biona001 / Knockoffs.jl

Variable Selection with Knockoffs
MIT License
6 stars 0 forks source link

Use squared Z-score for marginal feature importance statistic #56

Closed biona001 closed 1 year ago

biona001 commented 1 year ago

53 implemented fit_marginal which internally uses T[i] = |Zi| - |Zko[i]| as importance score for the ith feature. In Zihuai's ghostknockoff paper, he instead used T[i] = (Z[i])^2 - (Zko[i])^2 (see eq 15).

I should try this too to see if power is improved.