[BenjaminDEMAILLE]

SketchData() fails on any small object, at every size of sketch:

> SketchData(pbmc_small, ncells = 40, method = "LeverageScore")
Error: max(nu, nv) must be strictly less than min(nrow(A), ncol(A))

Cause

LeverageScore.default() sizes its decomposition to the object, and then does not use the value:

nv <- ifelse(nrow(x = object) < 50, nrow(x = object) - 1, 50)
...
Z <- irlba(A = object, nv = 50, nu = 0, verbose = FALSE)$v

nv is computed and dropped; 50 components are always requested. Anything with fewer than 50 features or cells on the side being decomposed therefore stops inside irlba, and since SketchData() is how LeverageScore() is usually reached, that is where it surfaces.

Fix

Use nv, bounded by both dimensions, and say so when there is nothing to decompose at all:

Cannot compute leverage scores from 1 features and 1 cells: at least two of each are needed

Behaviour is unchanged for objects with 50 or more of each, which is where the constant and the computed value agree.

Tests

tests/testthat/test_leverage_score_small.R: leverage scores on pbmc_small, sketching to 40 cells, asking for more cells than exist (all of them are kept), and the message for a matrix with nothing to decompose.

Four assertions fail without the change. Full suite NOT_CRAN=true: 1187 passing, the only 2 failures being the pre-existing Read10X_Image ones fixed in #10454.

🤖 Generated with Claude Code

[BenjaminDEMAILLE]

Superseded by #10492, which groups five small independent fixes into one review. The commit is carried over unchanged; closing here to keep the queue down rather than because anything was wrong with it.

[anashen]

Thanks for the catch!