ovaskain
commented
4 years ago Hi,
There is probably something wrong how the random effects are set up to. Just a quick comment. When you define:
| rL0 = HmscRandomLevel(units = studyDesign$Site) |
|---|
||and then ...("Site" = rL0, ...)||
| it is maybe safer to define |
|---|
|rL0| = |||HmscRandomLevel(units = levels(studyDesign$Site))|||
|||If this does not solve the error message, we should look in more detail how you defined the model object.|||
| Otso |
|---|
On 29.1.2020 13.42, AdMePo wrote:
Hi,
First of all, thank you very much for developping the |Hmsc |package in |R|, and for the very clear papers and vignettes about this very interesting and useful framework.
I'm reporting the following error that popped up while I was estimating a model with a handfull of random effects, while the model was working fine without them:
|[1] "Setting updater$Gamma2=FALSE due to specified phylogeny matrix" [1] "Computing chain 1" Error in Eta[[r]][hM$Pi[, r], ] : out of range index |
Here is the |traceback()|:
|3: computeInitialParameters(hM, initPar) 2: sampleChain(chain) 1: sampleMcmc(m4, thin = thin, samples = samples, transient = transient, nChains = nChains, verbose = verbose) |
I would appreciate some help and feedback to figure out what is happening. One part of my struggle comes from deciding which categorical predictors to treat as random or fixed effects, and the other about how to specify these in the models. Reading the issue "Specification of random effects for nested+longitudinal study design
32 https://github.com/hmsc-r/HMSC/issues/32" did not helped. I
provide some context below.
I'm intending to use |Hmsc |to study microbial communities. I'm currently working on a reduced set of 50 species (|YData|) to get some experience with Hmsc and figure out a script that works fine (typically, I could work with up to 10,000 species). I have environmental (|XData|) and phylogenetic correlation (|CData|) data in hands, trait data (|TData|) will come later.
*
The 108 sampling units have a unique |sample.id| that can be grouped in the following sets of observations: 3 |Replicates |X 2 |Fractions |X 3 |Compartments |X 6 |Sites|. Because I expect these sets share a common property, I would be inclined to treat them all as random effects. I am rigth? What about the |Sample.id|?*
I've defined each |rL| with |HmscRandomLevel()|, /e.g./ for the effect of Site |HmscRandomLevel(units = studyDesign$Site)|. Then I specified |ranLevels = list("Site" = rL0, "Compartment" = rL1, "Fraction" = rL2, "Replicate" = rL3, "Sample.id" = rL4)| in my model. Is that right or I am missing something?Thanks for reading this post. Feel free to ask more details if needed, I'll be happy to provide them.
Adrien
— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/hmsc-r/HMSC/issues/39?email_source=notifications&email_token=AEIYMZUAI7NYRENGUBGV3PTRAFTRRA5CNFSM4KNDFEIKYY3PNVWWK3TUL52HS4DFUVEXG43VMWVGG33NNVSW45C7NFSM4IJQG6DQ, or unsubscribe https://github.com/notifications/unsubscribe-auth/AEIYMZQ45LBI6DX3CD42BKTRAFTRRANCNFSM4KNDFEIA.
AdMePo
Hi,
First of all, thank you very much for developping the
Hmscpackage inR, and for the very clear papers and vignettes about this very interesting and useful framework.I'm reporting the following error that popped up while I was estimating a model with a handfull of random effects, while the model was working fine without them:
Here is the
traceback():I would appreciate some help and feedback to figure out what is happening. One part of my struggle comes from deciding which categorical predictors to treat as random or fixed effects, and the other about how to specify these in the models. Reading the issue "Specification of random effects for nested+longitudinal study design #32" did not helped. I provide some context below.
I'm intending to use
Hmscto study microbial communities. I'm currently working on a reduced set of 50 species (YData) to get some experience with Hmsc and figure out a script that works fine (typically, I could work with up to 10,000 species). I have environmental (XData) and phylogenetic correlation (CData) data in hands, trait data (TData) will come later.The 108 sampling units have a unique
sample.idthat can be grouped in the following sets of observations: 3ReplicatesX 2FractionsX 3CompartmentsX 6Sites. Because I expect these sets share a common property, I would be inclined to treat them all as random effects. I am rigth? What about theSample.id?I've defined each
rLwithHmscRandomLevel(), e.g. for the effect of SiteHmscRandomLevel(units = studyDesign$Site). Then I specifiedranLevels = list("Site" = rL0, "Compartment" = rL1, "Fraction" = rL2, "Replicate" = rL3, "Sample.id" = rL4)in my model. Is that right or I am missing something?Thanks for reading this post. Feel free to ask more details if needed, I'll be happy to provide them.
Adrien