Command-Line Analyses

announcements (new versions, changes, bugs, installation problems...) related to program MARK

Command-Line Analyses

Postby lscopel » Fri Jun 18, 2021 4:17 pm

Hello,

This approach may sound rather odd, which I am aware of, so apologies for that. I am a very late-stage PhD student running a set of multistate models (8-10 states, ~15k individuals) on an ok pair of PCs. I have discovered that my data likely have local minima, and they won't converge unless I use the simulated annealing algorithm. My models take anywhere from 12 hours to 3 days to run, and when the individual covariates are included they can run twice as long. I have fortunately discovered that my university offers access to a high-speed computing array, and I can upload my data to run on their servers. Unfortunately, all of their machines run Linux, and given that I have been running all my models thus far in the GUI-based Windows program, the solution we have reached thus far is for me to save my model in the GUI, then upload and run the .tmp output from my saved model using the command-line interface that calls mark.exe in Linux. (I realize it would be much more straightforward to call mark.exe using R, but I am much more comfortable using the DM in the GUI, which is how I got to this point.)

That has actually been working ok, but my issue now is what to do with the output. Ideally, when the models have finished running, I'd like to be able to create a results table, perform model-averaging, and export the variance-covariance matrices for each model. I have looked at the help documentation and seen that when creating the results database, there's a .dbf, .fpt, and .cdx file. When I run an individual model command-line in Linux, I get a .VCM file (which I assume is the variance-covariance matrix), a .RES file, and a standard text output file. I have a good knowledge of R but didn't initially have the RAM to run my analyses in RMark; I was able to run the tutorial models, and those output files also look slightly different (.vcv, .res, .out).

So! My questions: is there a way to take my command-line output and recreate a "results database", resembling either GUI-based MARK or RMark? Are there additional command-line functions that can be called to update the .dbf/.cdx files? Or would I need to build each model initially in R to call the results properly? Again, this is quite a mess of softwares and OSes, but any help is very much appreciated. I have had a very hard time finding any information on the syntax for command-line functions, so any help there would also be appreciated.
lscopel
 
Posts: 1
Joined: Fri Jun 18, 2021 3:06 pm

Re: Command-Line Analyses

Postby jlaake » Fri Jun 18, 2021 5:28 pm

First of all this is the wrong sublist for this post. It sounds like you are trying to re-create RMark and it would certainly be easier to just use it. You are welcome to look at RMark code on my GitHub site to see how to specify command line run. You can use save structure in GUI, port .inp file to Linux, run on server, port models back to your machine (note Linux only uses line feed for end of line and windows uses lf/ cr so will need to convert), then you should be able to import back into MARK GUI.
jlaake
 
Posts: 1295
Joined: Fri May 12, 2006 12:50 pm
Location: Escondido, CA


Return to software problems/news

Who is online

Users browsing this forum: No registered users and 1 guest

cron