torch.distributions.transforms.Transform
class Transformnew Transform(options?: TransformOptions)
- readonly
domain(Constraint) - – Constraint representing valid inputs to this transform.
- readonly
codomain(Constraint) - – Constraint representing valid outputs of this transform.
- readonly
bijective(boolean) - – Whether this transform is bijective. A transform is bijective iff t.inv(t(x)) == x and t(t.inv(y)) == y.
- readonly
sign(number) - – For bijective univariate transforms, this is +1 or -1 depending on whether transform is monotone increasing or decreasing.
- readonly
event_dim(number) - – Number of dimensions that are correlated together in the transform. 0 for pointwise transforms, 1 for vector transforms, 2 for matrix.
- readonly
cache_size(number) - – Size of the cache (0 for no caching, 1 for single value).
- readonly
inv(Transform) - – Get the inverse transform.
Abstract base class for invertible transformations with computable log det jacobians.
Caching is useful for transforms whose inverses are either expensive or numerically unstable.