regularization in Machine Learning

Evaluating the effectiveness of regularized logistic regression for the
Netflix movie rating prediction task

L1 regularization, on the other hand, results in non-zero weight for less than 1% of the features.
对于 glmnet
这个1%的数据和 lambda 有很大关系，lambda 越大，非零的系数越少.
做了一个实验，2000 features, binary classification
系数和 lambda 结果如下