GAN mode coverage / collapse (Geron Ch 17, morie.fn grmcol)

A sample counts for its nearest true mode when within tol; the default tol is half the smallest inter-mode distance. Mode indices are 0-based.

Usage

morie_geron_gan_mode_collapse_metric(samples, true_modes, tol = NULL)

Arguments

samples

(m, d) generated samples.

true_modes

(K, d) mode centres.

tol

Optional positive radius.

Value

List with coverage, mode_collapse_rate, modes_hit, modes_missed, samples_per_mode, n_off_distribution, tol.

Examples

M <- matrix(c(1, 2, 3, 4, 5, 6), nrow = 2)
morie_geron_gan_mode_collapse_metric(M, M)
#> $coverage
#> [1] 1
#> 
#> $mode_collapse_rate
#> [1] 0
#> 
#> $modes_hit
#> [1] 0 1
#> 
#> $modes_missed
#> integer(0)
#> 
#> $samples_per_mode
#> [1] 1 1
#> 
#> $n_off_distribution
#> [1] 0
#> 
#> $tol
#> [1] 0.8660254
#> 
#> $estimate
#> [1] 1
#> 
#> $n
#> [1] 2
#>