Struct MultinomialPredictiveModel 

pub struct MultinomialPredictiveModel<'a> {
    pub training_design: ArrayView2<'a, f64>,
    pub training_class_index: &'a [u32],
    pub training_weights: ArrayView1<'a, f64>,
    pub joint_penalty: ArrayView2<'a, f64>,
    pub n_classes: usize,
}

The training data and penalty a saved multinomial model needs in order to evaluate its own log-posterior away from the mode.

A Laplace SUMMARY (β̂, H⁻¹) is not enough to compute a posterior mean: the summary is precisely the quadratic model whose inadequacy is the defect. The predictive therefore needs the likelihood itself, which means the rows.

Fields

training_design: ArrayView2<'a, f64>

Training design in the SAME (raw) basis as the saved coefficients and as the design rebuilt for prediction, shape (n, P).

training_class_index: &'a [u32]

Training class index per row, values in 0..K, aligned to class_levels.

training_weights: ArrayView1<'a, f64>

Training row weights, length n.

joint_penalty: ArrayView2<'a, f64>

The joint penalty S_λ at the selected smoothing parameters, in the stacked class-major coefficient order θ[a·P + i] = β[i, a], shape (P·M, P·M) with M = K − 1.

n_classes: usize

Total class count K (the reference class K − 1 carries η ≡ 0).

Implementations

impl<'a> MultinomialPredictiveModel<'a>

pub fn predictive_moments( &self, mode: ArrayView1<'_, f64>, x_new: ArrayView2<'_, f64>, want_second_moments: bool, ) -> Result<MultinomialPredictiveMoments, EstimationError>

Posterior-predictive moments at each row of x_new.

mode is the un-augmented posterior mode in the same stacked class-major order; it is the warm start for every augmented solve and the base of every ratio.

Trait Implementations

impl<'a> Clone for MultinomialPredictiveModel<'a>

fn clone(&self) -> MultinomialPredictiveModel<'a>

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl<'a> Copy for MultinomialPredictiveModel<'a>

impl<'a> Debug for MultinomialPredictiveModel<'a>

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.