search

JenniNiku / gllvm

Generalized Linear Latent Variable Models
https://jenniniku.github.io/gllvm/
48 stars 20 forks source link

Time series example #42

Open jebyrnes opened 3 years ago

jebyrnes commented 3 years ago

Hello! Still loving this package. I was wondering if you ever put together/had a good example of a multivariate time series? e.g., to see if some external driver affected community structure over time.

(and if you are looking for one, I've got a great example!)

JenniNiku commented 3 years ago

Hi! Nice to hear that. I have started working on such model that could take into account time dependency in the data, so at least something related to that might come sometime in a few months.

jebyrnes commented 3 years ago

That's awesome! If you are interested, I have access to a 30 year quarterly sampled fixed quadrat (n=45) study with ~150 species, if you are interested in looking that as a demo dataset. I've also got access to a shorter-term LTER dataset of ~100 species at fixed transects from 2000-2020 in kelp forests sampled annually. If that's of interest, feel free to email me or let me know here!

JenniNiku commented 3 years ago

Yeah. I think we have rather short time series data at the moment, so those datasets you mentioned sound very interesting.

JonJup commented 11 months ago

Hi, I was wondering whether you made progress on the time series aspect. Thanks !

BertvanderVeen commented 11 months ago

Hello Jonathan. Structured row effects are now implement in the package, see #1, which allow for an AR1 random row effect (see row.eff in gllvm(.)). @JenniNiku is working on functionality for structured latent variables, but (I think) that is not ready yet.

JonJup commented 11 months ago

Thanks Bert for the quick answer! Is it possible to have a combined structured/ unstructured row effect? We have time series data from different mesocosms. I have gotten the model working with a corAR1 row effect for time, but dont know if I can still add a random effect for the mesocosm now.

BertvanderVeen commented 11 months ago

I'm afraid not. At present gllvm is limited to a single random row effect. Could you consider having an additional fixed row effect instead?

JonJup commented 11 months ago

Thanks again! Yes thats what we will porbably do in that case.

JenniNiku commented 11 months ago

Hi, there is also a possibility for defining correlation structure for LVs, and I will write a vignette for these correlation structure options as soon as I have time, but my time for developing the package is limited at the moment.

rialsc commented 9 months ago

Hello Jonathan. Structured row effects are now implement in the package, see #1, which allow for an AR1 random row effect (see row.eff in gllvm(.)). @JenniNiku is working on functionality for structured latent variables, but (I think) that is not ready yet.

Hello! Ive looked through the documentation and many examples, Do you have an example of how to write this out? Ive tried different variations of this but it hasnt worked for me: row.eff = corAR1(~1| Age_myr)

Thank you!

BertvanderVeen commented 9 months ago

Hello Jonathan. Structured row effects are now implement in the package, see #1, which allow for an AR1 random row effect (see row.eff in gllvm(.)). @JenniNiku is working on functionality for structured latent variables, but (I think) that is not ready yet.

Hello! Ive looked through the documentation and many examples, Do you have an example of how to write this out? Ive tried different variations of this but it hasnt worked for me: row.eff = corAR1(~1| Age_myr)

Thank you!

Sorry about that, the vignette is still under construction. row.eff=~corAR1(1|time) should do it I'd think!