Skip to contents

Set up dsem series in a simulation list

Source: R/sim_dsem.R
Setup_Sim_DSEM.Rd

Two ways in. From a fit: pass data and pars and the operating model draws from the fitted arrows at their estimates, either every year or with the fitted years kept and only later years drawn. From scratch: pass dsem_arrows and dsem_values and it draws from arrows you wrote at values you chose, which is how a self test states its truth.

Usage

Setup_Sim_DSEM(
  sim_list,
  data = NULL,
  pars = NULL,
  rep = NULL,
  dsem_cov_use = NULL,
  condition_on_fit = FALSE,
  dsem_arrows = NULL,
  dsem_values = NULL,
  pars_by_sim = NULL,
  rep_by_sim = NULL,
  dsem_processes = "rec",
  dsem_cov_mu = NULL,
  dsem_cov_obs_sd = NULL,
  dsem_cov_family = NULL,
  dsem_cov_tweedie_p = NULL,
  dsem_cov_link = NULL,
  mod_var_logscale = FALSE,
  dsem_variance = "conditional"
)

Arguments

sim_list

Simulation list, for example from condition_closed_loop_simulations or the Setup_Sim_* calls.

data

Data list of the fitted model, set up by Setup_Mod_DSEM. NULL from scratch.

pars

Parameter list at the fitted values, as get_optim_param_list or obj$env$parList() returns it. NULL from scratch.

rep

Report of the fit. Needed for a numbers at age series, whose NAA_pred turns the fitted log state into innovations, and for a growth series linked under selectivity at length, which the rebuilt keys are read through.

dsem_cov_use

Optional matrix [year, covariate] of 0/1 flags for the years each covariate is observed in the simulation. Defaults to the fitted pattern for fitted years and every year after, or every year from scratch.

condition_on_fit

TRUE keeps every fitted year at its estimate and draws later years given them. FALSE (default) draws every year.

dsem_arrows

From scratch: the arrow lines, one per element or one per line of a string, naming covariates and deviation series.

dsem_values

From scratch: named vector giving every arrow parameter its value, sds on the natural scale, for example c(b_env = 0.5, rho_env = 0.6, sd_env = 1, sd_rec = 0.8).

pars_by_sim

List of one parameter list per replicate, as parList returns them, used when every replicate has its own draw so the conditioning years read that replicate's values rather than the fit's. NULL (default) uses pars for all of them.

rep_by_sim

List of one report per replicate, alongside pars_by_sim. NULL (default) uses rep for all of them.

dsem_processes

From scratch: processes the arrows may name, "rec" (default), "NAA", "fish_q" or "srv_q", in any combination.

dsem_cov_mu

From scratch: named vector of covariate means, one per covariate the arrows name. Default zero for each.

dsem_cov_obs_sd

From scratch: named vector, one per covariate, of the spread of its observations on its family's own scale: the sd for "normal", the CV for "gamma", the sd of the log for "lognormal", the dispersion for "tweedie"; NA for a covariate observed without error. Default NA.

dsem_cov_family

From scratch: named character vector, one per covariate, of the family the replicates' observations are drawn from, as in Setup_Mod_DSEM. Default "normal" where dsem_cov_obs_sd is given and "fixed" otherwise. A link-scale family reads dsem_cov_mu and the arrows on that scale.

dsem_cov_tweedie_p

From scratch: named vector of tweedie powers, strictly between 1 and 2. Default 1.5.

From scratch: named character vector of links, as in Setup_Mod_DSEM. Default each family's own. A "gaussian_fixed_sd" covariate from scratch takes its one sd from dsem_cov_obs_sd for every year.

mod_var_logscale

From scratch: passed to read_dsem_arrows.

dsem_variance

From scratch: passed to read_dsem_arrows as variance, "conditional" (default) or "diagonal". From a fit the data list's own setting is used.

Value

sim_list with the dsem arrows, values and observation pattern added, plus the deviation arrays and the growth or movement arguments the operating model needs to consume a drawn series.

What a drawn series feeds

Each linked series is written into the array the operating model reads, at the cells the fit linked, so those cells are never copied from the report:

  • recruitment into ln_RecDevs, which generate_recruitment reads. Linking recruitment also replaces both rows of the operating model's ln_sigmaR with the series' settled marginal sd under the arrows, since that is what the fit's initial age deviation penalty reads: whatever Setup_Sim_Rec was given is overwritten, and the initial age deviations are then drawn at the same spread the fit assumes. A moderated variance arrow on recruitment leaves ln_sigmaR as given, since the initial ages need one number and a series whose variance is driven by another never settles on one. Only the hand off is skipped: every deviation is still bias corrected, the initial ages at whatever ln_sigmaR holds and the recruitment cells year by year off the marginal variance. Both sides read the same gate, so a moderated series still has the operating model and the fit agreeing, on the supplied ln_sigmaR rather than on an arrow derived value.

  • catchability into ln_fish_q_devs or ln_srv_q_devs, and draw_sim_q_devs scales that fleet's catchability by them whatever the operating model's own fish_q_model or srv_q_model says. The catchability the deviations scale is the block mean, so the series drives the conditioning years and every year past them alike.

  • the numbers at age into naa_eta_all, as the log state less its deterministic prediction, replacing those cells of the replicate's own innovations.

  • growth and movement into arrays built here, since the operating model has none of its own: fitted years at their estimates and later years at zero. The fit's own Get_Growth or Get_Movement is then rerun per replicate to rebuild weight at age, the size-age keys and the movement matrix, reading every argument from data and pars by name. Under selectivity at length a fleet's selectivity at age is its curve read through the rebuilt key, as in the fit.

Growth propagated cohort by cohort (growth_tv_type = 1) runs the way the fit runs it: the years before growth_cohort_styr are built up front, and from there run_annual_cycle advances each replicate one year at a time from its own start of year numbers. From scratch recruitment, the numbers at age and catchability can be linked, since those are the arrays the operating model has itself, and a catchability series there needs its fleet set to "dsem" in Setup_Sim_q_devs to be read.

See also

Setup_Mod_DSEM