A Diagonal Distance Metric Learning (DDML) approach and popular regularization methods including Lasso (L1), Ridge (L2), ElasticNet, group Lasso, and fused Lasso, along with solvers including augmented Lagrangian method, penalty function method, and two versions of ADMM (feature separation and sample separation).
The motivation and application of this DML approach can be found in the paper "Feature Selection and Grouping Effect Analysis for Credit Evaluation via Regularized Diagonal Distance Metric Learning", which is still under review.