Profile the inner Laplace solve by phase — tulpa

Times the sparse joint Laplace solver one phase at a time – scatter (the Hessian and gradient assembly), factorize (numeric Cholesky), eta, line search, and the rest – and returns the breakdown as a data frame. The accumulator aggregates across the parallel outer-grid worker threads, so the reported times cover the whole fit rather than only the calling thread.

Usage

tulpa_profile(expr, sort = TRUE)

Arguments

expr: An expression that runs a fit (for example a call to tulpa_nested_laplace_joint()). Evaluated once, after the profile counters are reset.
sort: Logical; order rows by descending time. Default TRUE.

Value

A data frame with one row per phase and columns phase, seconds, calls, ms_per_call (mean wall time per phase call), and share (fraction of total timed seconds). The fit result is attached as the "value" attribute.

Details

Use it to settle where a per-cell solve spends its time, e.g. whether a slow joint occu_cover() fit is bound by the assembly scatter or the Cholesky factorize:


  p <- tulpa_profile(
    tulpa_nested_laplace_joint(..., control = list(integration = "ccd"))
  )
  print(p)            # rows ordered by time; scatter vs factorize at top
  fit <- attr(p, "value")

Examples

# \donttest{
set.seed(1)
n <- 200L; X <- cbind(1, rnorm(n))
y <- rbinom(n, 1, plogis(X %*% c(0, 0.5)))
tulpa_profile(tulpa_laplace(y, rep(1L, n), X, family = "binomial"))
# }