Predict using dump model for ranking task


#1

I run xgboost model with the objective of rank:ndcg on python.
I want to test manually the prediction by running on trees from dump model function.

my questions are:

  1. is the result of the model for ranking task is the sum of all each tree leaf?
  2. is dump model order the trees like the predict does? which mean if I set the ntree_limit to be 1 I can check only the first tree in the dump?
  3. is there a problem with the numeric representation using python and dumping it to json?

Thanks!


#2
  1. Yes
  2. The order between items in each query group is determined by the relative size of the prediction for the items.