o Me2@slddlZddlZddlmZmZzddlmZWn ddlmZYGdddejZddZ dd Z dS) N) categorical distribution)IterablecsbeZdZdZfddZeddZeddZdd Zd d Z dd dZ ddZ ddZ Z S) Multinomiala, Multinomial distribution parameterized by :attr:`total_count` and :attr:`probs`. In probability theory, the multinomial distribution is a generalization of the binomial distribution, it models the probability of counts for each side of a k-sided die rolled n times. When k is 2 and n is 1, the multinomial is the bernoulli distribution, when k is 2 and n is grater than 1, it is the binomial distribution, when k is grater than 2 and n is 1, it is the categorical distribution. The probability mass function (PMF) for multinomial is .. math:: f(x_1, ..., x_k; n, p_1,...,p_k) = \frac{n!}{x_1!...x_k!}p_1^{x_1}...p_k^{x_k} where, :math:`n` is number of trials, k is the number of categories, :math:`p_i` denote probability of a trial falling into each category, :math:`{\textstyle \sum_{i=1}^{k}p_i=1}, p_i \ge 0`, and :math:`x_i` denote count of each category. Args: total_count (int): Number of trials. probs (Tensor): Probability of a trial falling into each category. Last axis of probs indexes over categories, other axes index over batches. Probs value should between [0, 1], and sum to 1 along last axis. If the value over 1, it will be normalized to sum to 1 along the last axis. Examples: .. code-block:: python import paddle multinomial = paddle.distribution.Multinomial(10, paddle.to_tensor([0.2, 0.3, 0.5])) print(multinomial.sample((2, 3))) # Tensor(shape=[2, 3, 3], dtype=float32, place=Place(gpu:0), stop_gradient=True, # [[[1., 4., 5.], # [0., 2., 8.], # [2., 4., 4.]], # [[1., 6., 3.], # [3., 3., 4.], # [3., 4., 3.]]]) cst|tr |dkr td|dkrtd||jddd|_||_tj| |d|_ t t | |jdd|jdddS)NzBinput parameter total_count must be int type and grater than zero.z9probs parameter shoule not be none and over one dimensionT)Zkeepdim)logits) isinstanceint ValueErrordimsumprobs total_countrZ Categorical_probs_to_logits _categoricalsuperr__init__shape)selfrr __class__OD:\Projects\ConvertPro\env\Lib\site-packages\paddle/distribution/multinomial.pyrJs *zMultinomial.__init__cCs |j|jS)z[mean of multinomial distribuion. Returns: Tensor: mean value. )rrrrrrmean[s zMultinomial.meancCs|j|jd|jS)zdvariance of multinomial distribution. Returns: Tensor: variance value. r)rrrrrrvariancedszMultinomial.variancecCst||S)zprobability mass function evaluated at value. Args: value (Tensor): value to be evaluated. Returns: Tensor: probability of value. )paddleexplog_prob)rvaluerrrprobms zMultinomial.probcCst|r t||jj}tt|j|g\}}d||dkt|@<t| ddt|d d|| dS)zprobability mass function evaluated at value Args: value (Tensor): value to be evaluated. Returns: Tensor: probability of value. rrr) r is_integercastrdtypeZbroadcast_tensorslogisinflgammar )rr rrrrrxs  zMultinomial.log_probrcCsRt|ts td|j|jgt|}tjj ||j j d |j jdS)zdraw sample data from multinomial distribution Args: sample_shape (tuple, optional): [description]. Defaults to (). z%sample shape must be Iterable object.rr)r r TypeErrorrsamplerlistrnnZ functionalZone_hotrrr#r$r )rrZsamplesrrrr)s zMultinomial.samplecCstjdg|j|jjd}tj|jd|jjdddt|jjdd}t | ||}||j t |d|t |dddgS) z`entropy of multinomial distribution Returns: Tensor: entropy value r)rZ fill_valuer$r$)r)rNrr)rfullrrr$arangeZreshapelenrr_binomial_logpmfrentropyr'r )rnZsupportZ binomial_pmfrrrr1s"zMultinomial.entropyc Cs~|j|jdd}t|d}t|d}t||d}|t||ttt| |}|||||S)NT)Z is_binaryr)rrrr' _clip_by_zerolog1prabs)rcountr rZfactor_nZfactor_kZ factor_nmkZnormrrrr0s zMultinomial._binomial_logpmf)r)__name__ __module__ __qualname____doc__rpropertyrrr!rr)r1r0 __classcell__rrrrrs 0   rcCstj|d|dS)Nrr,)rr.)r6r$rrr_binomial_supportsr=cCs |jdd||jdddS)Nr)min)max)Zclip)xrrrr3s r3) collectionsrZpaddle.distributionrrcollections.abcr Distributionrr=r3rrrrs#