Mj=#ddlZddlmZddlmZddlmZddlmZddlm Z m Z ddl m Z ddl mZd gZGd d eZy) N)Tensor) constraints) Distribution) Independent)ComposeTransform Transform)_sum_rightmost)_sizeTransformedDistributionc >eZdZUdZiZeeejfe d< dde de e e zde dzddffd Zdfd Zej d d Zede fd Zej*fdZej*fdedefdZdZdZdZdZxZS)r a Extension of the Distribution class, which applies a sequence of Transforms to a base distribution. Let f be the composition of transforms applied:: X ~ BaseDistribution Y = f(X) ~ TransformedDistribution(BaseDistribution, f) log p(Y) = log p(X) + log |det (dX/dY)| Note that the ``.event_shape`` of a :class:`TransformedDistribution` is the maximum shape of its base distribution and its transforms, since transforms can introduce correlations among events. An example for the usage of :class:`TransformedDistribution` would be:: # Building a Logistic Distribution # X ~ Uniform(0, 1) # f = a + b * logit(X) # Y ~ f(X) ~ Logistic(a, b) base_distribution = Uniform(0, 1) transforms = [SigmoidTransform().inv, AffineTransform(loc=a, scale=b)] logistic = TransformedDistribution(base_distribution, transforms) For more examples, please look at the implementations of :class:`~torch.distributions.gumbel.Gumbel`, :class:`~torch.distributions.half_cauchy.HalfCauchy`, :class:`~torch.distributions.half_normal.HalfNormal`, :class:`~torch.distributions.log_normal.LogNormal`, :class:`~torch.distributions.pareto.Pareto`, :class:`~torch.distributions.weibull.Weibull`, :class:`~torch.distributions.relaxed_bernoulli.RelaxedBernoulli` and :class:`~torch.distributions.relaxed_categorical.RelaxedOneHotCategorical` arg_constraintsNbase_distribution transforms validate_argsreturnc t|tr |g|_nCt|tr%t d|Ds t d||_nt d||j |jz}t|j}t|j}t||jjkr&t d|jjd|d|j|}|j|}||k7r"|dt||z } |j| }|jj|z } | dkDr t|| }||_|j"j|jjz } t%|j"j|| z} t|| krt'dt|d | t|| z } |d| }|| d}t(|U||| y) Nc3<K|]}t|tywN) isinstancer).0ts k/var/www/html/ia-enhanced/venv/lib/python3.12/site-packages/torch/distributions/transformed_distribution.py z3TransformedDistribution.__init__..?sDAz!Y/Dsz6transforms must be a Transform or a list of Transformsz0transforms must be a Transform or list, but was z9base_distribution needs to have shape with size at least z , but got .rzforward_shape length z must be >= event_dim r)rrrlistall ValueError batch_shape event_shapelenrdomain event_dim forward_shape inverse_shapeexpandr base_distcodomainmaxAssertionErrorsuper__init__)selfrrr base_shapebase_event_dim transformr$expanded_base_shapebase_batch_shapereinterpreted_batch_ndimstransform_change_in_event_dimr#cutrr __class__s rr,z TransformedDistribution.__init__4s6 j) ,DO D )DDD L)DOB:,O  '225F5R5RR .::;$T__5 z?Y--77 7KIL\L\LfLfKggqr|q}}~ "// ; '55mD , ,2;#)*^;  !2 8 89I J $-$4$4$>$>$O! $q ( +!#<! +    ( (9+;+;+E+E E &    ( ( : :  }  ) 'M(:';;QR[Q\] - 9,#DS) #CD)  kOc|jt|}tj|}||jz}t |j D]}|j|}|dt|t|jjz }|jj||_ |j |_tt|3||jd|j|_ |S)NFr)_get_checked_instancer torchSizer reversedrr%r!r'r&r+r,_validate_args)r-r _instancenewshaperr2r6s rr&zTransformedDistribution.expandps(()@)Ljj- d...$//* +AOOE*E + !O3u:DNN4N4N0O#OP--.>?  %s4 )) 5 "00 r7F) is_discretec,|js|jjS|jdj}t |j |j kDr6tj|t |j |j z }|S)N) rr'supportr(r!r r#r independent)r-rDs rrDzTransformedDistribution.supportsz>>)) )//"%.. t 7#4#4 4!--T--.1B1BBGr7c.|jjSr)r' has_rsample)r-s rrGz#TransformedDistribution.has_rsamples~~)))r7ctj5|jj|}|jD] }||} |cdddS#1swYyxYw)a Generates a sample_shape shaped sample or sample_shape shaped batch of samples if the distribution parameters are batched. Samples first from base distribution and applies `transform()` for every transform in the list. N)r:no_gradr'samplerr- sample_shapexr0s rrJzTransformedDistribution.samplesU]]_ %%l3A!__ ! aL !    s 6AArLcn|jj|}|jD] }||} |S)a$ Generates a sample_shape shaped reparameterized sample or sample_shape shaped batch of reparameterized samples if the distribution parameters are batched. Samples first from base distribution and applies `transform()` for every transform in the list. )r'rsamplerrKs rrOzTransformedDistribution.rsamples9 NN " "< 0 I! A r7c(|jr|j|t|j}d}|}t |j D]{}|j |}||jj|jjz z }|t|j||||jjz z }|}}|t|jj||t|jjz z}|S)z Scores the sample by inverting the transform(s) and computing the score using the score of the base distribution and the log abs det jacobian. g)r=_validate_sampler!r r<rinvr"r#r(r log_abs_det_jacobianr'log_prob)r-valuer#rTyr0rMs rrTz TransformedDistribution.log_probs     ! !% ((() #& !$//2 I a A ))33i6H6H6R6RR RI...q!4I,,666#HA n NN # #A & C8R8R4S(S  r7cd}|jD]}||jz}t|tr|dk(r|S||dz zdzS)zu This conditionally flips ``value -> 1-value`` to ensure :meth:`cdf` is monotone increasing. g?)rsignrint)r-rUrYr0s r_monotonize_cdfz'TransformedDistribution._monotonize_cdfsS  )I)..(D ) dC TQYLus{#c))r7c|jdddD]}|j|}|jr|jj ||jj |}|j |}|S)z Computes the cumulative distribution function by inverting the transform(s) and computing the score of the base distribution. NrC)rrRr=r'rQcdfr[r-rUr0s rr]zTransformedDistribution.cdfsr 2. )IMM%(E )    NN + +E 2""5)$$U+ r7c|j|}|jj|}|jD] }||} |S)z Computes the inverse cumulative distribution function using transform(s) and computing the score of the base distribution. )r[r'icdfrr^s rr`zTransformedDistribution.icdfsJ $$U+##E* %Ie$E % r7r)__name__ __module__ __qualname____doc__r dictstrr Constraint__annotations__rrrboolr,r&dependent_propertyrDpropertyrGr:r;rJr rrOrTr[r]r` __classcell__)r6s@rr r sB:rus7 +97F4 % %RlRr7