs`.
+
+        """
+        if not (paddle.equal(self.total_count, other.total_count)).all():
+            raise ValueError(
+                "KL divergence of two binomial distributions should share the same `total_count` and `batch_shape`."
+            )
+        support = self._enumerate_support()
+        log_prob_1 = self.log_prob(support)
+        log_prob_2 = other.log_prob(support)
+        return (
+            paddle.multiply(
+                paddle.exp(log_prob_1),
+                (paddle.subtract(log_prob_1, log_prob_2)),
+            )
+        ).sum(0)

python/paddle/distribution/kl.py

@@ -17,6 +17,7 @@
 import paddle
 from paddle.distribution.bernoulli import Bernoulli
 from paddle.distribution.beta import Beta
+from paddle.distribution.binomial import Binomial
 from paddle.distribution.categorical import Categorical
 from paddle.distribution.cauchy import Cauchy
 from paddle.distribution.continuous_bernoulli import ContinuousBernoulli
@@ -28,6 +29,7 @@
 from paddle.distribution.lognormal import LogNormal
 from paddle.distribution.multivariate_normal import MultivariateNormal
 from paddle.distribution.normal import Normal
+from paddle.distribution.poisson import Poisson
 from paddle.distribution.uniform import Uniform
 from paddle.framework import in_dynamic_mode
 
@@ -167,6 +169,11 @@ def _kl_beta_beta(p, q):
     )
 
 
+@register_kl(Binomial, Binomial)
+def _kl_binomial_binomial(p, q):
+    return p.kl_divergence(q)
+
+
 @register_kl(Dirichlet, Dirichlet)
 def _kl_dirichlet_dirichlet(p, q):
     return (
@@ -269,5 +276,10 @@ def _kl_lognormal_lognormal(p, q):
     return p._base.kl_divergence(q._base)
 
 
+@register_kl(Poisson, Poisson)
+def _kl_poisson_poisson(p, q):
+    return p.kl_divergence(q)
+
+
 def _sum_rightmost(value, n):
     return value.sum(list(range(-n, 0))) if n > 0 else value