Algorithms for manipulation of level sets of nonparametric density estimates (Q2488405)

The author considers functions \(f:R^d\to(0,+\infty)\) defined by some evaluation tree and describes an algorithm for finding a level set tree for such a function. I.e., for a set of levels \(\lambda\) the level sets \(\Lambda_\lambda=\{x\in R^d\;: f(x)\geq\lambda\}\) are decomposed to separated (non-connected) components. A level set tree is a tree structure formed by taking as root nodes the maximally separated regions of the support of \(f\). The child nodes of a given node consist of the maximally separated regions of the level set corresponding to the one step higher level than the level of this parent node. This technique is applied to the kernel, CART and average shifted histogram estimates of multivariate densities.