Suppose \(\tau\) is a training set with \(n\) elements and \(\tau^{*}\), also of size \(n\), is obtained from \(\tau\) by bootstrapping; that is, resampling with replacement. Show that for large \(n, \tau^{*}\) does not contain a fraction of about \(\mathrm{e}^{-1} \approx 0.37\) of the points from \(\tau\).