@rui__carvalho I thought every statistician knew that L1 is typically only good for variable screening since the work of Peter Buhlmann & Berkeley colleagues wrote a bunch of papers on the terrible variable selection properties of the Lasso. https://t.
5,809 followers
394 followers
@economeager @jt_kerwin See e.g. Prop. 1 of https://t.co/3IVUQXiN8u, or Section 2.5.1 of https://t.co/I3VlK14QjX for a textbook reference. Heuristics exist to compensate for this; e.g. in the R glmnet package, the default is to choose the largest penalty w
17 followers
Chapter 6:Lassoの理論で度々登場する「oracle」が、本当に天下りすぎてoracle(神託)な感じ。 https://t.co/DTn2BP6SaF
17 followers
本読み始めました(Lassoの理論周り中心に)。 https://t.co/DTn2BP6SaF 2Q〜3Qくらいで読めたら良いな。 登竜門の鯉の稚魚にはなれた気でいます。