Testing hasn't demonstrated any significant differences between code that used lasso + lsei (lasso for feature selection and lsei for actual fitting the distribution) and code that relied exclusively for lasso. Since lasso supports sparse matrices and hence is much faster and more memory-efficient than lsei, it suggest dropping lsei from the current code.
Testing hasn't demonstrated any significant differences between code that used lasso + lsei (lasso for feature selection and lsei for actual fitting the distribution) and code that relied exclusively for lasso. Since lasso supports sparse matrices and hence is much faster and more memory-efficient than lsei, it suggest dropping lsei from the current code.