Figure 4

Simulation study: the square error of feature weights ( w ) in models derived by the proposed approach with chosen λ 's: SE( w ) and that of resulted quantitative traits ( y ): SE( y ), comparing to the optimal (implanted) model coefficients $\left(\widehat{w}\right)$ and traits $\left(\widehat{\mathbf{\text{y}}}\right)$. SE(w) is calculated as $\left|\right|\mathbf{\text{w}}-\widehat{w}|{|}_2^2$; and SE(y) is calculated as $\left|\right|\mathbf{\text{y}}-\widehat{y}|{|}_2^2/n$, where n is the total number of subjects in 2 2 the data. Results from both settings (setting 1 - without covariates, setting 2 - with covariates) are shown.