search

dbspitz / migrateR

Shared tools for analyzing animal movement behavior in R
6 stars 0 forks source link

Force model fit? #7

Closed briana-abrahms closed 5 years ago

briana-abrahms commented 5 years ago

Hi Derek,

Great package, thanks for putting so much work into it! Would it be possible to build into mvmtClass a way to force a model type to be fit? For example, I want all my individuals to be fit to a mixmig model, but currently that doesn't appear for all individuals.

It would also be fantastic if there could be a mixmig option for the elevation model; I've replaced elevation with latitude and it works well. I also like this better since it's not sensitive to starting point. Is this something you'd consider adding?

Thanks, Briana

dbspitz commented 5 years ago

Thanks, Briana.

Glad you're playing with other gradients--cool hearing that latitude is doing the trick for ya. What kind of critters are you working with?

When models fail to fit, it is almost always because of poor starting estimates. (Generating guesses at appropriate starting values for each parameter is the most complex thing migrateR does behind the scenes.) This is why it's not possible to "force" e.g., a mixmig model to fit-- if a model is missing it's because migrateR's best guess isn't good enough and better manual inputs are needed. If you want these models fit for all individuals, you'll have to play around with manually setting starting values (and/or bounds) using the "pEst" function and the "p.est" argument in "mvmtClass".

Right now expansion of migrateR is limited to work I'm directly involved with (such are the vagaries of the paycheck...). I'm certainly willing to consider adding a mixmig model for elevation (and, for that matter, tweaking the code to make altering axis labels easier) and have added these to my list of potential projects for the future. Maintaining the package is a top priority, though. If you encounter any problems I'm always appreciative of a heads up.

Thanks,

Derek Spitz Postdoc, UCSC Bishop, CA (831) 737-3120 CityOfKitesAndCrows.com http://cityofkitesandcrows.com Wilmers Lab http://wildlife.ucsc.edu // Wittmer Lab http://cms.victoria.ac.nz/sbs/research/ecology-biodiversity-research/ecology-and-conservation-biology

On Tue, Jul 2, 2019 at 10:17 AM briana-abrahms notifications@github.com wrote:

Hi Derek,

Great package, thanks for putting so much work into it! Would it be possible to build into mvmtClass a way to force a model type to be fit? For example, I want all my individuals to be fit to a mixmig model, but currently that doesn't appear for all individuals.

It would also be fantastic if there could be a mixmig option for the elevation model; I've replaced elevation with latitude and it works well. I also like this better since it's not sensitive to starting point. Is this something you'd consider adding?

Thanks, Briana

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/dbspitz/migrateR/issues/7?email_source=notifications&email_token=ACTZKUSGHMEMBB67UQRGFSLP5OEQ7A5CNFSM4H46EQX2YY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4G46CPNA, or mute the thread https://github.com/notifications/unsubscribe-auth/ACTZKUUST7UVCS7PZMDLML3P5OEQ7ANCNFSM4H46EQXQ .

briana-abrahms commented 5 years ago

Hi Derek,

Thanks so much for your thoughtful responses here. I've played around with the starting parameters, but it's not doing the trick. I'm working on whooping crane migrations and they perform really nice clean latitudinal migrations, but unfortunately for some birds the data cuts off partway through the year so it looks like it was a disperser. I'd love to tell it to assign it as a migrant or mixed migrant and still estimate the migration timings. I'll make do for now!

Best, Briana