inlabru : matrix dimensions problem

[Mouniaz96]

Dear all,

I've sent recently a question about spatial covariates effects varying over time (here named Shop), and since then I've tried to compute something like this :

formula = Cars ~ Shop_time (hour.index, Shop, model='ar', order=2) + ...

inlabru.shop=bru(formula , data = data.est , family = "poisson",
                 options=list(inla.mode="experimental",control.compute=list(cpo = TRUE)))

predict(inlabru.shop, data.pred, ~ exp(Intercept + Shop_time + ...)

and I have this error message :

'Error: Matrices must have same dimensions in Matrix(e1) * e2'

The prediction data 'data.pred' has the same names and number of columns as data.

I have this message only when I include this Shop_time term.

The code is the following :

require(INLA)
require(inlabru)

data1.csv data2.csv

data.est=read.csv('data1.csv', header=TRUE, sep=',')
data_pred.mds=read.csv('data2.csv', header=TRUE, sep=',')  #Prediction data

#Formula
shop.mds  <- Cars ~ Shop_time(hour.index, Shop, model='ar', order=2)+
  noise(main=loc.index, model = "iid", constr = TRUE) + # noise
  sitehour(main=cbind(Xutm, Yutm), model = prior_spde.mds, group=hour.index,  # space-hour interaction
           control.group = list(model = "ar", order=2, scale.model= TRUE)) + 
  hourday(main=hour.index, model='rw2', scale.model = TRUE, cyclic=TRUE,     # space-day interaction
          group=day.index, control.group = list(model = "ar", order=2, scale.model= TRUE))

shopp.mds <- bru(shop.mds , data = data.est, family = "poisson", 
                 options=list(inla.mode="experimental",control.compute=list(cpo = TRUE)))

# CPO - DIC  - PIT
shopp.mds$dic$dic 
-mean(log(shopp.mds$cpo$cpo))
shopp.mds$failure
hist(shopp.mds$cpo$pit)
shopp.mds$waic$waic 

##########################################
# PROBLEM Error: Matrices must have same dimensions in Matrix(e1) * e2
##########################################

myPred.mds <- predict(shopp.mds, data_pred.mds, ~ exp(Intercept + Shop_time + sitehour + hourday ))

[#####################################################] Do you know how I can fix this ? Thank you for your help. Best regards.

Mounia

[finnlindgren]

Hi, can you supply the prior_spde.mds object as well so I can test the code here?

Also the traceback() output after you get the error, and your sessionInfo().

[Mouniaz96]

Sorry i forgot, here are the mesh and the prior_spde.mds object

nb_sens=17
mesh <- inla.mesh.2d(loc=data.est[1:nb_sens,c(2,1)], max.edge= c(0.4,2), offset=0.4, cutoff = 0.1)
prior_spde.mds <- inla.spde2.pcmatern(
  mesh, alpha=3/2,
  prior.range = c(1.8, 0.5), # Probability(range < 1.8) = 0.5
  prior.sigma = c(0.6, 0.7)  # Probability(sigma > 0.6) = 0.7
)

[Mouniaz96]

traceback() 19: stop(gettextf("Matrices must have same dimensions in %s", deparse(sys.call(sys.parent()))), call. = FALSE, domain = NA) 18: dimCheck(e1, e2) 17: Matrix(e1) e2 16: Matrix(e1) e2 15: eval(call, parent.frame()) 14: eval(call, parent.frame()) 13: callGeneric(Matrix(e1), e2) 12: val[["weights"]] A 11: val[["weights"]] A 10: amatrix_eval.component(x, data = data, ...) 9: amatrix_eval(x, data = data, ...) 8: FUN(X[[i]], ...) 7: lapply(components, function(x) amatrix_eval(x, data = data, ...)) 6: amatrix_eval.component_list(model$effects[included], data = data, inla_f = TRUE) 5: amatrix_eval(model$effects[included], data = data, inla_f = TRUE) 4: evaluate_model(model = object$bru_info$model, state = state, data = data, predictor = formula, include = include, exclude = exclude) 3: generate.bru(object, data = data, formula = formula, n.samples = n.samples, seed = seed, num.threads = num.threads, include = include, exclude = exclude, ...) 2: predict.bru(shopp.mds, data_pred.mds, ~exp(Intercept + Shop_time + sitehour + hourday)) 1: predict(shopp.mds, data_pred.mds, ~exp(Intercept + Shop_time + sitehour + hourday)) ##################### sessionInfo()

