Exceptions to drop-in replacement: 1. don't require that the attributes are in the same order 2. don't require that the r version returns everything that the tmb version does (e.g. flexmodel) 3. don't require that the row.names are identical (is this ok? the r version counts iterations with skips, but is this informative?) 4. don't require that the call is identical (obvious i guess, but being exhaustive) 5. don't look for parameter called S0 in classic simulation results (FIXME: this one might not be necessary anymore) 6. don't require the ordering of params_timevar rows to be the same (FIXME: this should probably be handled in run_sim itself)
compare_sims(
classic_sim,
tmb_sim,
tolerance = NULL,
compare_attr = TRUE,
na_is_zero = FALSE
)
compare_sims_cbind(classic_sim, tmb_sim, index)
compare_sims_plot(...)
compare_sims_diffmat(classic_sim, tmb_sim, state, op = `-`)
compare_sims_unlist(classic_sim, tmb_sim, tolerance = testthat_tolerance())
result of `run_sim` without using TMB
result of `run_sim` using TMB
numerical tolerance
compare attributes or just the simulations themselves
should NAs be replaced with zeros?
length-one vector for extracting the simulated variable to be compared
Arguments passed on to compare_sims_cbind
vector of state names to compare
binary comparison operator
Other compare:
compare_grads()
,
time_wrap()