RMark Errors

[musselman]

Does anyone have any familiarity with this error? I think it has to do with the number of parameters in my mutistate model (i.e., 5 states) and size of the capture histories (i.e., 130 recaptures).

Code: Select all

Program received signal SIGSEGV: Segmentation fault - invalid memory reference.

Backtrace for this error:
#0  ffffffffffffffff

Warning: output from MARK was not complete
 Show Traceback
 
 Rerun with Debug
 Error in if (x4 > x2) { : argument is of length zero In addition: Warning message:
running command '"c:/Program Files (x86)/Mark/mark64.exe" i=mark320.inp o=mark320.out v=mark320.vcv r=mark320.res threads=-1' had status 3


Any thoughts on how I can deal with this? Should I keep reducing the size of my capture histories/number of individuals until I get a file size that is estimable in Mark-- which seems to work, but inevitably reduces model power and generally just sucks to have to figure out...

Please send help... I'm running out of Previcid.

- musselman

[jlaake]

First thing you should do is look at the .out file to see what errors MARK is giving if any. In version 2.1.8 of RMark(not yet posted to CRAN), an error like this will open the .out file automatically.

If you don't have a problem with your data (any errors from MARK.exe), then it could be memory if you meant 130 recapture occasions. The number of capture histories will typically not be a problem. Are all of your transitions being estimated or are some set to 0? If some are set to 0, then you can set the mlogit0 argument to TRUE in the call to mark or mark.wrapper and it will reduce the number of parameters by using logit links for those parameters. If for example only 10 of the 20 possible transition parameters are non-zero, it will reduce the number of parameters nearly in half.

Beyond that make sure you are running R64 because it will run mark64.exe and you'll be able to access more memory. If you don't have enough memory in your computer, stop buying previcid and buy a new computer or more memory.

regards --jeff

[musselman]

Warning from the .out file:

Code: Select all

   * *  WARNING  * *   Error number 5014 from memory allocation of VCRMAT.

   * *  WARNING  * *   Attempt to allocate an allocated object


Yeah, I meant recapture occassions... in total, 400 individual capture histories with 130 recapture occassions each. Lengthy datasets, however it doesn't appear to be a RAM memory issue (6GB) and I am running all 64-bit programs...

Thanks for the reply.

- musselman

[jlaake]

I'm guessing a bit but I'd think vcrmat is the variance-covariance matrix of the real parameters which will be quite large for Psi if you have 20 parameters for each occasion. Just Psi would have a v-c matrix of the real parameters containing over 6 million values and that assumes you have only one group. You'll have to do one of the following:

1) restrict some Psi to 0 if appropriate,
2) reduce the number of capture occasions
3) use pim.type="constant" or "time" which restricts what you can model
4) convince Gary to make calculation of v-c for real parameters an option.

I tried #4 a long time ago and he said no so you better stick with the first three.

--jeff

[bacollier]

Warning from the .out file:

Code: Select all

   * *  WARNING  * *   Error number 5014 from memory allocation of VCRMAT.

   * *  WARNING  * *   Attempt to allocate an allocated object


Yeah, I meant recapture occassions... in total, 400 individual capture histories with 130 recapture occassions each. Lengthy datasets, however it doesn't appear to be a RAM memory issue (6GB) and I am running all 64-bit programs...

Thanks for the reply.

- musselman

This is a memory error, its trying to allocate memory in your system you don't have so its telling you what you want to allocate is already used. 6GB ram is not much relative to what is likely the size of what MARK is building in the background (Variance-covariance matrix and such for MS models are large as I have learned). Not knowing about your model structure, I have hit memory issues on much smaller data sets with significantly fewer occasions than this. You will probably have to either 1) include some restrictions so your models are not so complex, or 2) probably get more memory.

\bret