Abstract:
|
We describe a way to construct hypothesis tests and confidence intervals after having used the Lasso for feature selection, allowing the regularization parameter to be chosen via an estimate of prediction error. Our estimate of prediction error is a slight variation on cross-validation. Using this variation, we are able to describe an appropriate selection event for choosing a parameter by cross-validation. Adjusting for this selection event, we derive a pivotal quantity that has an asymptotically Unifp0, 1q distribution which can be used to test hypotheses or construct intervals. To enhance power, we consider the randomized Lasso with cross-validation. We derive a similar test statistic and develop MCMC sampling scheme to construct valid post-selective confidence intervals empirically. Finally, we demonstrate via simulation that our procedure achieves high-statistical power and FDR control, yielding results comparable to knockoffs (in simulations favorable to knockoffs).
|