16、基因选择方法：从微阵列数据到时间序列数据的探索

基因选择方法：从微阵列数据到时间序列数据的探索

1. 集成逻辑回归特征选择算法

在处理高维数据的特征选择问题时，集成逻辑回归（Ensemble Logistic Regression，ELR）是一种有效的方法。下面是ELR算法的详细介绍：

Algorithm 1. Ensemble Logistic Regression with Feature Resampling
Algorithm elr
Input: A learning sample X ∈Rn×p and class labels y ∈{−1, 1}n
Input: A regularization parameter λ for estimating a l2-LR model (3)
Output: A vector prob ∈[0, 1]p of feature relevance
Initialize prob according to a t-test ranking
BCR ←0.5
// Default initialization of the average BCR
repeat
    Randomly split X into training (80%) and validation (20%)
    Draw n out of p features at random according to prob
    (w, v) ←a l2-LR model M learned on training restricted to n features
    Compute BCR of M on validation
    quality