[upstream rENA 0.3.1] ena.correlations(dims=) breaks for any subset that is not 1:n

[HUDongpin]

This is an issue in rENA 0.3.1, not in ena-python. It is tracked here because it came out of porting rENA to Python. Reported as an observation, not an accusation — it may be fixed upstream or intended, in which case please correct me. Full write-up: docs/rena-upstream-issues.md

Summary

ena.correlations's documented dims parameter only works for a contiguous range starting at 1. Any other subset raises subscript out of bounds, even when every requested dimension exists.

Severity: low — the parameter works for its default-shaped values, so this only bites when selecting other dimensions.

Reproduce

Tested with rENA 0.3.1, R 4.4.2.

library(rENA)
rows <- data.frame(
  unit  = rep(paste0("U", 1:6), each = 4),
  conv  = rep(rep(c("c1", "c2"), each = 2), times = 6),
  grp   = rep(c("g1", "g2"), each = 12),
  score = rep(c(10, 20, 30, 40, 50, 60), each = 4),
  A = c(1,0,1,0, 0,1,1,1, 1,1,0,1, 0,0,1,0, 1,0,0,1, 0,1,1,0),
  B = c(0,1,1,1, 1,1,0,0, 0,1,1,0, 1,0,0,1, 0,1,1,0, 1,0,0,1),
  C = c(1,1,0,0, 1,0,1,0, 1,0,0,1, 0,1,1,1, 1,1,0,0, 0,0,1,1),
  D = c(0,0,1,1, 0,1,0,1, 0,1,1,0, 1,1,0,0, 0,0,1,1, 1,1,0,0),
  stringsAsFactors = FALSE)
acc <- ena.accumulate.data(
  units = rows[, "unit", drop = FALSE], conversation = rows[, "conv", drop = FALSE],
  metadata = rows[, c("grp", "score")], codes = rows[, c("A","B","C","D")],
  window.size.back = 2)
set <- ena.make.set(acc, dimensions = 2)

colnames(as.matrix(set$points))
#> [1] "SVD1" "SVD2" "SVD3" "SVD4" "SVD5" "SVD6"      <- all six exist

ena.correlations(set, dims = c(1, 2))     #> OK
ena.correlations(set, dims = c(1, 2, 3))  #> OK
ena.correlations(set, dims = c(1))        #> OK

ena.correlations(set, dims = c(2, 3))     #> Error in optDiff[, dim] : subscript out of bounds
ena.correlations(set, dims = c(1, 3))     #> Error in optDiff[, dim] : subscript out of bounds
ena.correlations(set, dims = c(3, 4))     #> Error in optDiff[, dim] : subscript out of bounds

Root cause

ena.correlations.R uses dims in two different coordinate systems. First to slice, producing a matrix whose columns are numbered 1..length(dims):

svdDiff = matrix(points[point1, dims] - points[point2, dims], ncol=length(dims), nrow=length(point1))
optDiff = matrix(centroids[point1, dims] - centroids[point2, dims], ncol=length(dims), nrow=length(point1))

then to index that matrix, still with the original values:

lapply(dims, function(dim) {
  cor(as.numeric(svdDiff[,dim]), as.numeric(optDiff[,dim]), method=method)
});

With dims = c(2, 3), svdDiff has two columns but column 3 is requested. It works for 1:n only because the sliced positions happen to coincide with the requested indices.

Suggested fix

Index by position after slicing:

lapply(seq_along(dims), function(i) {
  cor(as.numeric(svdDiff[,i]), as.numeric(optDiff[,i]), method=method)
});

What ena-python does

Takes dimension names (dims=["SVD1", "SVD3"]) and enumerates positionally, so any subset works. A different constraint applies there: ena-python slices points to the model's dimensions, so correlating SVD3 needs make_set(..., dimensions=3) — rENA keeps every dimension in points and never needs this.

[HUDongpin]

Reported upstream: rENA #54 (2026-07-16).

The report there pins its source references to rENA main @ a09a4b20, verified byte-identical to the released 0.3.1 this was found against — so it applies to their current tree, not just the CRAN build.

This issue stays open here as ena-python's record of the divergence; it can close when the upstream one resolves.