Error in mgmsampler function

[LiuJL1100]

Hi Jonas,

I run the exampling code in the documantation of mgmsampler function, but got the warning message:

Error in if (!(class(interactions[[i]][[row]]) %in% c("array", "matrix"))) stop("The parameters of each interaction have to be provided as k-dimensional array.") : 
  the condition has length > 1

How to slove this problem?

[Jiahao-ZHU]

I also came across this problem.

My mgm version is "1.2-14", R version is "4.3.2 (2023-10-31)", and RStudio version is "2023.09.1+494"

When I ran example 1 in the mgmsampler() function, I got the same warning message. However, sexamples 2 and 3 (focus on 3-way interaction) are fine. I wonder if there is any problem with 2-way interaction or else.

Hoping for solutions. Thanks.

[Jiahao-ZHU]

I went through the code of the function, and I think this problem might occur because the code: f (!(class(interactions[[i]][[row]]) %in% c("array", "matrix"))) in

for (i in 1:n_order) { n_ints <- length(interactions[[i]]) if (n_ints > 0) for (row in 1:n_ints) if (!(class(interactions[[i]][[row]]) %in% c("array", "matrix"))) stop("The parameters of each interaction have to be provided as k-dimensional array.") }

returns more than one logic element (as for example 1, the result is TRUE TRUE). Therefore, the condition has length > 1.

To solve this problem, I defined a new function mgm_sampler on my own, and replaced the original code with for(i in 1:n_order) { n_ints <- length(interactions[[i]]) if(n_ints > 0) { for(row in 1:n_ints) { if(!inherits(interactions[[i]][[row]], "array") && !inherits(interactions[[i]][[row]], "matrix")) { stop("The parameters of each interaction have to be provided as k-dimensional array.") } } } }

Hope this solution can help. :)

[TomRow90]

Hello,

I was hoping to use this for some simulations I need to run. I just installed the latest developmental version and it looks like this error is still present (it was still present in the CRAN version as well). I tried the workaround posted by Jiahao-ZHU above but I can't quite get that to work. So I was just wondering whether anyone has found another workaround or whether an update to the developmental version could be made? Many thanks.

[jmbh]

I'll try to fix this asap. Meanwhile, an alternative and possibly more convenient way to simulate data with multivariate dependencies between variables with different domains is to simulate from a DAG. This avoids the issues arising from needing a proper joint distribution over all variables (see Appendix F in https://osf.io/preprints/psyarxiv/926pv and the accompanying code in corresponding reproducibility repo).

[TomRow90]

Great, thanks Jonas, and for the tip and link as well.