torch.distributions.transforms.SoftplusTransform
class SoftplusTransform extends Transformnew SoftplusTransform(options?: TransformOptions)
- readonly
domain(unknown) - readonly
codomain(unknown) - readonly
bijective(unknown) - readonly
sign(unknown)
Softplus transform: y = ln(1 + exp(x)) (smooth ReLU, maps ℝ → (0, ∞)).
Transform via the mapping Softplus(x) = log(1 + exp(x)). This is a smooth approximation to ReLU that maps real numbers to positive numbers. It's commonly used to transform unbounded parameters to positive values (e.g., standard deviations).