Lasso-implementation-in-R

Self-implemented Lasso algorithm

This file contains a lasso implementation in R, using coordinate descent algorithm, based on the essay by Tibshirani.

• The png files: coordinate iteration formula.png and Lasso- essay by Tibshirani(Part),png describe the basis of my code
• Sor soft thresholding function.R is the code for soft thresholding function(see in coordinate iteration formula.png)
• update function.R is the code for coordinate wise update function
• Iteration Function.R for one iteration step
• coordinate descent.R is the final code

Here's a problem. When I test my code on the R-built-in data set mtcars, it works well for other coefficients(supress some to zero and pick the most significant ones) except for one variable (hp in the data set), compared with the official package glmnet; and I can't figure out why.

Besides, I test my code in some self-create data, like, x1,x2,x3 are independent variables, $x_{i}$~N(0,1), and set y=x1+x2 and train the data; but it works well only when the mean value of y is pretty close to zero. So I guess there is a fault in updating the intercept term. Hope someone can give a hint as to how to modify the code.