"Advantages of LDA over logistic regression", it was mentioned that logistic regression may be unstable in certain settings. Consider the following example: In binary logistic regression, suppose that p>n,X has full row rank (rank=n), and there is no intercept. Show that the likelihood function is not maximized by any finite ; in fact we can let diverge while the likelihood tends to 1 . Hint: pick a so that there is no training error in classification, and then set t with the scalar t