Trouble running multi-season two species

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

Trouble running multi-season two species

Postby Bill » Wed Feb 01, 2012 7:04 am

Hello,
I am having trouble running a two species multi-season analysis for 70 sites over 6 seasons (each season is made up of 7 or 14 surveys). The analysis appears to run well for simpler models, however when I try to include detection as a full identity, my computer will run for over an hour and not produce a result. I am wondering if it is an unreasonable expectation for a computer with 2GB of RAM to be able to process an analysis such as this, or whether there may be another underlying problem. I can run full identity detection models using the same data in single species multi season analyses.
Bill
 
Posts: 10
Joined: Tue Jan 31, 2012 11:18 pm

Re: Trouble running multi-season two species

Postby bacollier » Wed Feb 01, 2012 12:44 pm

Bill,
i suspect Jim/Darryl will chime in later, but a while back in an earlier version of Presence I had some issues with the non-result on a similar sized dataset (90 sites over 5 seasons) but my analysis worked on most updated version. For what its worth, you have plenty of computer, so I suspect there is an underlying problem-whether that is either with the design matrix you have specified, or Presence, but I don't know which.

Can you run a simpler model for p under the 2 species approach, does it converge?

Bret
bacollier
 
Posts: 230
Joined: Fri Nov 26, 2004 10:33 am
Location: Louisiana State University

Re: Trouble running multi-season two species

Postby darryl » Wed Feb 01, 2012 4:12 pm

Hi Bill,
I guess my first question is whether a full identity design matrix makes biological sense? Given the logistics of your sampling, does it make sense to say that detection was one value for all sites in the first survey, then they all had a different value the next time? To me, such a situation really only makes sense if all sites where being surveyed at the same time (or at least within a relatively short time period). If some sites didn't get their 1st survey until after other sites had been surveyed 2 or 3 times, then I don't think an identity matrix is reasonable.

That aside, has PRESENCE hung? Click on the 'black-box' window that hiding behind the GUI and see if there's any action on that screen. Depending on how complicated the model is, it may just be taking a long time to converge.
Darryl
darryl
 
Posts: 495
Joined: Thu Jun 12, 2003 3:04 pm
Location: Dunedin, New Zealand

Re: Trouble running multi-season two species

Postby jhines » Wed Feb 01, 2012 4:38 pm

Hi Bill,

If you check the black command window, which might be behind other windows, you should see a likelihood value printed which changes periodically. For models with lots of parameters, it might take a while to change, but I don't think it should take over a minute. If you see 'Nan' (not a number) or -1ENa, or some other strange number, then the optimization has failed and you can abort it by hitting 'Ctrl-C'.

As Bret suggested, you should try a simpler model first. I'd suggest setting all p's and r's constant over time. Actually, I'd set all parameters constant over time at first. If that works, you can use the beta estimates obtained from that model as starting values for a more complicated model (perhaps making detection probs season-specific). Just repeat a particular starting value for each season. Note: Excel is a nice place to store things for cutting/pasting back into parts of PRESENCE.

If you get stuck, feel free to send me the input file. I haven't seen many people use this model yet, so I'm curious to see it anyway.

Cheers,

Jim
jhines
 
Posts: 599
Joined: Fri May 16, 2003 9:24 am
Location: Laurel, MD, USA

Re: Trouble running multi-season two species

Postby jhines » Thu Feb 02, 2012 4:01 pm

Hi Billie,

I was able to run a model with an identity design matrix for detection, but I used the 2nd parameterization (which computes conditional probabilities, psiA,psiBA,psiBa,pA,pB,rA,rBA,rBa, instead of interaction ratios). Also, I constrained colonization and extinction to be constant over time. The reason for this is that I don't think you have enough data (18 sites per species) to estimate time-specific colonization, extinction and all of the detection probabilities. In addition, I ran some simpler models for comparison.

If you had trouble with the default parameterization (phiA,phiB,phi,...), it's not surprising that you had trouble with some models. This parameterization attempts to estimate the species interaction factor, phi, directly, as opposed to the 2nd parameterization, which estimates the conditional probabilities. The program needs to compute the likelihood value for the model, given the parameters and if you're using the interaction factor, it has to translate that ratio back into conditional probabilities. This can result in numerical problems for the optimization routine when the denominator of the ratio is small. This problem is avoided when using the conditional probabilities (2nd parameterization).

If you're interested in the species interaction factor, you can compute it from the estimates of psiA, psiBA, and psiBa as:

phi = psiBA/(psiA*psiBA+(1-psiA)*psiBa)

Cheers,

Jim
jhines
 
Posts: 599
Joined: Fri May 16, 2003 9:24 am
Location: Laurel, MD, USA

Re: Trouble running multi-season two species

Postby Bill » Fri Feb 03, 2012 8:16 pm

Many thanks to everybody for your helpful suggestions. Jim had a look at the data, and together with your suggestions, was able to put me on the right track (see post above), consequently my two species multi-season models are up and running yielding some interesting results.
Bill
 
Posts: 10
Joined: Tue Jan 31, 2012 11:18 pm


Return to software problems/news

Who is online

Users browsing this forum: No registered users and 10 guests