exporting to Mark

[roche042]

Hi there,

I am trying to export a multistrata model from RMark to Mark. Although the model seems to export properly using the command export.MARK(), when I attempt to open the .Rinp file in MARK, program Mark ends up not-responding and eventually shutting down. I attempted to run this model (prior to exporting) with reduced PIMs, hoping that this would avoid such a problem. However, my 'full' model will not run with the PIMs reduced (at least, not with the Psi PIMs reduced_. Are there any tricks to reducing PIMs for a multistrata model (specifically the PIMs for Psi) or for importing the exported RMark model into Mark that could help me out?

I'm sorry if I'm being too vague, please let me know what additional information (if any) would help answer this question.

[jlaake]

I assume that you mean use of logit for Psi when you say reduced PIMS. What do you mean when you say that the model with reduced PIMS didn't run? It didn't converge? If so did you try alternate starting values? For the mlogit version, it sounds like MARK is choking on the size of the model. If that is the case, then there is little you can do but to reduce the problem size. This just happened in one other case recently. If you want to email me a zip of the .Rinp, .inp and .tmp files, I can try it on my machine and if that doesn't work I can ask Gary to see if he can tell why it is balking in the import.

--jeff

[roche042]

Yes, I was using logit (I thought) with Psi when I attempted to reduce the PIMs. I can't remember anymore what the error message was that I received from RMark, so I will try again and let you know what I find. I am sure MARK is choking on the size of the model... there are way too many parameters with PIM=all! I will send it along so you can see. Thanks!

erin

[roche042]

Oh, hilarious... the files are too big to be zipped and sent! That's a problem. I'll try the 'reduce PIMs' thing again and cross my fingers.

erin

[jlaake]

If you have some parameters that are only time-dependent then you can use pim.type="time" or if they are all constant then pim.type="constant". However, that shouldn't matter as long as the formula for the global model is only ~time or ~1 respectively because RMark will simplify any parameter to time or constant unless you use the mlogit link. So even though it may start with pim.type="all", as long as you don't use link="mlogit" then it will only use as many parameters as necessary and only the smaller set is sent to MARK in the export. The only exception is mlogit parameters.

I just saw your other post. You can send them with yousendit (https://www.yousendit.com/) and I can retreive. It is free for the occasional use as long as it doesn't exceed a certain size. It sounds like the size is likely the problem. You may have to trim down the global model to get it to work.

--jeff

[roche042]

Hi, thanks, I will try to load the files. I tried running the reduced model again and got the following error:

"Error in if (x4 > x2) { : argument is of length zero"

which seems to be an error I see frequently but do not understand.

erin

[jlaake]

I need to trap that error. That happens when the MARK output file is not complete. The R code searching for strings in the output doesn't find something and that is the error message you get. I need to trap the error and output something more sensible. It is one of the errors that is described in appendix C (C25 I believe). Anyhow, open the most recent marknnn.out file in the working directory and you can work out what the problem was. Most likely it was numerical convergence issue.

--jeff