Skip to contents

Hyperparameter tuning by grid or random search (Geron Ch 1, morie.fn hmhpt)

Source: R/geron_w4b_native.R
morie_geron_hyperparameter_tuning.Rd

The grid branch is exact and deterministic (delegates to the existing morie_geron_cross_validation_score core). The random branch's candidate draws use the shared LCG stream, not numpy's PCG64 – non-portable.

Usage

morie_geron_hyperparameter_tuning(
  param_grid,
  X,
  y,
  estimator = NULL,
  search = "grid",
  n_iter = 10,
  K = 3,
  seed = 0,
  score = NULL
)

Arguments

param_grid

Named list of candidate value vectors.

X, y

Data.

estimator

Optional function(Xtr,ytr,...) -> function(Xnew) predict.

"grid" or "random".

n_iter

Random draws.

K

Folds.

seed

Seed.

score

Optional scorer.

Value

List with best_params, best_score, results, n_candidates, n_fits, distinct_values_per_param, estimate, n, method.

Examples

X <- matrix(c(1, 2, 3, 4), ncol = 1)
y <- c(3, 6, 9, 12)
morie_geron_hyperparameter_tuning(list(alpha = c(0, 1, 100)),
    X, y, K = 2)
#> $best_params
#> $best_params$alpha
#> [1] 0
#> 
#> 
#> $best_score
#> [1] 1
#> 
#> $results
#> $results[[1]]
#> $results[[1]]$params
#> $results[[1]]$params$alpha
#> [1] 0
#> 
#> 
#> $results[[1]]$cv_score
#> [1] 1
#> 
#> $results[[1]]$fold_scores
#> [1] 1 1
#> 
#> 
#> $results[[2]]
#> $results[[2]]$params
#> $results[[2]]$params$alpha
#> [1] 1
#> 
#> 
#> $results[[2]]$cv_score
#> [1] 0.2981591
#> 
#> $results[[2]]$fold_scores
#> [1]  0.9852071 -0.3888889
#> 
#> 
#> $results[[3]]
#> $results[[3]]$params
#> $results[[3]]$params$alpha
#> [1] 100
#> 
#> 
#> $results[[3]]$cv_score
#> [1] -24.87574
#> 
#> $results[[3]]$fold_scores
#> [1]  -5.40000 -44.35147
#> 
#> 
#> 
#> $n_candidates
#> [1] 3
#> 
#> $n_fits
#> [1] 6
#> 
#> $estimate
#> [1] 1
#> 
#> $n
#> [1] 4
#> 
#> $method
#> [1] "Hyperparameter tuning by grid or random search"
#> 
#> $distinct_values_per_param
#> $distinct_values_per_param$alpha
#> [1] 3
#> 
#> 
#> $search
#> [1] "grid"
#>