Bug in mstop for non-cyclical fitting

[hofnerb]

require("gamboostLSS")

###negbin dist, linear###

set.seed(2611)
x1 <- rnorm(1000)
x2 <- rnorm(1000)
x3 <- rnorm(1000)
x4 <- rnorm(1000)
x5 <- rnorm(1000)
x6 <- rnorm(1000)
mu    <- exp(1.5 + x1^2 +0.5 * x2 - 3 * sin(x3) -1 * x4)
sigma <- exp(-0.2 * x4 +0.2 * x5 +0.4 * x6)
y <- numeric(1000)
for (i in 1:1000)
  y[i] <- rnbinom(1, size = sigma[i], mu = mu[i])
dat <- data.frame(x1, x2, x3, x4, x5, x6, y)

model <- glmboostLSS(y ~ ., families = NBinomialLSS(), data = dat,
                     control = boost_control(mstop = 3), method = "inner")
cvr2 <- cvrisk(model, grid = 1:100)

mstop(cvr2)  #  a scalar
mstop(model) <- mstop(cvr2) #works
## but
mstop(model) # a vector
## and
mstop(model) <- c(mu = 10, sigma = 20) # breaks
mstop(model) ## now mstop = 10...

Does the current behavior makes sense? I don't think so.

I do see why it is interesting to know how many iterations were fitted for mu and sigma but at the same time, we cannot reuse this information. We only care about the total number of steps.

• [ ] So it would preferable to primarily show the single mstop value.
• [ ] Furthermore, we need to check all documentation where we state that mstop can be assigned as a named vector etc. That must be consistent and correct.

[ja-thomas]

I see your point but:

1) in most cases it is more interesting to see the distribution of the iterations that the overall iterations (which is specified by the user).

2) this would break a lot of my code since I use the mstop() function a lot in the internal fitting process.

What we could do is if mstop(model) <- c(mu = 10, sigma = 20) is called we use sum(c(mu = 10, sigma = 20)) as new mstop and give a warning to the user?

[hofnerb]

I see. Let's discuss this at the phone. That is easier.

[hofnerb]

return scalar with one attribute for individual steps per distribution parameter. perhaps add nice print function...