Use Ridge Regression and Lasso Regression in prostate cancer data

http://web.stanford.edu/~hastie/ElemStatLearn/

• Python3.6

  • sklearn 0.18.1
  • numpy 1.17.3
  • pandas 0.20.1
  • scipy 1.2.1

This model contains regression and evaluation. You can also use your own data, just simply change the data path.

Note: The RidgeCV function is for finding Alpha, and I choose MSE, MAE and Standard Error to evaluate the model.

Result:

• Alpha is:3.0721
• Intercept is:-0.2565
• Estimated coefficients are:[('lcavol', 0.4769), ('lweight', 0.4624), ('age', -0.0117), ('lbph', 0.1007), ('svi', 0.6278), ('lcp', 0.0045), ('gleason', 0.1349), ('pgg45', 0.0033)]
• Std Error is:0.1889
• MSE is:0.3931
• MAE is:0.4443

This model is similar to Ridge, and I use coordinate gradient descent, proximal gradient and ADMM methods seperately to solve Lasso.

Note: The parameters in proximal gradient descent Lasso need to be adjusted if you want to predict other data. All in all, the rule is to make it iterate enough in a short time. You can try the other two methods before using proximal gradient and set the value as the initial value in proximal gradient descent Lasso.