Control Numerical Design-MFA Calculations
Usage
design_mfa_control(
rank_tol = sqrt(.Machine$double.eps),
estimability_tol = 1e-08,
max_numeric_grid = 20L,
max_grid_rows = 5000L,
feature_chunk_size = 25000L,
multilevel_max_iter = 5000L,
multilevel_tol = 1e-08,
soft_max_iter = 500L,
soft_tol = 1e-08
)Arguments
- rank_tol
Relative eigenvalue threshold for numerical design rank and effect-component rank.
- estimability_tol
Relative tolerance for reference-grid estimability.
- max_numeric_grid
Maximum number of distinct numeric values included automatically in a reference grid. Variables with more distinct values use this many bounded sample quantiles.
- max_grid_rows
Maximum allowed reference-grid size.
- feature_chunk_size
Maximum number of response features processed at once while accumulating shared-grid sufficient statistics.
- multilevel_max_iter
Maximum block-coordinate iterations for ridge subject deviations. The default leaves headroom for weak-pooling fits, which can converge substantially more slowly than strongly pooled fits.
- multilevel_tol
Relative penalized-objective convergence tolerance.
- soft_max_iter
Maximum conjugate-gradient iterations for soft alignment.
- soft_tol
Relative normal-equation residual tolerance for soft alignment.
Examples
control <- design_mfa_control(feature_chunk_size = 1000L, soft_tol = 1e-7)
control[c("feature_chunk_size", "soft_tol")]
#> $feature_chunk_size
#> [1] 1000
#>
#> $soft_tol
#> [1] 1e-07
#>