0: Given training data set with, , the desired quantile value, and the total number of iterations M, initialize; 1: for to M do 2: Compute the negative gradient and evaluate at:
3: Fit the negative gradient vector to by the base procedure (weak leaner, e.g., linear regression, decision stump)
4: Update the estimation by
where is a step size factor for the mth iteration. 5: end for 6: Output the obtained classifier. |