Based on the quadratic approximation (LQA) penalty terms and Newton's method for solving Lasso. MM algorithm, Optimal Subset Selection (BIC&AIC) and Adaptive Lasso are also simulated for comparison. This experiment is for practice purposes only, and the PDF does not contain mathematical derivations or theoretical properties (for the former, please refer to my other repository on GLM penalized likelihood, and for the latter, please refer to the articles by Fan, Li, and other teachers). For a more precise estimation of Lasso please resort to Minimum Angle Regression or Coordinate Descent Methods.
基于对惩罚项的二次逼近（LQA）和牛顿法求解Lasso，也模拟了MM算法，最优子集选择（BIC&AIC）和Adaptive Lasso以作对比。本实验仅作练习之目的，PDF中不含数学推导和理论性质证明（前者请移步本人另一个GLM惩罚似然的repo，后者请查阅Fan,Li等老师的文章），需要对Lasso更精准的估计还请诉诸最小角回归或坐标下降法。