Overfitting diagnostic (Geron Ch 4, morie.fn hmovf)

Generalisation gap E_val - E_train with the early-stopping point. best_epoch is 0-based, as in Python.

Usage

morie_geron_overfitting(train_err, val_err, tol = 0)

Arguments

train_err, val_err

Equal-length non-negative error curves.

tol

Non-negative gap tolerance.

Value

List with gap, gaps, ratio, overfitting, best_epoch, best_val, epochs_past_best, verdict, estimate, n.

Examples

V <- c(1, 2, 3, 4, 5, 6, 7, 8)
morie_geron_overfitting(V, V)
#> $gap
#> [1] 0
#> 
#> $gaps
#> [1] 0 0 0 0 0 0 0 0
#> 
#> $ratio
#> [1] 1
#> 
#> $overfitting
#> [1] FALSE
#> 
#> $best_epoch
#> [1] 0
#> 
#> $best_val
#> [1] 1
#> 
#> $epochs_past_best
#> [1] 7
#> 
#> $verdict
#> [1] "no overfitting gap: errors agree, so look at their LEVEL for underfitting"
#> 
#> $estimate
#> [1] 0
#> 
#> $n
#> [1] 8
#> 
#> $method
#> [1] "Generalisation gap E_val - E_train with early-stopping point"
#>