cpoDropConstants does not work as intendet

[ja-thomas]

set.seed(3)
N = 2000
d = transform(data.frame(
    x1 = rnorm(N),
    x2 = rnorm(N),
    x3 = rnorm(N)),
    y = 2*x2 + (abs(x3) < 1) + rnorm(N))

train = (1 : N) <= 1000

task = makeRegrTask(data = d, target = "y")

lrn1 = makeLearner("regr.lm")
lrn2 = cpoDropConstants() %>>% lrn1

benchmark(list(lrn1, lrn2), task)

In this case, it randomly drops up to 2 features, even though they are standard normal

*added seeding for reproducibility

[ja-thomas]

Found the bug here:

return(!(all(abs(col - cmean) < abs.tol) || all(abs(col - cmean) / cmean < rel.tol)))

should be

return(!(all(abs(col - cmean) < abs.tol) || all(abs((col - cmean) / cmean) < rel.tol)))

The first version will always drop features that have a negative mean

[mb706]

Thanks!