Structural comparison of two symbolized models — compare

Compares two symbolized_model objects and returns a structured diff covering:

meta — class, family, response, and n_obs for each side.
submodels — which submodels appear only on the left, only on the right, or on both sides.
terms — within each shared submodel, which term labels appear only on one side or on both.
assumptions — for each assumption, the statuses on each side and whether they match.
metrics (optional, when metrics = TRUE) — AIC, BIC, log- likelihood, and residual degrees of freedom on each side plus their delta. The metrics block refuses to compute deltas when the two fits are obviously incomparable (different family, different response, different n_obs) and instead carries a note column explaining why.

Use it to ask questions like "what's different between this fit and the previous one?" — a structural answer rather than a numeric one. Coefficient values are not compared; this is the structural-symbolic diff. For coefficient-level differences, use the parameter_interpretation() tibbles directly. For fit-level identifiability and convergence diagnostics, run drmTMB::check_drm() (or the analogous diagnostic for gllvmTMB) on each fit before interpreting the structural diff.

Usage

compare_symbolic(sym_a, sym_b, metrics = FALSE, ...)

Arguments

sym_a, sym_b: Two symbolized_model objects.
metrics: Logical, default FALSE. When TRUE, joins the AIC / BIC / log-likelihood / df from each side as a fifth slot diff_metrics and adds a delta column. Requires that each side carries its fit object on metadata$fit (this is the default for symbolize.drmTMB() and symbolize.gllvmTMB()).
...: Reserved for future use.

Value

A list classed c("symbolic_comparison", "list") with four slots (meta, diff_submodels, diff_terms, diff_assumptions), plus a fifth diff_metrics slot when metrics = TRUE.

Examples

# a <- symbolize(lm(mpg ~ wt, data = mtcars))
# b <- symbolize(lm(mpg ~ wt + hp, data = mtcars))
# compare_symbolic(a, b)