Struct OffsetChannelResiduals 

pub struct OffsetChannelResiduals {
    pub exit: Array1<f64>,
    pub entry: Array1<f64>,
    pub derivative: Array1<f64>,
    pub right: Array1<f64>,
}

Per-observation gradients of the unpenalized survival NLL with respect to each additive offset channel, at a given β. See WorkingModelSurvival::offset_channel_residuals for the algebra.

Contract: all four arrays have length n = number of observations. Rows with non-positive sampleweight are 0 in every channel. The derivative channel is 0 in all non-event rows. The right channel is the interval upper-bound (R) η-offset sensitivity and is exactly 0 for every NON-interval-censored model and every non-interval row of the latent interval model (only the dedicated SurvInterval(L, R, event) latent fit populates it); the baseline-θ chain rule contracts it against the age_right-evaluated η-partial.

Fields

exit: Array1<f64>

∂NLL/∂o_X: w·(exp(η_exit) − δ) per row.

entry: Array1<f64>

∂NLL/∂o_E: −w·exp(η_entry) if row has a positive entry interval else 0.

derivative: Array1<f64>

∂NLL/∂o_D: −w·δ / s (event-row only).

right: Array1<f64>

∂NLL/∂o_R: interval upper-bound (R) η-offset sensitivity, −w·∂(log-lik)/∂q_right. Nonzero only for interval-censored latent rows; exactly 0 for every other channel/model.

Trait Implementations

impl Clone for OffsetChannelResiduals

fn clone(&self) -> OffsetChannelResiduals

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Debug for OffsetChannelResiduals

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

Formats the value using the given formatter.