Skip to contents

Set likelihood and penalty weights for the estimation model

Source: R/setup_weighting.R
Setup_Mod_Weighting.Rd

Assigns the \(\lambda\) multiplier on each likelihood component and penalty term, which is how a noisy data source is down-weighted, how an iterative reweighting such as Francis is applied, and how a component is switched off entirely, with a weight of 0. Call after every data setup function.

Usage

Setup_Mod_Weighting(
  input_list,
  addtocomp = 0.001,
  comp_const_obs = 1,
  addtofishidx = 1e-04,
  addtosrvidx = 1e-04,
  addtotag = 1e-10,
  Wt_Catch = 1,
  Wt_FishIdx = 1,
  Wt_SrvIdx = 1,
  Wt_Catch_pop = 1,
  Wt_FishIdx_pop = 1,
  Wt_SrvIdx_pop = 1,
  Wt_Rec = 1,
  Wt_Init_Rec = NULL,
  Wt_F = 1,
  Wt_Tagging = 1,
  Wt_FishAgeComps = array(1, dim = c(input_list$data$n_regions,
    length(input_list$data$years), input_list$data$n_seas, input_list$data$n_sexes,
    input_list$data$n_fish_fleets)),
  Wt_SrvAgeComps = array(1, dim = c(input_list$data$n_regions,
    length(input_list$data$years), input_list$data$n_seas, input_list$data$n_sexes,
    input_list$data$n_srv_fleets)),
  Wt_FishLenComps = array(1, dim = c(input_list$data$n_regions,
    length(input_list$data$years), input_list$data$n_seas, input_list$data$n_sexes,
    input_list$data$n_fish_fleets)),
  Wt_SrvLenComps = array(1, dim = c(input_list$data$n_regions,
    length(input_list$data$years), input_list$data$n_seas, input_list$data$n_sexes,
    input_list$data$n_srv_fleets)),
  Wt_FishAgeComps_pop = array(1, dim = c(input_list$data$n_pop,
    input_list$data$n_regions, length(input_list$data$years), input_list$data$n_seas,
    input_list$data$n_sexes, input_list$data$n_fish_fleets)),
  Wt_SrvAgeComps_pop = array(1, dim = c(input_list$data$n_pop, input_list$data$n_regions,
    length(input_list$data$years), input_list$data$n_seas, input_list$data$n_sexes,
    input_list$data$n_srv_fleets)),
  Wt_FishLenComps_pop = array(1, dim = c(input_list$data$n_pop,
    input_list$data$n_regions, length(input_list$data$years), input_list$data$n_seas,
    input_list$data$n_sexes, input_list$data$n_fish_fleets)),
  Wt_SrvLenComps_pop = array(1, dim = c(input_list$data$n_pop, input_list$data$n_regions,
    length(input_list$data$years), input_list$data$n_seas, input_list$data$n_sexes,
    input_list$data$n_srv_fleets)),
  Wt_Discard = 1,
  Wt_Discard_pop = 1,
  Wt_D = 1,
  Wt_FishAgeComps_discard = array(1, dim = c(input_list$data$n_regions,
    length(input_list$data$years), input_list$data$n_seas, input_list$data$n_sexes,
    input_list$data$n_fish_fleets)),
  Wt_FishLenComps_discard = array(1, dim = c(input_list$data$n_regions,
    length(input_list$data$years), input_list$data$n_seas, input_list$data$n_sexes,
    input_list$data$n_fish_fleets)),
  Wt_FishAgeComps_discard_pop = array(1, dim = c(input_list$data$n_pop,
    input_list$data$n_regions, length(input_list$data$years), input_list$data$n_seas,
    input_list$data$n_sexes, input_list$data$n_fish_fleets)),
  Wt_FishLenComps_discard_pop = array(1, dim = c(input_list$data$n_pop,
    input_list$data$n_regions, length(input_list$data$years), input_list$data$n_seas,
    input_list$data$n_sexes, input_list$data$n_fish_fleets)),
  Wt_Fish_caal = array(1, dim = c(input_list$data$n_regions,
    length(input_list$data$years), input_list$data$n_seas,
    length(caal_row_lens(input_list$data)), input_list$data$n_sexes,
    input_list$data$n_fish_fleets)),
  Wt_Srv_caal = array(1, dim = c(input_list$data$n_regions,
    length(input_list$data$years), input_list$data$n_seas,
    length(caal_row_lens(input_list$data)), input_list$data$n_sexes,
    input_list$data$n_srv_fleets)),
  fish_sel_pen_wts = NULL,
  ret_sel_pen_wts = NULL,
  srv_sel_pen_wts = NULL
)

Arguments

input_list

Named list with $data, $par, $map and $verbose.

addtocomp

Small constant added to the composition proportions to avoid log(0). Default 1e-3. Ignored by the logistic normal, which handles zeros itself.

comp_const_obs

Integer switch for where addtocomp enters the multinomial, not a constant to tune. 1 (default) adds it to the observed proportions that weight the likelihood as well as inside the logarithms, so the likelihood is stationary at pred = obs; 0 weights by the raw observed proportions. With a Dirichlet-multinomial conditional age-at-length fleet, 1 warns, since the constant biases theta upward when most age bins in a length bin are structurally empty.

addtofishidx, addtosrvidx

Small constants added to the fishery and survey indices. Default 1e-4.

addtotag

Small constant added to the tag recovery observations. Default 1e-10.

