[Developers] Control files and basic MCMC

[Arni Magnusson]

I have uploaded the pella.tpl model with data and documentation to the

   http://admb-project.org/examples/fisheries

directory, next to Dave's pella-t.tpl model.

(I wrote the model several years ago and today I would handle the 
Pella-Tomlinson shape parameter differently. The current implementation 
does not allow Bmsy to go below 0.368K. If we ignore that unnecessary 
constraint, this example is useful as a discussion platform about control 
files and basic MCMC.)

To run the model, unzip the input files and try:

   pella -ind hake_schaefer.dat

With this data filename, the model will automatically look for parameter 
settings in hake_schaefer.ctl.

---

My point was not that all models should automatically look for a control 
file - although we might opt for that in the end. Rather, my point was 
that we could probably make it easier for the ADMB programmer to implement 
such control files.

Lines 55-60, 64-87, and 115-120 are just a lot of repeated patterns. To a 
programmer, this means that a new function that handles this general task 
would make the code shorter, more readable, and less likely to have 
errors.

In pseudocode, plui(logk,1,2,3,4) would set the estimation phase, lower 
bound, upper bound, and initial value for that parameter, all in one 
function call. This functionality would be very handy for a model with 60 
parameters.

---

Less important, but easier to improve, is the fact that ADMB does not give 
the programmer access to the *.dat filename. As a result, lines 47-53 are 
an ugly hack to find the *.dat filename. This is a simple matter of having 
ADMB export the existing variable "datname" to user scope, where it can be 
useful.

---

My second point was that ADMB users deserve something prettier and more 
accessible than mymodel.psv for their basic MCMC output. In other words, 
text files with a header line showing column names. Lines 24, 125-126, and 
171-194 show that this is currently somewhat more arduous than just typing 
mymodel -mcmc. For one thing, I could see MCMC_SECTION be a synonym for 
if(mceval_phase()).

Arni