Underfitting diagnosis (Geron Ch 4, morie.fn hmuf)

Verdict from the training error against a threshold plus the train-validation gap. The plateau slope is the OLS slope of the last max(2, floor(n/3)) points; converged compares it against 1 percent of the final training error.

Usage

morie_geron_underfitting(
  train_err,
  threshold = NULL,
  val_err = NULL,
  baseline = NULL,
  tol = 0.05
)

Arguments

train_err

Training error curve.

threshold

Bar to beat.

val_err

Optional validation curve.

baseline

Alias for threshold.

tol

Gap tolerance.

Value

List with diagnosis, underfitting, train_error, val_error, gap, threshold, baseline, plateau_slope, converged, estimate, n.

Examples

morie_geron_underfitting(c(0.9, 0.85, 0.84, 0.84), threshold = 0.2,
    val_err = c(0.95, 0.9, 0.87, 0.86))
#> $diagnosis
#> [1] "underfitting"
#> 
#> $underfitting
#> [1] TRUE
#> 
#> $train_error
#> [1] 0.84
#> 
#> $val_error
#> [1] 0.86
#> 
#> $gap
#> [1] 0.02
#> 
#> $threshold
#> [1] 0.2
#> 
#> $baseline
#> NULL
#> 
#> $plateau_slope
#> [1] 5.119863e-17
#> 
#> $converged
#> [1] TRUE
#> 
#> $estimate
#> [1] 0.84
#> 
#> $n
#> [1] 4
#> 
#> $method
#> [1] "Bias/variance verdict from the training error against a threshold plus the train-validation gap"
#>