EiB(SrSSKJrJr SSKJr /SQr"SS5r\"5r\"5r \R\R5\ R\R5S55r \R\R5S5r\ R\R5S 5r\R\R 5\R\R"5\ R\R 5\ R\R"5S 5555r\R\R&5\R\R(5\ R\R&5\ R\R(5S 5555r\R\R,5\ R\R,5S 55r\R\R05\R\R25\ R\R05\ R\R25S 5555r\R\R65S5r\ R\R65S5r\ R\R<5S5r\ R\R@5\ R\RB5S55r"\R\RF5\ R\RF5S55r$\R\RJ5S5r&\ R\RJ5S5r'\R\RP5S5r)\ R\RP5S5r*g)aF PyTorch provides two global :class:`ConstraintRegistry` objects that link :class:`~torch.distributions.constraints.Constraint` objects to :class:`~torch.distributions.transforms.Transform` objects. These objects both input constraints and return transforms, but they have different guarantees on bijectivity. 1. ``biject_to(constraint)`` looks up a bijective :class:`~torch.distributions.transforms.Transform` from ``constraints.real`` to the given ``constraint``. The returned transform is guaranteed to have ``.bijective = True`` and should implement ``.log_abs_det_jacobian()``. 2. ``transform_to(constraint)`` looks up a not-necessarily bijective :class:`~torch.distributions.transforms.Transform` from ``constraints.real`` to the given ``constraint``. The returned transform is not guaranteed to implement ``.log_abs_det_jacobian()``. The ``transform_to()`` registry is useful for performing unconstrained optimization on constrained parameters of probability distributions, which are indicated by each distribution's ``.arg_constraints`` dict. These transforms often overparameterize a space in order to avoid rotation; they are thus more suitable for coordinate-wise optimization algorithms like Adam:: loc = torch.zeros(100, requires_grad=True) unconstrained = torch.zeros(100, requires_grad=True) scale = transform_to(Normal.arg_constraints["scale"])(unconstrained) loss = -Normal(loc, scale).log_prob(data).sum() The ``biject_to()`` registry is useful for Hamiltonian Monte Carlo, where samples from a probability distribution with constrained ``.support`` are propagated in an unconstrained space, and algorithms are typically rotation invariant.:: dist = Exponential(rate) unconstrained = torch.zeros(100, requires_grad=True) sample = biject_to(dist.support)(unconstrained) potential_energy = -dist.log_prob(sample).sum() .. note:: An example where ``transform_to`` and ``biject_to`` differ is ``constraints.simplex``: ``transform_to(constraints.simplex)`` returns a :class:`~torch.distributions.transforms.SoftmaxTransform` that simply exponentiates and normalizes its inputs; this is a cheap and mostly coordinate-wise operation appropriate for algorithms like SVI. In contrast, ``biject_to(constraints.simplex)`` returns a :class:`~torch.distributions.transforms.StickBreakingTransform` that bijects its input down to a one-fewer-dimensional space; this a more expensive less numerically stable transform but is needed for algorithms like HMC. The ``biject_to`` and ``transform_to`` objects can be extended by user-defined constraints and transforms using their ``.register()`` method either as a function on singleton constraints:: transform_to.register(my_constraint, my_transform) or as a decorator on parameterized constraints:: @transform_to.register(MyConstraintClass) def my_factory(constraint): assert isinstance(constraint, MyConstraintClass) return MyTransform(constraint.param1, constraint.param2) You can create your own registry by creating a new :class:`ConstraintRegistry` object. ) constraints transforms)_Number)ConstraintRegistry biject_to transform_toc<^\rSrSrSrU4SjrSSjrSrSrU=r $)rPz- Registry to link constraints to transforms. c0>0Ul[TU] 5 gN) _registrysuper__init__)self __class__s d/home/mande/repo/quber/.venv/lib/python3.13/site-packages/torch/distributions/constraint_registry.pyrConstraintRegistry.