Closed nnorton24 closed 2 weeks ago
Quantile regression is currently not supported.
It can be achieved by defining the objective function on user's side: https://github.com/dmlc/xgboost/blob/master/R-package/demo/custom_objective.R
Quantile regression with xgboost requires custom gradient and hessian functions. Here is an implementation in Python: http://www.bigdatarepublic.nl/regression-prediction-intervals-with-xgboost/
Note that implementation is not very useful for most users, since it specifies grid-searching three parameters (very costly) to get a quantile estimate. In sklearn, you define only the quantile value and you are given a very robust and reliable estimate of that quantile.
any progress ?
link above is broken http://www.bigdatarepublic.nl/regression-prediction-intervals-with-xgboost/
??
Supported by the Python package, and R package (still in master).
I know that sklearn.ensemble.GradientBoostingRegressor supports quantile regression and the production of prediction intervals. Are there any plans for the XGBoost package to offer similar support?