Figure 1.

The performance comparison between Lasso and AGL on identification of main effects. We generate 100 replicates under each setting. Both 1000 samples and 2000 samples with balanced design are simulated. For Model 1-1, the main effect can be well approximated by the additive effect. Thus, Lasso outperforms AGL slightly. For Model 1-2, the main effect can not be approximated by additive, dominant or recessive effect. Thus, AGL outperforms Lasso significantly.