Skip to main content

Module lognormal_kernel

Module lognormal_kernel 

Source
Expand description

Shared analytic kernel for latent-variable families with lognormal structure.

The kernel object K_{k,m}(μ, σ) := E[exp(k·U − m·exp(U))], where U ~ N(μ, σ²), is the only special function required by all latent families.

It satisfies exact μ-recurrences (see kernel_ratio_jet) and the corresponding heat-equation σ-identities, so fixed-σ latent families reduce to evaluating kernel bundles at shifted arguments.

Row likelihoods for binary and survival models are small signed sums of kernel terms; LogKernelSumJet evaluates their log-derivatives from log-space kernel bundles and treats non-positive signed sums as invalid rows.

Structs§

KernelSumTerm
A single signed term in a kernel sum: coefficient × K_{k,m}.
LatentCLogLogJet5
Fifth-order latent-cloglog inverse-link jet.
LatentSurvivalRow
Row-level sufficient statistics for one latent survival observation.
LatentSurvivalRowJet
Row-level log-likelihood and μ-derivatives for the latent survival model.
LogKernelSumJet
Derivatives of log(Σ_j a_j · K_{k_j, m_j}(μ, σ)) with respect to μ.
LogLognormalKernelBundle
Kernel bundle storing log K_{k,m} values instead of K_{k,m}.
ProbitFrailtyScaleJet
Probit frailty scaling factor with t-derivatives (t = log σ).

Enums§

FrailtySpec
Frailty modifier specification at the family level.
HazardLoading
How the hazard multiplier frailty loads onto the hazard components.
LatentSurvivalEventType
Event type for compiled survival sufficient statistics.
LognormalKernelError
Errors produced by the lognormal-kernel frailty/marginal-slope validators.

Functions§

kernel_ratio_jet
Computes the value-space derivative ratios ∂ⁿ_μ K_{k,m} / K_{k,m} from a log-space bundle.
latent_cloglog_inverse_link_jet
latent_cloglog_jet5
log_kernel_bundle
Builds a log-space kernel bundle for k = 0, 1, …, max_k at fixed (m, μ, σ).
log_kernel_term