Skip to main content

VectorLikelihood

gam_models::vector_response

Trait VectorLikelihood 

Source 
pub trait VectorLikelihood {
    // Required methods
    fn log_lik(&self, eta: ArrayView2<'_, f64>, y: ArrayView2<'_, f64>) -> f64;
    fn grad_eta(
        &self,
        eta: ArrayView2<'_, f64>,
        y: ArrayView2<'_, f64>,
    ) -> Array2<f64>;
    fn hess_diag(
        &self,
        eta: ArrayView2<'_, f64>,
        y: ArrayView2<'_, f64>,
    ) -> Array2<f64>;

    // Provided method
    fn hess_block(
        &self,
        eta: ArrayView2<'_, f64>,
        y: ArrayView2<'_, f64>,
    ) -> Array3<f64> { ... }
}
Expand description

Connector trait the inner solver (Piece 1) plugs into.

eta is the (N, M) linear predictor; y is the (N, M) target. The implementation is responsible for any link inversion. The hess_diag return is the per-element diagonal of the per-row Hessian block; for a Diagonal-noise Gaussian this is exactly (N, M) of per-output precisions.

Required Methods§

Source

fn log_lik(&self, eta: ArrayView2<'_, f64>, y: ArrayView2<'_, f64>) -> f64

log p(Y | η).

Source

fn grad_eta( &self, eta: ArrayView2<'_, f64>, y: ArrayView2<'_, f64>, ) -> Array2<f64>

∂ log p(Y | η) / ∂ η, shape (N, M).

Source

fn hess_diag( &self, eta: ArrayView2<'_, f64>, y: ArrayView2<'_, f64>, ) -> Array2<f64>

Diagonal of the per-row Hessian −∂² log p / ∂ η ∂ η, shape (N, M). This is the per-row block consumed by solver/arrow_schur.rs.

Provided Methods§

Source

fn hess_block( &self, eta: ArrayView2<'_, f64>, y: ArrayView2<'_, f64>, ) -> Array3<f64>

Per-row dense Hessian block −∂² log p / ∂η_a ∂η_b, shape (N, M, M).

Default implementation lifts Self::hess_diag onto the per-row diagonal, valid only when the per-row Hessian is genuinely diagonal across outputs (e.g. Gaussian with Isotropic/Diagonal noise). Likelihoods with off-diagonal output coupling must override this: GaussianVectorLikelihood with a low-rank precision factor F (block w·(diag(precision) + F·Fᵀ), off-diagonals w·Σ_k F[a,k]·F[b,k]) and multinomial-logit (per-row Fisher block p_a (δ_ab − p_b)).

The returned array is consumed by gam_solve::pirls::dense_block_xtwx / gam_solve::pirls::dense_block_xtwy to build XᵀWX and XᵀWy for vector-response IRLS in output-major coefficient ordering.

Dyn Compatibility§

This trait is dyn compatible.

In older versions of Rust, dyn compatibility was called "object safety".

Implementors§