Functionalise control script

[samueldnj]

Wrap the current code in the control script in tools.R and simulation.R functions.

Tools

• Write dat and pin files
• read rep files
• call estimators (msProd, ssProd)
• compare operating model and assessment parameters (MCMC and MLE)

  Simulation:

• Simulate data
• Create lists of biological and observational data, for passing to write tools
• Covariance across time and species

[samueldnj]

Estimator call functions need a trial fit subroutine to modify pin file and retry fitting in case of failure.

[samueldnj]

Fit trials working, for the most part, using mseR code as a reference:

Each trial i = 0:N

• increments leading scale par (lnMSY) by log (i + 1)
• decreases variance of the MSY prior by dividing by (i+1), become increasingly more informative with each trial (I think this might be too aggressive)

The trials are stopped if both a local min is found and the hessian is positive definite, or if N is reached.

If N is reached and either of the conditions are unsatisfied the rep file is returned unchanged, and the mc output from the same species/complex with previous seed value is used - not ideal i know

This seems to work often, but not always.

Other idea:

• read in rep file, use current estimates for new pin file values (not sure how this will go, as the number of reps are often reached)
• increase maxReps? This has addressed most of the problem seed values, tends max effectiveness around somewhere between 5k and 10k iterations. There are a couple of holdouts.
• modify other priors
• fix q? right now it's concentrated out NO, this is one of the only shared priors. Maybe modify the prior in some way - also, allow the mean value to be estimated

[samueldnj]

Using seed values 1:100, model has 86 successful runs that perform MCMC. Given that there are other issues defined for the MCMC comparisons (issue #11) and covariance simulation (issues #3 and #4), I'm closing this issue here.