Mode Trees for Multivariate Data

We consider visualizing scales of multivariate density estimates with the help of mode trees. Mode trees visualize the locations of the modes of density estimates, when the smoothing parameter of the estimates ranges over an interval. We define multiframe mode graphs which generalize classical mode trees to the multivariate setting. We give examples of the application of multiframe mode graphs with kernel estimates and with multivariate adaptive histograms.     

Visualization of Multivariate Density Estimates With Level Set Trees

This article presents a method for visualization of multivariate functions. The method is based on a tree structure—called the level set tree—built from separated parts of level sets of a function. The method is applied for visualization of estimates of multivarate density functions. With different graphical representations of level set trees we may visualize the number and location of modes, excess masses associated with the modes, and certain shape characteristics of the estimate. Simulation examples are presented where projecting data to two dimensions does not help to reveal the modes of the density, but with the help of level set trees one may detect the modes. I argue that level set trees provide a useful method for exploratory data analysis.     

Estimation of level set trees using adaptive partitions

We present methods for the estimation of level sets, a level set tree, and a volume function of a multivariate density function. The methods are such that the computation is feasible and estimation is statistically efficient in moderate dimensional cases (\(d\approx 8\)) and for moderate sample sizes (\(n\approx \) 50,000). We apply kernel estimation together with an adaptive partition of the sample space. We illustrate how level set trees can be applied in cluster analysis and in flow cytometry.     

The spanning trees forced by the path and the star

A set S of trees of order n forces a tree T if every graph having each tree in S as a spanning tree must also have T as a spanning tree. A spanning tree forcing set for order n that forces every tree of order n. A spanning-tree forcing set S is a test set for panarboreal graphs, since a graph of order n is panarboreal if and only if it has all of the trees in S as spanning trees. For each positive integer n ≠ 1, the star belongs to every spanning tree forcing set for order n. The main results of this paper are a proof that the path belongs to every spanning-tree forcing set for each order n ∉ {1, 6, 7, 8} and a computationally tractable characterization of the trees of order n ≥ 15 forced by the path and the star. Corollaries of those results include a construction of many trees that do not belong to any minimal spanning tree forcing set for orders n ≥ 15 and a proof that the following related decision problem is NP-complete: an instance is a pair (G, T) consisting of a graph G of order n and maximum degree n - 1 with a hamiltonian path, and a tree T of order n; the problem is to determine whether T is a spanning tree of G. © 1996 John Wiley & Sons, Inc.     

Hypergraphs as a mean of discovering the dependence structure of a discrete multivariate probability distribution

Most everyday reasoning and decision making is based on uncertain premises. The premises or attributes, which we must take into consideration, are random variables, therefore we often have to deal with a high dimensional multivariate random vector. A multivariate random vector can be represented graphically as a Markov network. Usually the structure of the Markov network is unknown. In this paper we construct special type of junction trees, in order to obtain good approximations of the real probability distribution. These junction trees are capable of revealing some of the conditional independences of the network. We have already introduced the concept of the t-cherry junction tree (E. Kovács and T. Szántai in Proceedings of the IFIP/IIASA//GAMM Workshop on Coping with Uncertainty, 2010), based on the t-cherry tree graph structure. This approximation uses only two and three dimensional marginal probability distributions. Now we use k-th order t-cherry trees, also called simplex multitrees to introduce the concept of the k-th order t-cherry junction tree. We prove that the k-th order t-cherry junction tree gives the best approximation among the family of k-width junction trees. Then we give a method which starting from a k-th order t-cherry junction tree constructs a (k+1)-th order t-cherry junction tree which gives at least as good approximation. In the last part we present some numerical results and some possible applications.     

The Polya Tree Sampler: Towards Efficient and Automatic Independent Metropolis-Hastings Proposals

We present a simple, efficient, and computationally cheap sampling method for exploring an un-normalized multivariate density on ?(d), such as a posterior density, called the Polya tree sampler. The algorithm constructs an independent proposal based on an approximation of the target density. The approximation is built from a set of (initial) support points - data that act as parameters for the approximation - and the predictive density of a finite multivariate Polya tree. In an initial "warming-up" phase, the support points are iteratively relocated to regions of higher support under the target distribution to minimize the distance between the target distribution and the Polya tree predictive distribution. In the "sampling" phase, samples from the final approximating mixture of finite Polya trees are used as candidates which are accepted with a standard Metropolis-Hastings acceptance probability. Several illustrations are presented, including comparisons of the proposed approach to Metropolis-within-Gibbs and delayed rejection adaptive Metropolis algorithm.     

