Exhaustive grid search with K-fold CV (Geron Ch 2, morie.fn grgs)

Cartesian product of param_grid (first name varies slowest, matching itertools.product) crossed with morie_geron_kfold_cv() splits. best_index is 0-based; std_scores uses ddof = 1.

Usage

morie_geron_grid_search_cv(
  X,
  y,
  param_grid,
  K,
  fit_score,
  shuffle = FALSE,
  seed = 0
)

Arguments

X, y

Data.

param_grid

Named list of candidate value vectors.

K

Number of folds.

fit_score

function(X_tr, y_tr, X_va, y_va, params) -> numeric, higher is better.

shuffle, seed

Passed to the fold generator.

Value

List with best_params, best_score, best_index, best_std, mean_scores, std_scores, all_scores, candidates, n_fits.

Examples

if (requireNamespace("ranger", quietly = TRUE)) {
  set.seed(5)
  X <- matrix(rnorm(60), 30, 2)
  y <- as.numeric(X[, 1] > 0)
  fs <- function(Xtr, ytr, Xte, yte, params) {
    b <- qr.solve(cbind(1, as.matrix(Xtr)), as.numeric(ytr))
    pred <- as.numeric(cbind(1, as.matrix(Xte)) %*% b)
    -mean((pred - yte)^2) * params$w
  }
  morie_geron_grid_search_cv(X, y, param_grid = list(w = c(1, 2)),
                             K = 3, fit_score = fs)
}
#> $best_params
#> $best_params$w
#> [1] 1
#> 
#> 
#> $best_score
#> [1] -0.08928899
#> 
#> $best_index
#> [1] 0
#> 
#> $best_std
#> [1] 0.04236076
#> 
#> $mean_scores
#> [1] -0.08928899 -0.17857797
#> 
#> $std_scores
#> [1] 0.04236076 0.08472153
#> 
#> $all_scores
#>            [,1]       [,2]        [,3]
#> [1,] -0.1139315 -0.1135600 -0.04037546
#> [2,] -0.2278631 -0.2271199 -0.08075092
#> 
#> $candidates
#> $candidates[[1]]
#> $candidates[[1]]$w
#> [1] 1
#> 
#> 
#> $candidates[[2]]
#> $candidates[[2]]$w
#> [1] 2
#> 
#> 
#> 
#> $n_fits
#> [1] 6
#> 
#> $estimate
#> $estimate$w
#> [1] 1
#> 
#> 
#> $n
#> [1] 30
#>