%X% with base-learners that have design matrices with different row numbers

[sbrockhaus]

When two base-learners are connected by %X% and one of them has shorter data, the design matrix is simply repeated. See

library(mboost)

mydat <- list(x_10 = 1:10, x_20 = 1:20, y = rnorm(20))

## works without throwing an error / warning, ... 
m1 <- mboost(y ~ bols(x_10) %X% bols(x_20), data = mydat)

## ... as x_10 is simply repeated 
extract(m1)

## see get("%X%")

I was surprised that this does not throw an error or a warning.

This is due to
mf <- cbind(model.frame(bl1), model.frame(bl2))
in
https://github.com/boost-R/mboost/blob/master/R/bl.R#L1043

I would suggest to throw a warning, something like

if (nrow(model.frame(bl1)) != nrow(model.frame(bl2))) 
  warning("The design matrices of the two marginal base-learners imply a different number of rows: ", 
                   nrow(model.frame(bl1)), ", ", nrow(model.frame(bl2)))

But maybe this behavior is intended?

@davidruegamer do you use %X% with differing row numbers?

[sbrockhaus]

btw. bols(x, by = z) behaves in the same way.

mydat <- list(x_10 = 1:10, x_20 = 1:20, y = rnorm(20))

## works without throwing an error / warning, ... 
m2 <- mboost(y ~ bols(x_20, by = x_10), data = mydat)

## ... as x_10 is simply repeated 
extract(m2)

[hofnerb]

Other base-learners and situations showed the same behavior. See tests/testthat/test-baselearner.R for a detailed list of cases.