row- vs. column-major ordering of nodes and char representations of nodes

[paciorek]

At the moment, we have row-major ordering of nodes, regardless of how the index vars are arranged in the declaration

nimble is inconsistent, so that may open the opportunity to decide to use column-major as most natural given how R's behavior.

code <- nimbleCode({
  for(i in 1:2)
    for(j in 1:3)
      y[i,j]~dnorm(0,1)
})

m <- nimbleModel(code)
sapply(m$getNodes(), \(x) x$toNodeChars())

[1,] "y[1, 1]"
[2,] "y[1, 2]"
[3,] "y[1, 3]"
[4,] "y[2, 1]"
[5,] "y[2, 2]"
[6,] "y[2, 3]"

code <- nimbleCode({
  for(i in 1:2)
    for(j in 1:3)
      y[j,i]~dnorm(0,1)
})

m <- nimbleModel(code)
sapply(m$getNodes(), \(x) x$toNodeChars())

     [,1]     
[1,] "y[1, 1]"
[2,] "y[1, 2]"
[3,] "y[2, 1]"
[4,] "y[2, 2]"
[5,] "y[3, 1]"
[6,] "y[3, 2]"

[perrydv]

See my comment on the nimble issue you linked to. Arguably nimble is not inconsistent but makes a natural default choice to order tied nodes in their declaration order. I agree since now we handle ties as ties we can also determine a different default ordering when listing them and I agree column-major makes sense as consistent with R's storage order. Do we have a separate feature to list elements of a (arbitrary subset of a) variable in column-major order without sorting (e.g. for purposes of setting up MCMC output columns)? That would be useful in the same way expandNodeNames is in nimble.