iscan.score_estimator.stein_hess¶

iscan.score_estimator.stein_hess(X: torch.Tensor, eta_G: float, eta_H: float, s: float | None = None) → torch.Tensor¶

Estimates the diagonal of the Hessian of \(\log p(x)\) at the provided samples points \(X\), using first and second-order Stein identities.

Parameters:¶

X : torch.Tensor¶

dataset X

eta_G : float¶

Coefficient of the L2 regularizer for estimation of the score.

eta_H : float¶

Coefficient of the L2 regularizer for estimation of the score’s Jacobian diagonal.

s : float, optional¶

Scale for the Kernel. If None, the scale is estimated from data, by default None.

Returns:¶

Estimation of the score’s Jacobian diagonal.

Return type:¶

torch.Tensor