Error in fitting spatial poisson BYM models: conflicts between inla function and Matrix package??

[Beth Purse]

Hi there,

I am running some spatial BYM models in INLA.

This code used to run fine on R and package versions in August but is now throwing an error during the inla function which I think may be related to package conflicts between INLA and the Matrix or MatrixModels package.

The error is: Error in validObject(.Object) :
  invalid class “dsparseModelMatrix” object: superclass "replValueSp" not defined in the environment of the object's class

This error is thrown whether I am fitting an intercept only INLA model or a model with covariates in it.

My code is as follows and I have attached the workspace and the session info is also given below. Another colleague can get it to run on his machine and his session info is also given.

Any assistance would be very much appreciated.

Best wishes

Beth

#####################################

load("Workspace_test_09Feb2017.Rdata")
require(INLA)
require(spdep)
require(nlme)
temp <- poly2nb(sind)
#The order of the areas needs to be the same between the data and the spatial polygon object obtained importing the shapefile, so we re-order the data.
str(temp)
ID<-seq(1,80)
nb2INLA("South_India.graph", temp)
Yc$ID <- c(1:80)
load("South_India_data_Feb2017_centred.RData")
nullmod <- inla(bt ~ 1 +f(ID, model="bym",graph="South_India.graph"),
data=Yc,
family="poisson",
#param = c(prior.iid, prior.besag),
#initial = c(initial.iid, initial.besag),
#control.data = list(link="log"),
control.compute=list(dic=TRUE))

My session info is as follows:

Session 3.3.2 (2016-10-31)
Platform: x86_64-w64-mingw32/x64 (64-bit)
Running under: Windows 7 x64 (build 7601) Service Pack 1

locale:
[1] LC_COLLATE=English_United Kingdom.1252
[2] LC_CTYPE=English_United Kingdom.1252  
[3] LC_MONETARY=English_United Kingdom.1252
[4] LC_NUMERIC=C                          
[5] LC_TIME=English_United Kingdom.1252   

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base    

other attached packages:
[1] spdep_0.6-9         nlme_3.1-130        INLA_0.0-1485844051
[4] Matrix_1.2-7.1      sp_1.2-3          

loaded via a namespace (and not attached):
 [1] lattice_0.20-34    deldir_0.1-12      gtools_3.5.0       MASS_7.3-45      
 [5] grid_3.3.2         MatrixModels_0.4-1 coda_0.19-1        gdata_2.17.0     
 [9] LearnBayes_2.15    gmodels_2.16.2     boot_1.3-18        splines_3.3.2    
[13] tools_3.3.2        compiler_3.3.2

A colleague managed to get the models to run successfully, his session info was:

R version 3.3.2 (2016-10-31)

Platform: x86_64-w64-mingw32/x64 (64-bit)

Running under: Windows 7 x64 (build 7601) Service Pack 1

 

locale:

[1] LC_COLLATE=English_United Kingdom.1252  LC_CTYPE=English_United Kingdom.1252    LC_MONETARY=English_United Kingdom.1252

[4] LC_NUMERIC=C                            LC_TIME=English_United Kingdom.1252   

attached base packages:

[1] stats     graphics  grDevices utils     datasets  methods   base    

other attached packages:

[1] spdep_0.6-9         INLA_0.0-1485844051 Matrix_1.2-7.1      maptools_0.8-39     sp_1.2-3          

loaded via a namespace (and not attached):

[1] lattice_0.20-34    deldir_0.1-12      gtools_3.5.0       MASS_7.3-45        grid_3.3.2         MatrixModels_0.4-1

[7] nlme_3.1-128       coda_0.19-1        gdata_2.17.0       LearnBayes_2.15    gmodels_2.16.2     boot_1.3-18      

[13] splines_3.3.2      tools_3.3.2        foreign_0.8-67     compiler_3.3.2   

[INLA help]

Beth,

thanks for your email. I ran your example here without any problems on
linux and on a Mac.

Yes, there is certainly some package crash. I would suggest just to
remove your R and all packages you have installed, and simply reinstall
everything. In my experience, it is way easier to bypass the problem
than to try to solve it.

