tdhock
commented
5 years ago note that t ~ weibull is positive valued (data and parameters) whereas y = exp(t) ~ extreme_value is real valued. (potentially negative)
@theadityasam please try to avoid duplication in the code for cv.iregnet. please use transformed.distributions list when computing validation log likelihood in cv.iregnet.
seems to me that the best way to do that would be via three base distributions for real-valued numbers, normal, logistic, extreme value. all the other distributions should be computable in terms of these.
extreme value distribution/density functions do not seem to exist in base R. so to implement them either (1) use dweibull/pweibull if that is numerically stable enough, (2) otherwise try one of the packages that explicitly implements it.
look at formulas in http://members.cbio.mines-paristech.fr/~thocking/survival.pdf