Simulation results for different values of the smoothing parameter.

Comparison of the discriminatory power resulting from the gradient boosting approach when applying different values of the smoothing parameter . Numbers refer to to the median value and interquartile range (in parentheses) of the final on 100 simulation runs. The amount of pre-selected genes is denoted as , is the size of the training samples and cens. refers to the censoring rate. We recommend to use the value , which is also the default value of the new Cindex family for the R add-on package mboost.

本研究对比了不同平滑参数（smoothing parameter）取值下，梯度提升（Gradient Boosting）方法所获得的判别能力。数值代表100次模拟运行后最终结果的中位数，以及括号内标注的四分位距（interquartile range）。其中，预选基因的数量记为$k$，$n$为训练样本量，cens. 指代删失率（censoring rate）。我们推荐采用该参数取值，其同时也是R语言扩展包mboost中新提出的C指数族（Cindex family）的默认参数值。