compute_nll: Compute negative log-likelihood for EcoState model
In ecostate: State-Space Mass-Balance Model for Marine Ecosystems

compute_nll

R Documentation

Compute negative log-likelihood for EcoState model

Description

Compute negative log-likelihood for EcoState model

Usage

compute_nll(
  p,
  taxa,
  years,
  noB_i,
  type_i,
  n_species,
  project_vars,
  DC_ij,
  Bobs_ti,
  Cobs_ti,
  Nobs_ta_g2,
  Wobs_ta_g2,
  log_prior,
  fit_eps,
  fit_nu,
  stanza_data,
  settings,
  control
)

Arguments

`p`	list of parameters
`taxa`	Character vector of taxa included in model.
`years`	Integer-vector of years included in model
`noB_i`	Boolean vector indicating which taxa have no B value
`type_i`	character vector indicating whether a taxon is "hetero", "auto", or "detritus"
`n_species`	number of species
`project_vars`	function to integrate differential equation
`DC_ij`	Diet projections matrix
`Bobs_ti`	formatted matrix of biomass data
`Cobs_ti`	formatted matrix of catch data
`Nobs_ta_g2`	formatted list of age-comp data
`Wobs_ta_g2`	formatted list of weight-at-age data
`log_prior`	A user-provided function that takes as input the list of parameters `out$obj$env$parList()` where `out` is the output from `ecostate()`, and returns a numeric vector where the sum is the log-prior probability. For example `log_prior = function(p) dnorm( p$logq_i[1], mean=0, sd=0.1, log=TRUE)` specifies a lognormal prior probability for the catchability coefficient for the first `taxa` with logmean of zero and logsd of 0.1
`fit_eps`	Character-vector listing `taxa` for which the model should estimate annual process errors in dB/dt
`fit_nu`	Character-vector listing `taxa` for which the model should estimate annual process errors in consumption `Q_ij`
`stanza_data`	output from `make_stanza_data`
`settings`	Output from `stanza_settings()`, used to define age-structured dynamics (called stanza-groups).
`control`	output from ecostate_control

Details

Given a list of parameters, calculates the joint negative log-likelihood, where the Laplace approximation is then used to marginalize across random effects to calculate the log-marginal likelihood of fixed effects. The joint likelihood includes the fit to fishery catches, biomass indices, age-composition data, weight-at-age data, priors, and the distribution for random effects.