Meta densities and the shape of their sample clouds

This paper compares the shape of the level sets for two multivariate densities. The densities are positive and continuous, and have the same dependence structure. The density f is heavy-tailed. It decreases at the same rate-up to a positive constant-along all rays. The level sets {f>c} for c↓0, have a limit shape, a bounded convex set. We transform each of the coordinates to obtain a new density g with Gaussian marginals. We shall also consider densities g with Laplace, or symmetric Weibull marginal densities. It will be shown that the level sets of the new light-tailed density g also have a limit shape, a bounded star-shaped set. The boundary of this set may be written down explicitly as the solution of a simple equation depending on two positive parameters. The limit shape is of interest in the study of extremes and in risk theory, since it determines how the extreme observations in different directions relate. Although the densities f and g have the same copula-by construction-the shapes of the level sets are not related. Knowledge of the limit shape of the level sets for one density gives no information about the limit shape for the other density.     

The decomposition of trees into subtrees

A necessary condition for the decomposition of a tree T into subtrees, each isomorphic to a tree from a given set of trees is presented. We also present a characterization of the set of trees for which the condition is sufficient. Many examples are given.     

Probabilistic and fractal aspects of Lévy trees

We investigate the random continuous trees called Lévy trees, which are obtained as scaling limits of discrete Galton-Watson trees. We give a mathematically precise definition of these random trees as random variables taking values in the set of equivalence classes of compact rooted -trees, which is equipped with the Gromov-Hausdorff distance. To construct Lévy trees, we make use of the coding by the height process which was studied in detail in previous work. We then investigate various probabilistic properties of Lévy trees. In particular we establish a branching property analogous to the well-known property for Galton-Watson trees: Conditionally given the tree below level a, the subtrees originating from that level are distributed as the atoms of a Poisson point measure whose intensity involves a local time measure supported on the vertices at distance a from the root. We study regularity properties of local times in the space variable, and prove that the support of local time is the full level set, except for certain exceptional values of a corresponding to local extinctions. We also compute several fractal dimensions of Lévy trees, including Hausdorff and packing dimensions, in terms of lower and upper indices for the branching mechanism function which characterizes the distribution of the tree. We finally discuss some applications to super-Brownian motion with a general branching mechanism.     

Integral trees of arbitrarily large diameters

In this paper, we construct trees having only integer eigenvalues with arbitrarily large diameters. In fact, we prove that for every finite set S of positive integers there exists a tree whose positive eigenvalues are exactly the elements of S. If the set S is different from the set {1} then the constructed tree will have diameter 2|S|.     

Approximating the Minimum-Degree Steiner Tree to within One of Optimal

The problem of constructing a spanning tree for a graph G = (V, E) with n vertices whose maximal degree is the smallest among all spanning trees of G is considered. This problem is easily shown to be NP-hard. In the Steiner version of this problem, along with the input graph, a set of distinguished vertices D V is given. A minimum-degree Steiner tree is a tree of minimum degree which spans at least the set D. Iterative polynomial time approximation algorithms for the problems are given. The algorithms compute trees whose maximal degree is at most Δ* + 1, where Δ* is the degree of some optimal tree for the respective problems. Unless P = NP, this is the best bound achievable in polynomial time.     

On Rotations and the Generation of Binary Trees

The rotation graph, G_n, has vertex set consisting of all binary trees with n nodes. Two vertices are connected by an edge if a single rotation will transform one tree into the other. We provide a simpler proof of a result of Lucas that G_n, contains a Hamilton path. Our proof deals directly with the pointer representation of the binary tree. This proof provides the basis of an algorithm for generating all binary trees that can be implemented to run on a pointer machine and to use only constant time between the output of successive trees. Ranking and unranking algorithms are developed for the ordering of binary trees implied by the generation algorithm. These algorithms have time complexity O(n²) (arithmetic operations). We also show strong relationships amongst various representations of binary trees and amongst binary tree generation algorithms that have recently appeared in the literature.     

The number of maximum matchings in a tree

We determine upper and lower bounds for the number of maximum matchings (i.e., matchings of maximum cardinality) m(T) of a tree T of given order. While the trees that attain the lower bound are easily characterised, the trees with the largest number of maximum matchings show a very subtle structure. We give a complete characterisation of these trees and derive that the number of maximum matchings in a tree of order n is at most O(1.391664ⁿ) (the precise constant being an algebraic number of degree 14). As a corollary, we improve on a recent result by Górska and Skupień on the number of maximal matchings (maximal with respect to set inclusion).     

