Replay an Observed Path Through a Numeric Exponential Kernel

Decay, integrate and update components along an observed path, excluding the baseline, likelihood and random draws. This is an experimental R reference interface, independent of the likelihood engine option.

Usage

exp_kernel_replay(kernel, inter_arrival, type = NULL, mark = NULL,
                  lambda_component0, history = FALSE)

Arguments

kernel

An exp_kernel_spec or a convertible numeric hspec.

inter_arrival

Finite nonnegative numeric vector beginning with zero.

type

Event types, one per row. Required for multivariate kernels. Entries after the initial row must be integer event types; the initial entry is ignored.

mark

Finite numeric marks, one per row; defaults to zero initially and one at each event. The initial mark is ignored.

lambda_component0

Explicit finite initial intensity components excluding the baseline. A scalar expands to every component. A matrix must match the kernel shape; a vector of matching length uses R's column-major matrix order. Automatic initialization policies belong to the calling model.

history

Logical; retain full component histories?

Details

The initial row causes no update. Zero-length event intervals apply updates in input order. Negative components are not clipped. No terminal observation interval is added. Count paths and initial counts are the calling model's responsibility. Full component histories are allocated only when requested.

Value

A list. For M rows including the initial row, D event types and K kernel columns, kernel_integral and kernel_before are M by D matrices of interval integrals and pre-event row sums. Initial integrals are zero; initial sums are the initial component row sums. final_component is the D by K state after the last event, or the initial state for a one-row path. With history = TRUE, component_before and component_after are M by (D times K), with row-major kernel ordering within each path row.

See also

exp_kernel_spec

Examples

kernel <- exp_kernel_spec(.2, 1)
path <- exp_kernel_replay(kernel, c(0, .25), lambda_component0 = .1,
                          history = TRUE)
path$kernel_integral
#>            [,1]
#> [1,] 0.00000000
#> [2,] 0.02211992
path$final_component
#>           [,1]
#> [1,] 0.2778801