o h @s:ddlZddlmZddlmZdgZGdddeZdS)N)Tensor) DistributionExponentialFamilyc@sLeZdZdZedeedffddZddZede fdd Z d d Z d S) ra ExponentialFamily is the abstract base class for probability distributions belonging to an exponential family, whose probability mass/density function has the form is defined below .. math:: p_{F}(x; \theta) = \exp(\langle t(x), \theta\rangle - F(\theta) + k(x)) where :math:`\theta` denotes the natural parameters, :math:`t(x)` denotes the sufficient statistic, :math:`F(\theta)` is the log normalizer function for a given family and :math:`k(x)` is the carrier measure. Note: This class is an intermediary between the `Distribution` class and distributions which belong to an exponential family mainly to check the correctness of the `.entropy()` and analytic KL divergence methods. We use this class to compute the entropy and KL divergence using the AD framework and Bregman divergences (courtesy of: Frank Nielsen and Richard Nock, Entropies and Cross-entropies of Exponential Families). return.cCt)zv Abstract method for natural parameters. Returns a tuple of Tensors based on the distribution NotImplementedErrorselfr r/var/www/html/construction_image-detection-poc/venv/lib/python3.10/site-packages/torch/distributions/exp_family.py_natural_paramsz!ExponentialFamily._natural_paramscGr)z Abstract method for log normalizer function. Returns a log normalizer based on the distribution and input r)r natural_paramsr r r _log_normalizer'sz!ExponentialFamily._log_normalizercCr)zp Abstract method for expected carrier measure, which is required for computing entropy. rr r r r _mean_carrier_measure.rz'ExponentialFamily._mean_carrier_measurecCsv|j }dd|jD}|j|}tjj||dd}||7}t||D]\}}||||j dd8}q%|S)z_ Method to compute the entropy using Bregman divergence of the log normalizer. cSsg|]}|qSr )detachrequires_grad_).0pr r r ;sz-ExponentialFamily.entropy..T) create_graph)r) rr rtorchautogradgradsumzipreshape _batch_shape)r resultnparams lg_normal gradientsnpgr r r entropy6s  zExponentialFamily.entropyN) __name__ __module__ __qualname____doc__propertytuplerr rfloatrr&r r r r r s )rr torch.distributions.distributionr__all__rr r r r s