runJitter: Function to run TCSAM02 multiple times using jittered initial...

In wStockhausen/rTCSAM02: A set of tools to set up input for, and interact with output from, TCSAM02 model runs

Function to run TCSAM02 multiple times using jittered initial parameter values.

Description

This functions runs a TCSAM02 model multiple times, jittering the initial starting values to assess model convergence.

Usage

runJitter(
  os = "osx",
  path = ".",
  model = "tcsam02",
  path2model = "",
  configFile = "",
  numRuns = 3,
  minPhase = 1,
  onlyEvalJitter = FALSE,
  in.csv = "jitterInfo.csv",
  out.csv = "jitterResults.csv",
  calcOFL = FALSE,
  calcOFLJitter = FALSE,
  mcmc = FALSE,
  mc.N = 1e+06,
  mc.save = 1000,
  mc.scale = 1000,
  saveResults = FALSE,
  cleanup = TRUE,
  keepFiles = c("tmp.sh", "tcsam02.par"),
  cleanupAll = FALSE
)

Arguments

os	'win', 'mac', 'osx', or 'linux'
path	path for model output
model	TCSAM02 model executable name
path2model	path to model executable
configFile	path to model configuration file
numRuns	number of jitter runs to make
minPhase	phase in which to start optimization
onlyEvalJitter	flag (T/F) to only evaluate a (previous) set of jitter runs, not make new runs
in.csv	filename for jitter info (seed, obj fun value) from ADMB model run
out.csv	filename for jittered results
calcOFL	flag (T/F) to perform OFL calculations for "best" model
calcOFLJitter	flag (T/F) to perform OFL calculations while jittering
mcmc	flag (T/F) to run mcmc on "best" model
mc.N	number of mcmc iterations to make
mc.save	number of iterations to skip when saving mcmc calculations
mc.scale	number of iterations to adjust scale for mcmc calculations
saveResults	T/F to save final results to best/ModelResults.RData
cleanup	T/F to clean up SOME model output files after each run
keepFiles	vector of file names to keep, not clean up, after model run
cleanupAll	T/F to clean up ALMOST ALL model output files after each run

Details

For each model run, this function creates a shell script ('./tmp.sh') in the working directory and uses it to run the ADMB version of the TCSAM02 model. Initial model parameters are jittered based on the system clock time as a seed to the random number generator. The seed and final objective function value are saved for each model run in a csv file (the value of out.csv).

When all the models requested have been run, the function determines the seed associated with the 1st model run that yielded the smallest value for the objective function and re-runs the model using this seed to re-create the model run resulting in the minimum objectve function to recreate the model output files. The final model run is done estimating the hessian, so standard deviations for estimated model parameters are available in the .std file.

Value

• list w/ 4 elements: imx - index of (1st) smallest value for the objective function seed - seed resulting in the smallest objective function par - dataframe with par results from run w/ smallest objective function objFuns - vector of objective function values from all model runs parList - list of par dataframes for each model run

wStockhausen/rTCSAM02 documentation built on April 16, 2024, 6:25 a.m.