Code quit working about hmsc

[jarioksa]

Again a tibble! That tibble brings so much pain.

Variables Richness, Complexity and Dist.Refuge are not variables, but they are matrices. Please change them to variables (vectors). Currently their dimensions are given as [46, 1], but it should only read Richness: num -1.935, -0.935... without those dimensions.

See issue #65 which is a duplicate of this.

[morgan-j-black]

Oh! Those darn tibbles! I should have known. Thank you for your quick response. Morgan

…

On Mon., Mar. 28, 2022, 12:07 p.m. Jari Oksanen, ***@***.***> wrote: Again a tibble! That tibble brings so much pain. Variables Richness, Complexity and Dist.Refuge are not variables, but they are matrices. Please change them to variables (vectors). Currently their dimensions are given as [46, 1], but it should only read Richness: num -1.935, -0.935... without those dimensions. See issue #65 <#65> which is a duplicate of this. — Reply to this email directly, view it on GitHub <#138 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AUM5WNPRSNNAW3XA6IXF6KLVCH7PVANCNFSM5R3YRCQA> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[morgan-j-black]

Just a quick follow up... in this instance, the matrices that should have been vectors were generated using the scale function in base r. I had been avoiding the use of tibbles as I know they can be problematic in general.

dat$Richness = scale(dat$richAll) # this code created the matrices in the dataframe.

[brendanf]

@morgan-j-black You are right, base scale() always returns a matrix, even if its argument is a vector, and matrix columns are allowed in base data.frames. Single-column matrices are converted to a vector by the data.frame() constructor function, but this does not happen when they are assigned into an existing data.frame. Try dat$Richness = c(scale(dat$richAll)) to remove the dimension attribute.

(As an aside, as recently as 2018, tibble was more restrictive than base data.frame, and did not allow matrix columns by assignment.)

[jarioksa]

I too hastily put the blame on tibble where we have seen this problem earlier. Actually, it is possible to have even more pathological data frames, such as with poly(x, 2) which adds one variable that is a two-column matrix.

We (or probably I) added the test against matrix entries after issue #65. The basic Hmsc and sampleMcmc commands accept data frames with matrix entries, but then some posterior analysis tools fail for reasons that were outside the Hmsc package (that is, we called functions in other packages such as base and stats and these failed). So we considered it is better to catch these cases before sampleMcmc run, and not weeks later when you finished with sampling and tried to do something with the result. The change was quick and dirty. The minimum is that we need to improve error reporting. I don't have an instant idea to automatically remove the matrix entries from the data frames stored in the result object. This really concerns those auxiliary methods that need access to the original data frame (most of analytic Hmsc tools do not need data frame but they only operate on model matrix which is OK with matrix variables).