Question about predictions about xgboost HOT 4 CLOSED

Hi ivihrov:
The initial value is not dumped into the dump for now. There could be two behaviors of initial value:
(1) in the basic setting, initial value(initial prediction) is set via base_score, the default is 0.5. In logistic regression, this means initial probability, and corresponding initial margin is 0. In linear regression, however, this means initial margin is 0.5

(2) in the case when you provide base_margin via https://github.com/tqchen/xgboost/blob/master/R-package/demo/boost_from_prediction.R the initial margin is what you specified via the array
Tianqi

from xgboost.

Thank you!

from xgboost.

a correction, default value of base_score is 0.5, but it can be set to other values. For linear reg, base_score is your initial margin, for logistic regression, it will be inverse logistic-transformed to initial margin

from xgboost.

Hi Tianqi, I really appreciate your amazing work on XGBoost! I am still not clear about the meaning of the base_score and margin. How does it affect the training process?

from xgboost.