R version 4.2.1 (2022-06-23 ucrt) Platform: x86_64-w64-mingw32/x64 (64-bit) Running under: Windows 10 x64 (build 19044)

Matrix products: default

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

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

other attached packages: [1] inlabru_2.5.2 INLA_22.07.23 sp_1.5-0 foreach_1.5.2 Matrix_1.4-1

loaded via a namespace (and not attached): [1] Rcpp_1.0.9 plyr_1.8.7 pillar_1.8.0 compiler_4.2.1 RColorBrewer_1.1-3 [6] viridis_0.6.2 remotes_2.4.2 iterators_1.0.14 tools_4.2.1 lifecycle_1.0.1
[11] tibble_3.1.8 gtable_0.3.0 lattice_0.20-45 viridisLite_0.4.0 pkgconfig_2.0.3
[16] rlang_1.0.4 cli_3.3.0 rstudioapi_0.13 rgdal_1.5-32 gridExtra_2.3
[21] dplyr_1.0.9 MatrixModels_0.5-0 generics_0.1.3 vctrs_0.4.1 stats4_4.2.1
[26] grid_4.2.1 tidyselect_1.1.2 glue_1.6.2 R6_2.5.1 fansi_1.0.3
[31] sn_2.0.2 tidyr_1.2.0 ggplot2_3.3.6 purrr_0.3.4 magrittr_2.0.3
[36] splines_4.2.1 scales_1.2.0 codetools_0.2-18 mnormt_2.1.0 colorspace_2.0-3
[41] numDeriv_2016.8-1.1 utf8_1.2.2 munsell_0.5.0

[finnlindgren]

I haven't been able to reproduce the problem when running on a lower resolution mesh (to make it faster to get to the predict() part). I tried several inlabru versions, both with your Matrix version and the current version. I'm rerunning with your original mesh now, to see if that triggers the problem.

Can you try upgrading inlabru to at least the CRAN version 2.5.3 and see if that makes a difference?

[finnlindgren]

I also get no error with the full mesh, for inlabru 2.5.2, Matrix 1.4-1.

[finnlindgren]

I notice that trying to save the estimated object to an .rds file generates a huge file (10GB and it hasn't even finished saving it yet). Not sure if that's related, but I think that's a general problem.

[finnlindgren]

I traced the memory use to the inla "configs" output, which is required for predict() to work. To reduce the memeory needed when testing, add the option control.inla=list(int.strategy="eb").

[Mouniaz96]

I tried the 2.5.3 version and I have the same problem. Ok, I will try to reduce the memory right now

[Mouniaz96]

Hi, it didn't help either unfortunately. Is it working for you ? Should I run this on another laptop ? Thank you.

[finnlindgren]

I haven't been able to reproduce the problem here. Can you first try the inlabru development version, in case there's some strange issue that's been resolved there? devtools::install_github("inlabru-org/inlabru")

[finnlindgren]

Also reinstall and/or upgrade the Matrix package, since you have an old version of that.

[Mouniaz96]

Hi again, I think I found the problem, the class of my covariate Shop in the data_pred.mds is 'matrix array' and it is simply coming from that. Thank you for your help.

[finnlindgren]

Interesting; can you confirm that if you run exactly the code you posted above that you don't get the error (as your own code must be reading the variables in some other way to create an array)? A simple Shop_time(hour.index, as.vector(Shop), ...) should fix it.

I'll keep this issue open to remember to check to what extent the inlabru code might coerce inputs to avoid similar issues, or give more informative messages than the built-in R error.

[Mouniaz96]

Hi Finn, I tried to run the same code that I posted and it was incredibly slow and I think I didn't get the same error. So I believe that it is just coming from the table, and the way I saved it that transformed the covariate for some reason...