Recurring convergence error when running RMark

posts related to the RMark library, which may not be of general interest to users of 'classic' MARK

Recurring convergence error when running RMark

Postby rmark4salamanders » Wed Oct 19, 2016 3:05 pm

I previously posted about this same problem, and received a response which was, at the time, extremely helpful, but that fix is not working anymore. below is a link to that original conversation:
http://www.phidot.org/forum/viewtopic.php?f=21&t=3309

I am working with the same script and the same capture history and size class data sets as previously. I have updated the precipitation data sets, and would like to run the same models as before as well as several new ones.
Now, 9 times out of 10 when I run a model I get the same old "error -- numerical convergence never reached" in my outfile. This happens with complex models, but also with the null model in which all parameters are held constant, and using the fix described at the link above works only in rare instances, most of the itme even when very similar models are used to define "initial", the model still fails. Survival ~ body size will always run, and survival ~ time will always run, and the null model will run if initial=S~size, but that's as far as I have been able to get, and I would like to be running many more models than this.

Things I have tried that did not work:
- the fix described at the link above
- going back to the original precipitation datasets and running the original models
- running the original models with the original data sets in R version 3.3.1 and 3.3.0
- re-installing R, Mark, and RMark
- restarting the computer
- running the models on two different computers

Any help would be appreciated.
rmark4salamanders
 
Posts: 8
Joined: Wed Apr 20, 2016 12:48 pm

Re: Recurring convergence error when running RMark

Postby jlaake » Mon Oct 24, 2016 6:04 pm

Sorry for my late response. I was out sick last week.

When you get that error, you need to look at the .out file and see what it is saying. If it runs one time and not the other then something changed. Compare the .out files and see what changed. Are you using different versions of R - R32 and R64? Will it work without the precip covariate? Presumably yes if it worked before. If you can get a simple model to run, use it as initial values for the more complex model. You have to be methodical about it. If you can't make any headway, send me the code and data and I'll run on my end and see what is up as best I can. I may have to pass it to Gary as all of the optimization is done in MARK and I can't test that.

--jeff
jlaake
 
Posts: 1417
Joined: Fri May 12, 2006 12:50 pm
Location: Escondido, CA

Re: Recurring convergence error when running RMark

Postby rmark4salamanders » Mon Mar 27, 2017 11:35 am

Hi Jeff,

I did manage figure out what was wrong with my model. I had not standardized my covariates and some of them ranged from about -12 to nearly 30. Once i standardized them, the problem disappeared.

I appreciate your generous offer to look over the code, and I'm sorry I didnt' log back until so many months later.
rmark4salamanders
 
Posts: 8
Joined: Wed Apr 20, 2016 12:48 pm


Return to RMark

Who is online

Users browsing this forum: Google [Bot] and 14 guests