Apply a Function Over Matrix Columns with Shared Inputs

A convenience wrapper for the common "per-column apply" pattern. The matrix is shared once and each worker receives a zero-copy column view when possible.

Usage

shard_apply_matrix(
  X,
  MARGIN = 2,
  FUN,
  VARS = NULL,
  workers = NULL,
  ...,
  policy = shard_apply_policy()
)

Arguments

X

A numeric/integer/logical matrix (or a shared matrix created by share()).

MARGIN

Must be 2 (columns).

FUN

Function of the form function(v, ...) returning a scalar atomic.

VARS

Optional named list of extra variables. Large atomic VARS are auto-shared based on policy$auto_share_min_bytes.

workers

Number of workers (passed to shard_map()).

...

Additional arguments forwarded to FUN.

policy

A shard_apply_policy() object.

Value

An atomic vector of length ncol(X) with the results.

Details

Current limitation: MARGIN must be 2 (columns). Row-wise apply would require strided/gather slicing and is intentionally explicit in shard via views/kernels.

Examples

# \donttest{
X <- matrix(rnorm(400), 20, 20)
shard_apply_matrix(X, MARGIN = 2, FUN = mean)
#>  [1]  0.01543774 -0.19104418  0.12401901 -0.19082450 -0.22187576 -0.06540110
#>  [7]  0.03941378 -0.28938217  0.01727253  0.29918314 -0.49148688  0.02911144
#> [13]  0.09123634 -0.26772574  0.27751757 -0.59333675 -0.10850564 -0.20181439
#> [19] -0.07470518  0.10584459
pool_stop()
# }