tests/tinytest/test-cv.R

In texmex: Statistical Modelling of Extreme Values

set.seed(1234)

x <- rgev(1000, xi = .2, mu = 0, sigma = 1)

g <- evm(x, family = gev)

set.seed(4321)
cg <- cv(g, range = seq(.1, 64, length.out = 25))

## Parameter estimates shrink with increasing penalty
expect_true(all(diff(cg$cv$estimate) < 0))

set.seed(4321)
cvL1 <- cv(g, range = seq(.1, 64, length.out = 25), penalty = "lasso")

## Specifying penalty = 'lasso' gives different output
expect_true(all(cg$cv$estimate != cvL1$cv$estimatae))

x <- rgev(50, xi = .2, mu = 0, sigma = 1)
g50 <- evm(x, family = gev)
g50num <-

  cg50 <- cv(g50, range = seq(.1, 64, length.out = 25), folds = 5)

expect_equal(range(cg50$data$fold), c(1, 5),
             label = "Specifying number of cv folds passes through")

cg50all <- cv(g50, range = seq(.1, 10, length.out = 25), folds = 50)
cg50all_2 <- cv(g50, range = seq(.1, 10, length.out = 25), folds = 50)


expect_equal(cg50all$cv, cg50all_2$cv,
             label = "Leave-one-out CV produces the same results from one run to another")