The key problem of gradient boosting is that you must maintain N decision trees to sum over, so rather than keeping only a single tree's root node in a data structure, multiple trees must remembered. The trees are summed in each inference,…