o * i,#@sjddlmZddlmZddlmZddlZddlmZer*ddlm Z ddl m Z Gdd d ej Z dS) ) annotations)Sequence) TYPE_CHECKINGN) distribution)Tensor) _DTypeLiteralcseZdZUdZded<ded<ded<d$fd d Zd%ddZed&ddZed&ddZ gfd'ddZ d&ddZ d&ddZ d(ddZ d(dd Zd)d"d#ZZS)*Binomiala 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]) rdtyper total_countprobs int | Tensorfloat | TensorreturnNonecs6t|_|||\|_|_|jj}t|dS)N) paddleZget_default_dtyper _to_tensorr r shapesuper__init__)selfr r batch_shape __class__h/home/app/PaddleOCR-VL-test/.venv_paddleocr/lib/python3.10/site-packages/paddle/distribution/binomial.pyrQs zBinomial.__init__ list[Tensor]cCs^t|trtj||jd}n|j|_t|tr tj||jd}ntj||jd}t||gS)zConvert the input parameters into Tensors if they were not and broadcast them Returns: list[Tensor]: converted total_count and probs. r ) isinstancefloatrZ to_tensorr intcastZbroadcast_tensors)rr r rrrrZs  zBinomial._to_tensorcCs |j|jS)zYMean of binomial distribution. Returns: Tensor: mean value. r r rrrrmeanos z Binomial.meancCs|j|jd|jS)zaVariance of binomial distribution. Returns: Tensor: variance value. r!r"rrrvariancexszBinomial.variancer Sequence[int]cCst|ts tdtd9t|}t|j}t||}tj|j|d}tj|j |d}t tj |dd|}t ||j WdS1sJwYdS)a^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`. z%sample shape must be Sequence object.F)rZint32rN) rr TypeErrorrZset_grad_enabledtuplerZ broadcast_tor r Zbinomialr r )rrrZ output_shapeZ output_sizeZ output_probsamplerrrr)s     $zBinomial.samplecCs(|}||}t||d S)aeShannon entropy in nats. The entropy is .. math:: \mathcal{H}(X) = - \sum_{x \in \Omega} p(x) \log{p(x)} In the above equation: * :math:`\Omega`: is the support of the distribution. Returns: Tensor: Shannon entropy of binomial distribution. The data type is the same as `probs`. r)_enumerate_supportlog_probrexpsum)rvaluesr+rrrentropys zBinomial.entropycCs8tjdt|j|jd}|ddt|j}|S)zReturn the support of binomial distribution [0, 1, ... ,n] Returns: Tensor: the support of binomial distribution r$r))r$)rZarangemaxr r Zreshapelenr)rr.rrrr*s zBinomial._enumerate_supportvaluecCstj||jd}t|jdt|j|dt|d}t|jjj}tj|j|d|d}tj ||t ||j|t d|| dS)zLog probability density/mass function. Args: value (Tensor): The input tensor. Returns: Tensor: log probability. The data type is the same as `probs`. rg?r$)minr1)Zneginf) rr r lgammar Zfinfor epsZclipZ nan_to_numlog)rr3Zlog_combr6r rrrr+s"   zBinomial.log_probcCst||S)zProbability density/mass function. Args: value (Tensor): The input tensor. Returns: Tensor: probability. The data type is the same as `probs`. )rr,r+)rr3rrrprobs z Binomial.probothercCs<|}||}||}tt|t||dS)avThe 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`. r)r*r+rmultiplyr,subtractr-)rr9ZsupportZ log_prob_1Z log_prob_2rrr kl_divergences   zBinomial.kl_divergence)r r r r rr)r r r r rr)rr)rr&rr)r3rrr)r9rrr)__name__ __module__ __qualname____doc____annotations__rrpropertyr#r%r)r/r*r+r8r< __classcell__rrrrrs" 0       r) __future__rcollections.abcrtypingrrZpaddle.distributionrrZpaddle._typing.dtype_liker Distributionrrrrrs