Random search hyperparameter optimisation (R parity)

Native k-fold cross-validated random search: n_iter candidate hyperparameter settings are drawn at random (caret's random-search distributions: alpha ~ U(0, 1), lambda = 2^U(-10, 3) for "glmnet"; lambda = 10^U(-5, 1) for "ridge"), each is scored by cross-validation with rmorie's own learners, and the best is returned. Folds, metrics, tie-breaking and the supported learners ("lm", "glm", "glmnet", "ridge") are those of morie_grid_search_cv(), which documents them; the RNG protocol is caret's train(search = "random"): one draw seeds the (isolated) fold assignment, then the candidates are drawn from the main stream, so a given seed reproduces caret's candidates and folds.

Usage

morie_random_search_cv(
  x,
  y,
  method = NULL,
  n_iter = 20L,
  cv = 5L,
  task = "auto",
  seed = 0L,
  deterministic_seed = NULL
)

Arguments

x

Numeric predictor matrix.

y

Response.

method

Learner id, one of "lm", "glm", "glmnet", "ridge" (default "glmnet" for classification, "ridge" for regression).

n_iter

Number of random draws.

cv

CV folds.

task

"auto" / "classification" / "regression".

seed

RNG seed.

deterministic_seed

Integer or NULL. If supplied, the RNG state is derived from the SHA-keyed morie_det_rng() so Py<->R streams agree on the canonical fixture. When NULL (default), behaviour is unchanged: seed drives set.seed() directly.

Value

Named list: estimate (CV score of the best draw), best_params, best_score (same as estimate), sampled_params (the drawn parameters, sorted, with the fold standard deviations of the metrics), sampled_scores (mean CV RMSE for regression, accuracy for classification, aligned with sampled_params), n_iter, task, n, method.

Examples

set.seed(1)
x <- matrix(rnorm(120), 40, 3)
y <- drop(x %*% c(1, -0.5, 0)) + rnorm(40)
morie_random_search_cv(x, y, n_iter = 4L, cv = 3L, task = "regression")
#> $estimate
#> [1] 0.9549318
#> 
#> $best_params
#> $best_params$lambda
#> [1] 2.813499
#> 
#> 
#> $best_score
#> [1] 0.9549318
#> 
#> $sampled_params
#>         lambda     RMSESD RsquaredSD      MAESD
#> 1 0.0001622152 0.10496248 0.05806580 0.08892095
#> 2 0.0017090053 0.10411330 0.05809362 0.08869571
#> 3 0.0273602575 0.09276388 0.05871163 0.08608925
#> 4 2.8134988370 0.19278543 0.10883596 0.21221732
#> 
#> $sampled_scores
#> [1] 0.9696466 0.9691100 0.9610609 0.9549318
#> 
#> $n_iter
#> [1] 4
#> 
#> $task
#> [1] "regression"
#> 
#> $n
#> [1] 40
#> 
#> $method
#> [1] "Random search CV (ridge)"
#>