Regression Demo about xgboost HOT 5 CLOSED

OK, this is a bit confusing. There are two factors in most machine learning algorithm. (1) model, which means how you make predictions score(trees, linear combination of features y=w^Tx (2) objective function, how do you train the prediction score, this include square loss, logistic loss, rank objective.

So reg:linear actually stands for square loss, reg:logistic for logistic loss.If you use reg:linear and gbtree booster, this is GBTree with square loss. By default, it is gradient boosted tree.

from xgboost.

BTW, an unrelated question:) I am creating a list of resources of higgs solutions: https://github.com/tqchen/xgboost/tree/master/demo/kaggle-higgs

If you like to share your solution, send a pull req or just ping me and I would love to put that in the list

from xgboost.

My code is a bit messy now, i will clean it and send it to you soon, its score can be improved, i used 5 models by averaging (with equal weights), maybe stacking 5 models' prediction with xgboost may improve the score :)

from xgboost.

i have sent it to your academic email address

from xgboost.

Seems the original question has been answered by my post

from xgboost.