torch.distributions.transforms.PowerTransform

class PowerTransform extends Transform

new PowerTransform(exponent: number, options?: PowerTransformOptions)

Power transform: y = x^exponent (maps (0, ∞) → (0, ∞)).

Transform via the mapping y = x^exponent. This transform is typically used with positive inputs.

\begin{aligned} \text{Forward: } y = x^\alpha \\ \text{Inverse: } x = y^{1/\alpha} \\ \text{Log-det Jacobian: } \log\left|\frac{dy}{dx}\right| = \log|\alpha| + (\alpha-1)\ln(x) \\ \text{Derivative: } \frac{dy}{dx} = \alpha x^{\alpha-1} \\ \text{Domain: } (0, \infty), \text{Codomain: } (0, \infty) \end{aligned}

torch.distributions.transforms.PowerTransform

readonlydomain(unknown)

readonlycodomain(unknown)

readonlybijective(unknown)

readonlyexponent(number)

– The exponent for the power transform.

readonlysign(number)

– Sign of the Jacobian determinant.

Power transform: y = x^exponent (maps (0, ∞) → (0, ∞)).

Transform via the mapping y = x^exponent. This transform is typically used with positive inputs.

\begin{aligned} \text{Forward: } y = x^\alpha \\ \text{Inverse: } x = y^{1/\alpha} \\ \text{Log-det Jacobian: } \log\left|\frac{dy}{dx}\right| = \log|\alpha| + (\alpha-1)\ln(x) \\ \text{Derivative: } \frac{dy}{dx} = \alpha x^{\alpha-1} \\ \text{Domain: } (0, \infty), \text{Codomain: } (0, \infty) \end{aligned}