Greedy Function Approximation:A Gradient Boosting Machine

 

https://statweb.stanford.edu/~jhf/ftp/trebst.pdf

 

 

page10

90% to 95% of the observations were often deleted without sacrificing accuracy of the
estimates,using either influence measure.

 

 

 

 

 

 

【解释regularization】

page12
5 Regularization
In prediction problems,fitting the training data too closely can be counterproductive.