They just wrapped around numpy.seterr, which users should call directly instead.
This also changes the default behaviour from setting numpy.seterr(all="ignore") to leaving all at its numpy default.