I noticed that pbz.control lets the user choose a different objective criterion for the smooth than the default method = "ML". However, in my experiments I couldn't get anything other than that to work. When I try I get an error which seems to be buried deep in the code:
Error in regpen(y = y, X = X, w = w, lambda = lambda) :
unused argument (lambda = lambda)
gamlss(formula = y ~ pbz(x), sigma.formula = ~ pbz(x), nu.formula = ~ pbz(x, control = pbz.control(method = "ML")), family = BCPEo(), control = gamlss.control(nu.step = 0.01), method = mixed(10, 20), data = dset)
However, this triggers the error:
gamlss(formula = y ~ pbz(x), sigma.formula = ~ pbz(x), nu.formula = ~ pbz(x, control = pbz.control(method = "GAIC")), family = BCPEo(), control = gamlss.control(nu.step = 0.01), method = mixed(10, 20), data = dset)
I was interested in "GAIC" for parsimoniously describing the quantiles of the data set. Note that there is a stark drop in the response, which has trouble with the default optimization settings.
I noticed that
pbz.control
lets the user choose a different objective criterion for the smooth than the defaultmethod = "ML"
. However, in my experiments I couldn't get anything other than that to work. When I try I get an error which seems to be buried deep in the code:Here is an example
dset
:Here is a minimal example. This works:
However, this triggers the error:
I was interested in "GAIC" for parsimoniously describing the quantiles of the data set. Note that there is a stark drop in the response, which has trouble with the default optimization settings.