Wt_Catch, Wt_FishIdx

Weights on the catch and fishery index likelihoods, a scalar or an array [n_regions × n_years × n_seas × n_fish_fleets]. Default 1.

Wt_SrvIdx

Weight on the survey index likelihood, a scalar or an array [n_regions × n_years × n_seas × n_srv_fleets]. Default 1.

Wt_Catch_pop, Wt_FishIdx_pop

The population-specific catch and fishery index weights, a scalar or an array [n_pop × n_regions × n_years × n_seas × n_fish_fleets]. Default 1.

Wt_SrvIdx_pop

The population-specific survey index weight, a scalar or an array [n_pop × n_regions × n_years × n_seas × n_srv_fleets]. Default 1.

Wt_Rec

Weight on the recruitment deviation penalty, a scalar or an array [n_pop × n_regions × n_est_rec_devs], where the third dim is ln_RecDevs's own rather than the number of years, since dont_est_recdev_last and n_proj_yrs_devs both move it. Default 1. A zero leaves a deviation estimated but takes it out of the penalty, which is how a stock-recruit relationship is fit over a window of years while recruitment stays free in every year; dont_est_recdev_last instead removes the deviations, so recruitment reverts to the deterministic prediction.

Wt_Init_Rec

Weight on the initial age deviation penalty, a scalar or an array [n_pop × n_regions × (n_ages - 1) × n_sexes]. NULL (default) takes Wt_Rec when that is a scalar; supply it explicitly when Wt_Rec is an array, since the two penalties are dimensioned differently.

Wt_F

Scalar weight on the fishing mortality deviation penalty. Default 1.

Wt_Tagging

Scalar weight on the tag recovery likelihood. Default 1.

Wt_FishAgeComps, Wt_FishLenComps, Wt_FishAgeComps_discard, Wt_FishLenComps_discard

Weights on the fishery and discard composition likelihoods, a scalar or an array [n_regions × n_years × n_seas × n_sexes × n_fish_fleets]. Default one everywhere.

Wt_SrvAgeComps, Wt_SrvLenComps

Weights on the survey composition likelihoods, a scalar or an array [n_regions × n_years × n_seas × n_sexes × n_srv_fleets]. Default one everywhere.

Wt_FishAgeComps_pop, Wt_FishLenComps_pop, Wt_FishAgeComps_discard_pop, Wt_FishLenComps_discard_pop

The population-specific fishery and discard composition weights, a scalar or an array [n_pop × n_regions × n_years × n_seas × n_sexes × n_fish_fleets]. Default one everywhere.

Wt_SrvAgeComps_pop, Wt_SrvLenComps_pop

The population-specific survey composition weights, a scalar or an array [n_pop × n_regions × n_years × n_seas × n_sexes × n_srv_fleets]. Default one everywhere.

Wt_Discard

Weight on the aggregated discard amount or fraction likelihood, a scalar or an array [n_regions × n_years × n_seas × n_fish_fleets]. Default 1.

Wt_Discard_pop

The population-specific discard weight, a scalar or an array with a leading n_pop dim. Default 1.

Wt_D

Scalar weight on the discard mortality rate deviation penalty. Default 1.

Wt_Fish_caal

Weight on the fishery conditional age-at-length likelihood, multiplying each length row's input sample size. Array [n_regions x n_years x n_seas x n_caal_lens x n_sexes x n_fish_fleets], the shape of ISS_Fish_caal. Default one everywhere.

Wt_Srv_caal

The survey counterpart, with n_srv_fleets last. Default one everywhere.

fish_sel_pen_wts

NULL (default), or a named numeric vector or list weighting any subset of six selectivity smoothness penalties, which are evaluated on the fleet's realized selectivity by bin and year surface and so apply to any functional form: "smooth_bin_curve" and "smooth_bin_diff" are the second and first difference across bins, "smooth_yr_diff" and "smooth_yr_curve" the same across years, "smooth_dome" penalizes non-monotonicity across bins, and "smooth_mean_center" regularizes each year's mean. See resolve_sel_pen_wts and Get_Selex_Smoothness_Penalty. A name left out is 0. Each weight may instead be a vector with one value per model year, so a penalty can act in some years only or at a different strength in each. The specification may also hold "bin_range", the first and last bin the penalties act over. Pass an unnamed list of per-fleet specifications to give each fleet its own. Call after Setup_Mod_Fishsel_and_Q.

ret_sel_pen_wts

As fish_sel_pen_wts, for retained fishery selectivity.

srv_sel_pen_wts

As fish_sel_pen_wts, for survey selectivity. Call after Setup_Mod_Srvsel_and_Q.

Value

input_list with every weight stored in $data under its own name.

See also

Other Model Setup: Setup_Mod_Biologicals(), Setup_Mod_Catch_and_F(), Setup_Mod_Dim(), Setup_Mod_Discard_Comps(), Setup_Mod_FishIdx_and_Comps(), Setup_Mod_Fishsel_and_Q(), Setup_Mod_Movement(), Setup_Mod_Rec(), Setup_Mod_Retsel(), Setup_Mod_SrvIdx_and_Comps(), Setup_Mod_Srvsel_and_Q(), Setup_Mod_Tagging()