Struct BernoulliMarginalSlopePredictor 

pub struct BernoulliMarginalSlopePredictor {

    pub beta_marginal: Array1<f64>,
    pub beta_logslope: Array1<f64>,
    pub beta_score_warp: Option<Array1<f64>>,
    pub beta_link_dev: Option<Array1<f64>>,
    pub base_link: InverseLink,
    pub z_column: String,
    pub latent_z_normalization: SavedLatentZNormalization,
    pub latent_measure: LatentMeasureKind,
    pub baseline_marginal: f64,
    pub baseline_logslope: f64,
    pub covariance: Option<Array2<f64>>,
    pub score_warp_runtime: Option<SavedCompiledFlexBlock>,
    pub link_deviation_runtime: Option<SavedCompiledFlexBlock>,
    pub gaussian_frailty_sd: Option<f64>,
    pub latent_z_calibration: Option<LatentZRankIntCalibration>,
    pub latent_z_conditional_calibration: Option<LatentZConditionalCalibration>,
}

Fields

beta_marginal: Array1<f64>

beta_logslope: Array1<f64>

beta_score_warp: Option<Array1<f64>>

beta_link_dev: Option<Array1<f64>>

base_link: InverseLink

z_column: String

latent_z_normalization: SavedLatentZNormalization

latent_measure: LatentMeasureKind

baseline_marginal: f64

baseline_logslope: f64

covariance: Option<Array2<f64>>

score_warp_runtime: Option<SavedCompiledFlexBlock>

link_deviation_runtime: Option<SavedCompiledFlexBlock>

gaussian_frailty_sd: Option<f64>

latent_z_calibration: Option<LatentZRankIntCalibration>

latent_z_conditional_calibration: Option<LatentZConditionalCalibration>

Implementations

impl BernoulliMarginalSlopePredictor

pub fn likelihood_family(&self) -> LikelihoodSpec

pub fn mean_from_eta( &self, eta: &Array1<f64>, ) -> Result<Array1<f64>, EstimationError>

pub fn mean_derivative_from_eta( &self, eta: &Array1<f64>, ) -> Result<Array1<f64>, EstimationError>

pub fn from_unified( unified: &UnifiedFitResult, z_column: String, latent_z_normalization: SavedLatentZNormalization, latent_measure: LatentMeasureKind, baseline_marginal: f64, baseline_logslope: f64, base_link: InverseLink, frailty: FrailtySpec, score_warp_runtime: Option<SavedCompiledFlexBlock>, link_deviation_runtime: Option<SavedCompiledFlexBlock>, latent_z_calibration: Option<LatentZRankIntCalibration>, latent_z_conditional_calibration: Option<LatentZConditionalCalibration>, ) -> Result<Self, String>

pub fn theta(&self) -> Array1<f64>

pub fn final_eta_and_gradient_from_theta( &self, input: &PredictInput, theta: &Array1<f64>, need_gradient: bool, ) -> Result<(Array1<f64>, Option<Array2<f64>>), EstimationError>

pub fn final_eta_from_theta( &self, input: &PredictInput, theta: &Array1<f64>, ) -> Result<Array1<f64>, EstimationError>

Per-row final (base-scale) linear predictor for an arbitrary coefficient vector theta in the saved [marginal | logslope | score_warp? | link_dev?] block order. The marginal-slope rigid kernel is applied exactly per row, so the returned η is the same object the point predictor consumes — only parameterised by an external draw instead of self.theta(). Used by the posterior predictive path (#1049) to map each Laplace draw to its η surface before the shared eta→bands collapse; the response scale is the probit inverse link μ = Φ(η).

pub fn theta_len(&self) -> usize

Length of the concatenated coefficient vector this predictor consumes (marginal + logslope + score_warp? + link_dev?). The posterior predictive path validates each saved draw against this before mapping it through Self::final_eta_from_theta.

pub fn predict_eta_and_q_chain( &self, input: &PredictInput, ) -> Result<(Array1<f64>, Array1<f64>), EstimationError>

Per-row (eta, ∂eta/∂q_marginal) under the exact IFT pull-back.

Returns the same eta as predict_plugin_response/predict_linear_predictor plus the analytic derivative of the internal probit index with respect to the per-row marginal q (the linear predictor before the de-nested calibration). Survival prediction multiplies the second component by the per-row dq/dt to obtain the exact hazard time derivative under score-warp / link-deviation flex blocks.

Rigid path (no flex blocks): ∂eta/∂q = c = sqrt(1 + (s b)^2), recovering the rigid-path probit-frailty composition. Flex path: ∂eta/∂q = scale · link_c_obs · a_q where link_c_obs = 1 + Δ_w'(eta_base) is the link-deviation slope at the observed eta_base = a + b z and a_q = φ(q) / |F_a| is the implicit-function derivative of the calibration intercept (mirrors the bernoulli final_eta_and_gradient_from_theta flex branch lines 1399-1593).