On the distribution of binary search trees under the random permutation model

We study the distribution Q on the set B_n of binary search trees over a linearly ordered set of n records under the standard random permutation model. This distribution also arises as the stationary distribution for the move-to-root (MTR) Markov chain taking values in B_n when successive requests are independent and identically distributed with each record equally likely. We identify the minimum and maximum values of the functional Q and the trees achieving those values and argue that Q is a crude measure of the “shape” of the tree. We study the distribution of Q(T) for two choices of distribution for random trees T; uniform over B_n and Q. In the latter case, we obtain a limiting normal distribution for −ln Q(T). © 1996 John Wiley & Sons, Inc.     

The rotation graph of binary trees is Hamiltonian

In this paper we study the rotation transformation on binary trees and consider the properties of binary trees under this operation. The rotation is the universal primitive used to rebalance dynamic binary search trees. New binary search tree algorithms have recently been introduced by Sleator and Tarjan. It has been conjectured that these algorithms are as efficient as any algorithm that dynamically restructures the tree using rotations. We hope that by studying rotations in binary trees we shall gain a better understanding of the nature of binary search trees, which in turn will lead to a proof of this “dynamic optimality conjecture”. We define a graph, RG(n), whose vertex set consists of all binary trees containing n nodes, and which has an edge between two trees if they differ by only one rotation. We shall introduce a new characterization of the structure of RG(n) and use it to demonstrate the existence of a Hamiltonian cycle in the graph. The proof is constructive and can be used to enumerate all binary trees with n nodes in constant time per tree.     

The shape of large Galton‐Watson trees with possibly infinite variance

Let T be a critical or subcritical Galton‐Watson family tree with possibly infinite variance. We are interested in the shape of T conditioned to have a large total number of vertices. For this purpose we study random trees whose conditional distribution given their size is the same as the respective conditional distribution of T. These random family trees have a simple probabilistic structure if decomposed along the lines of descent of a number of distinguished vertices chosen uniformly at random. The shape of the subtrees spanned by the selected vertices and the root depends essentially on the tail of the offspring distribution: While in the finite variance case the subtrees are asymptotically binary, other shapes do persist in the limit if the variance is infinite. In fact, we show that these subtrees are Galton‐Watson trees conditioned on their total number of leaves. The rescaled total size of the trees is shown to have a gamma limit law. © 2004 Wiley Periodicals, Inc. Random Struct. Alg., 2004     

Noncrossing trees are almost conditioned Galton–Watson trees

A noncrossing tree (NC‐tree) is a tree drawn on the plane having as vertices a set of points on the boundary of a circle, and whose edges are straight line segments that do not cross. In this article, we show that NC‐trees with size n are conditioned Galton–Watson trees. As corollaries, we give the limit law of depth‐first traversal processes and the limit profile of NC‐trees. © 2002 John Wiley & Sons, Inc. Random Struct. Alg., 20, 115–125, 2002     

On the reconstruction of locally finite trees

We prove a theorem saying, when taken together with previous results of Bondy, Hemminger, and Thomassen, that every locally finite, infinite tree not containing a subdivision of the dyadic tree (i. e., the regular tree of degree 3) is uniquely determined, up to isomorphism, from its collection of vertex-deleted subgraphs. Furthermore, as another partial result concerning the reconstruction of locally finite trees, we show that the same is true for locally finite trees whose set of vertices of degree s is nonempty and finite (for some positive integer s).     

On spanning tree problems with multiple objectives

We investigate two versions of multiple objective minimum spanning tree problems defined on a network with vectorial weights. First, we want to minimize the maximum ofQ linear objective functions taken over the set of all spanning trees (max-linear spanning tree problem, ML-ST). Secondly, we look for efficient spanning trees (multi-criteria spanning tree problem, MC-ST).Problem ML-ST is shown to be NP-complete. An exact algorithm which is based on ranking is presented. The procedure can also be used as an approximation scheme. For solving the bicriterion MC-ST, which in the worst case may have an exponential number of efficient trees, a two-phase procedure is presented. Based on the computation of extremal efficient spanning trees we use neighbourhood search to determine a sequence of solutions with the property that the distance between two consecutive solutions is less than a given accuracy.Partially supported by Deutsche Forschungsgemeinschaft and HCº Contract no. ERBCHRXCT 930087.Partially supported by Alexander von Humboldt-Stiftung.