B

On Wed, 2017-02-08 at 06:35 -0800, Beth Purse wrote:
>
> load("Workspace_test_09Feb2017.Rdata")
> require(INLA)
> require(spdep)
> require(nlme)
> temp <- poly2nb(sind)
> #The order of the areas needs to be the same between the data and the
> spatial polygon object obtained importing the shapefile, so we re-
> order the data.
> str(temp)
> ID<-seq(1,80)
> nb2INLA("South_India.graph", temp)
> Yc$ID <- c(1:80)
> load("South_India_data_Feb2017_centred.RData")
> nullmod <- inla(bt ~ 1 +f(ID, model="bym",graph="South_India.graph"),
> data=Yc,
> family="poisson",
> #param = c(prior.iid, prior.besag),
> #initial = c(initial.iid, initial.besag),
> #control.data = list(link="log"),
> control.compute=list(dic=TRUE))
>

--
Håvard Rue
he...@r-inla.org

[Beth Purse]

Dear Harvard

Thanks so much for taking a look at this for me. I re-installed everything and it works fine now.

Good to know that it is best to "bypass" in these situations.

Thanks again for your time.

Beth

[Pedro Garrett]

Hi,

I also had the same problem and just updated the "Matrix" package and it worked.

Pedro

[Gustavo Machado]

I have the same and by updating MATRIX the problem was fixed

[Hossameldin Mahmoud]

Kindly be informed that I am getting the following error: 

Error in validObject(.Object) : invalid class “dsparseModelMatrix” object: superclass "replValueSp" not defined in the environment of the object's class

when I run the following code 

library(rqpd)

data(bwd)

### A fixed-effects model:

# formula is specified as

fe.form <- dbirwt ~ smoke + dmage + agesq + 

   novisit + pretri2 + pretri3 | as.factor(momid3) 

# estimate the model:

fe.fit <- rqpd(fe.form, panel(lambda=0.5), data=bwd)

can you help me please? 

[Finn Lindgren]

rqpd does not appear to be related to the INLA package.

Finn

--
You received this message because you are subscribed to the Google Groups "R-inla discussion group" group.
To unsubscribe from this group and stop receiving emails from it, send an email to r-inla-discussion...@googlegroups.com.
To post to this group, send email to r-inla-disc...@googlegroups.com.
Visit this group at https://groups.google.com/group/r-inla-discussion-group.
For more options, visit https://groups.google.com/d/optout.

[Ema Biscaia]

Hello everyone!

I just updated my R and RStudio versions to the latest ones and although I was able to install INLA, I'm getting this message when I run "library(INLA)":

• Not a validObject(): no slot of name "srs" for this object of class "RasterLayer"

I have already updated the Matrix package but it continues to give the error.

I found this error by noticing that the formula for my model had "unknown" written in the environment.

Can you help me?

Thank you a lot!

Have a nice weekend

Ema

[Finn Lindgren]

What lead you to think it has to do with the Matrix package? The RasterLayer message rather seems related to the raster package. Please provide the traceback() output you can get after the error message, and also the full sessionInfo() output.

Finn

To view this discussion on the web, visit https://groups.google.com/d/msgid/r-inla-discussion-group/8cb44c39-0a9d-4698-8136-cc8845cf7f02n%40googlegroups.com.

[Finn Lindgren]

Also make sure you run in a clean R environment, with no automatically loaded .RData session data etc.

Finn

[Ema Biscaia]

Thank you for your answer, Finn!

Well, as I read in this topic that upgrading the matrix package did work in the past I tried it if I could fix my problem but I couldn't.

I tried next to upgrade the raster package with no success either.

I was able to open the package INLA withou any problem but now I have something like this:

and I can't run my INLA model which I did using the R version 4.1.1.

If there's no solution I think I will just return to the older version where everything worked fine!

Ema

[Finn Lindgren]

Hi, that's a warning message on a specific line of code and not an error on library(INLA). First of all, check if its the as() call that gives the warning or if its the inla.sp2segment() call, and let me know.

Finn

To view this discussion on the web, visit https://groups.google.com/d/msgid/r-inla-discussion-group/be7d5e4a-974d-4929-8940-55b06fb569d0n%40googlegroups.com.

[Finn Lindgren]

Can you send me you bound.pt object so I can test it here?

Finn

[Finn Lindgren]

Correction, it's the pol.PT object I need.

What kind of object is it?

I suspect the error comes from the raster package, but it's not clear why that gets involved here, if pol.Pt is a polygon object of some type.

Finn

--

Finn Lindgren
email: finn.l...@gmail.com

[Ema Biscaia]

Hi Finn

My pol.PT is an "sf" "data.frame" object that I created from reading with st_read() the shapefile I send above.

[Ema Biscaia]

About this, I rerunned the code and it didn't gave me errors when compiling the bound.pt but it continues to give this massive error when I run my model that works perfectly when I use R 4.1.1.

I don't know if it helps but my "form" that's a formula object has written in the environment this: (unknow), but it runs I doesn't seem to be the problem because, again, in R 4.1.1 it has the same thing and the model works. 

I'm very grateful for your help!

