Currently, kliep() and lhss() are not parallelized, because then all kernel matrices need to be distributed. For kliep(), this is often more expensive than the computations themselves. For lhss(), many objects are created that need to be held separately, which complicates parallel computation substantively, and initially also reduced the speedup tremendously.
A simple solution might be to simple run the estimation function in parallel using pbapply, which is already used in the summary function. An alternative can be to let users take care of this themselves, and provide a bind function that stitches objects of the same class together.
Currently,
kliep()andlhss()are not parallelized, because then all kernel matrices need to be distributed. Forkliep(), this is often more expensive than the computations themselves. Forlhss(), many objects are created that need to be held separately, which complicates parallel computation substantively, and initially also reduced the speedup tremendously.A simple solution might be to simple run the estimation function in parallel using
pbapply, which is already used in the summary function. An alternative can be to let users take care of this themselves, and provide a bind function that stitches objects of the same class together.