LIBLINEAR训练L1正则化逻辑回归采用何种优化方法？

Hey there, great question! I can confirm that LIBLINEAR does rely on coordinate descent methods for its core optimization tasks. Let me point you to where you can verify this in the resources you referenced:

• The Official LIBLINEAR Paper: If you flip to Section 3 (covering solutions for L2-regularized L1-loss SVM and L2-regularized L2-loss SVM), you’ll find explicit descriptions of coordinate descent being used to solve the dual optimization problems. The paper even walks through the step-by-step iterative updates for each coordinate—this is the key telltale of coordinate descent.
• The GitHub Repository: Dig into the source code (check out linear.cpp for starters) and you’ll spot the coordinate descent logic in action. Look for loops that iterate over individual variables, updating each one while holding the rest fixed—that’s exactly how coordinate descent works.

For your convenience, here are the key LIBLINEAR resources you mentioned:

• Official Website
• Research Paper
• GitHub Repository

Pro tip: It’s easy to gloss over the technical optimization sections in the paper, so focusing specifically on the dual problem solving parts will help you find this confirmation quickly.

内容的提问来源于stack exchange，提问作者zyxue