Ema

[Finn Lindgren]

Please show str(form)

Finn

To view this discussion on the web, visit https://groups.google.com/d/msgid/r-inla-discussion-group/d0c70619-b5c2-4ee6-b15f-a255988a2653n%40googlegroups.com.

[Ema Biscaia]

This is the str(form):

[Finn Lindgren]

Where’s the “unknown” part you mentioned before?

Finn

On 1 Aug 2023, at 15:04, Ema Biscaia <e.bi...@campus.fct.unl.pt> wrote:

This is the str(form):

To view this discussion on the web, visit https://groups.google.com/d/msgid/r-inla-discussion-group/e2b321e0-9aa9-48be-b2a2-dfb82d1a0b68n%40googlegroups.com.

[Finn Lindgren]

But if the conversion to bound.pt no longer failed, what did you change?

It’s difficult to help when the problem seems to be changing all the time.

Finn

On 1 Aug 2023, at 15:17, Finn Lindgren <finn.l...@gmail.com> wrote:

Where’s the “unknown” part you mentioned before?

[Ema Biscaia]

Hello, Finn

I'm sorry I'm only replying now but I have been trying to understand what is happening to R and the new version and all of that.

I'm now able to run every line of my code before the model. When I try to run the model with this code:

inla(form, family = "weibullsurv", data = inla.stack.data(stk),

    control.predictor = list(A = inla.stack.A(stk), compute = TRUE),
    control.compute=list(config = TRUE))

I get the following error:

Error in inla.core(formula = formula, family = family, contrasts = contrasts, : With control.fixed = list(expand.factor.strategy='model.matrix'), then NA's in factor are not allowd. Please use strategy 'inla' instead.

*** inla.core.safe: inla.program has crashed: rerun to get better initial values. try=1/2 Error in inla.core(formula = formula, family = family, contrasts = contrasts, : With control.fixed = list(expand.factor.strategy='model.matrix'), then NA's in factor are not allowd. Please use strategy 'inla' instead. *** inla.core.safe: inla.program has crashed: rerun to get better initial values. try=2/2 Error in inla.core(formula = formula, family = family, contrasts = contrasts, : With control.fixed = list(expand.factor.strategy='model.matrix'), then NA's in factor are not allowd. Please use strategy 'inla' instead. Error in inla.core.safe(formula = formula, family = family, contrasts = contrasts, : *** Failed to get good enough initial values. Maybe it is due to something else.  

I then added control.fixed=list(expand.factor.strategy='inla') and it worked but I'm not sure that is what I want because I want to work with dummy variables in my factor variables (var1 and var2). Does this added line change the results of the model?

Besides all this I don't have any NA in my factor variables. I only have NA in my time2 variable, indicating the right censorship. Is this maybe the problem?

Thank you again for all your time spent

Ema

[Finn Lindgren]

I'm not sure about the time1 NAs, but for Var1 and Var2, the NAs are introduced by the inla.stack() construction.  You're much better off doing the conversion to dummy (0/1) variables yourself first, since that would give you complete control over how the factor variables are interpreted. E.g. first calling "MyVars = model.matrix(~ 1 + Var1 + Var2, data = ...)", and then using "... ~ 0 + MyVars" instead of "... ~ Var1 + Var2" in the inla formula, and as (matrix) input to inla.stack would probably give you what you want.

Finn

To view this discussion on the web, visit https://groups.google.com/d/msgid/r-inla-discussion-group/d0a54017-5117-44c1-824c-7e314ebcc009n%40googlegroups.com.

[Ema Biscaia]

Well, yes, I don't think the problem is with the NA's in time2 because I changed them to Inf and I still have the error.

I applied your suggestion to the formula but I don't understand how to do it in the stack. Should I change the "A" argument or the "effects" one?

Thank you

Ema

[Finn Lindgren]

The effects part. MyVars=MyVars instead of Var1=Var1. It’s been a while since I used it with a matrix, so there may be some detail I’ve forgotten, but essentially a matrix should work just as a regular variable.

Finn

On 11 Aug 2023, at 11:13, Ema Biscaia <e.bi...@campus.fct.unl.pt> wrote:

Well, yes, I don't think the problem is with the NA's in time2 because I changed them to Inf and I still have the error.

To view this discussion on the web, visit https://groups.google.com/d/msgid/r-inla-discussion-group/c3320383-24f3-4d65-9bd8-88aad4da43b5n%40googlegroups.com.

[Ema Biscaia]

It worked!! Thank you a lot, Finn!

I now have to compare the results with this version of R (4.3.1) with the older one I was working on. Is it expected that the results differ?

Ema