Similar to the random forest use case (both num_parallel_tree and num_boost_round are greater than 1),
but instead of just specifying the row and col random numbers, each step it will be
- a set of trees, each trained on a fixed sub-set (data split will not change during the boosting)
- the updated boosting tree will add an ensemble of the trees for next round’s boosting (same as random forest scenario)
I wonder how I can achieve this? Is there a way to load and specify each “parallel_tree” individually?