__init__Us c^^UcUU4Sj$[T[R5(a [T5m[T[5(a[ T[R5(d[ ST35eUTR T'U$)a Registers a :class:`~torch.distributions.constraints.Constraint` subclass in this registry. Usage:: @my_registry.register(MyConstraintClass) def construct_transform(constraint): assert isinstance(constraint, MyConstraint) return MyTransform(constraint.arg_constraints) Args: constraint (subclass of :class:`~torch.distributions.constraints.Constraint`): A subclass of :class:`~torch.distributions.constraints.Constraint`, or a singleton object of the desired class. factory (Callable): A callable that inputs a constraint object and returns a :class:`~torch.distributions.transforms.Transform` object. c(>TRTU5$r )register)factory constraintrs r-ConstraintRegistry.register..ls4==W#ErzLExpected constraint to be either a Constraint subclass or instance, but got ) isinstancer Constrainttype issubclass TypeErrorr rrrs`` rrConstraintRegistry.registerYs$ ?E E j+"8"8 9 9j)J*d++:  ..4 4 ^_i^jk &-z"rcUR[U5nU"U5$![a$ [S[U5RS35Sef=f)a Looks up a transform to constrained space, given a constraint object. Usage:: constraint = Normal.arg_constraints["scale"] scale = transform_to(constraint)(torch.zeros(1)) # constrained u = transform_to(constraint).inv(scale) # unconstrained Args: constraint (:class:`~torch.distributions.constraints.Constraint`): A constraint object. Returns: A :class:`~torch.distributions.transforms.Transform` object. Raises: `NotImplementedError` if no transform has been registered. zCannot transform z constraintsN)r rKeyErrorNotImplementedError__name__r!s r__call__ConstraintRegistry.__call__|s`( nnT*%56G z""  %#D$4$=$=#>lK  s ".A)r r ) r& __module__ __qualname____firstlineno____doc__rrr'__static_attributes__ __classcell__)rs@rrrPs!F##rrc"[R$r )ridentity_transformrs r_transform_to_realr2s  ( ((rcl[UR5n[R"XR5$r )rbase_constraintrIndependentTransformreinterpreted_batch_ndimsrbase_transforms r_biject_to_independentr9s.z99:N  * *<< rcl[UR5n[R"XR5$r )rr4rr5r6r7s r_transform_to_independentr;s.!*"<"<=N  * *<< rc,[R"5$r )r ExpTransformr1s r_transform_to_positiver>s  " " $$rc[R"[R"5[R"URS5/5$)N)rComposeTransformr=AffineTransform lower_boundr1s r_transform_to_greater_thanrDs>  & &  # # %  & &z'='=q A  rc[R"[R"5[R"URS5/5$)N)rrAr=rB upper_boundr1s r_transform_to_less_thanrHs>  & &  # # %  & &z'='=r B  rc[UR[5=(a URS:Hn[UR[5=(a URS:HnU(aU(a[R "5$URnURUR- n[R "[R "5[R"X45/5$)Nrr@)rrCrrGrSigmoidTransformrArB)r lower_is_0 upper_is_1locscales r_transform_to_intervalrOs :))73S 8N8NRS8S :))73S 8N8NRS8Sj**,,  C  " "Z%;%; ;E  & &  $ $ & (B(B3(NO rc,[R"5$r )rStickBreakingTransformr1s r_biject_to_simplexrR  , , ..rc,[R"5$r )rSoftmaxTransformr1s r_transform_to_simplexrVs  & & ((rc,[R"5$r )rLowerCholeskyTransformr1s r_transform_to_lower_choleskyrYrSrc,[R"5$r )rPositiveDefiniteTransformr1s r_transform_to_positive_definiter\s  / / 11rc,[R"5$r )rCorrCholeskyTransformr1s r_transform_to_corr_choleskyr_s  + + --rc[R"URVs/sHn[U5PM snURUR 5$s snfr )r CatTransformcseqrdimlengthsrcs r_biject_to_catrg s@  " ")/!1/ASAS /Ac[R"URVs/sHn[U5PM snURUR 5$s snfr )rrarbrrcrdres r_transform_to_catrjs@  " "",//2/Qa/2JNNJDVDV 2rhc[R"URVs/sHn[U5PM snUR5$s snfr )rStackTransformrbrrcres r_biject_to_stackrms8  $ $)/!1/ /Ac[R"URVs/sHn[U5PM snUR5$s snfr )rrlrbrrcres r_transform_to_stackrps8  $ $",//2/Qa/2JNN 2rnN)+r,torch.distributionsrr torch.typesr__all__rrrrrealr2 independentr9r;positive nonnegativer> greater_thangreater_than_eqrD less_thanrHintervalhalf_open_intervalrOsimplexrRrVlower_choleskyrYpositive_definitepositive_semidefiniter\ corr_choleskyr_catrgrjstackrmrprrrsAF8 F#F#R  !#  K$$% {''())&) K++,-{../0 K(() K++, {++, {../%0--*% K,,- K//0 {//0 {223411. K))* {,,-.+ K(() K223 {++, {5567-4*$ K''(/)/{**+),) {112/3/{445 {8892:62 K--. {001.2/. KOO$% {'(  K%%&' {(()*r