Package 'gromovlab'

Gromov-Hausdorff-type distances of labelled metric spaces

Description

The function gromovdist calculates the matched Gromov-$\ell^p$ distances of two metrics on a finite space $X$:

$D_p(\rho_1,\rho_2)=\inf\{\|(|d(\phi_1(x),\phi_2(x))|)_{x \in X}\|_p\}.$

There, the infimum is taken over all isometric embeddings $\phi_1$ of $(X,\rho_1)$, $\phi_2$ of $(X,\rho_2)$ into a common metric space $(Y,d)$. Only $1\le p\le \infty$ is considered.

At the basis is the reformulation of the metric as value of a convex program, see Liebscher (2015).

Methods for various classes are provided:

• dist,dissimilarity which are distance matrices.

• matrix for matrices containing the individual distances of the elements of $X$.

• igraph for connected graphs. The metric on the nodes or just the leaves (nodes of degree 1) of the graph is the length of the shortest path.

• phylo for phylogenetic trees. Again the metric is induced by the graph.

Usage

gromovdist(d1,d2=NULL,type="l1",p=NULL,...)

## S3 method for class 'list'
gromovdist(d1,d2=NULL,type="l1",p=NULL,...)
## S3 method for class 'multiPhylo'
gromovdist(d1,d2=NULL,type="l1",p=NULL,...)
## S3 method for class 'phylo'
gromovdist(d1,d2=NULL,type="l1",p=NULL,...)
## S3 method for class 'dist'
gromovdist(d1,d2=NULL,type="l1",p=NULL,...)
## S3 method for class 'dissimilarity'
gromovdist(d1,d2=NULL,type="l1",p=NULL,...)
## S3 method for class 'matrix'
gromovdist(d1,d2=NULL,type="l1",p=NULL,...)
## S3 method for class 'igraph'
gromovdist(d1,d2=NULL,type="l1",p=NULL,leavesonly=TRUE,...)

Arguments

type	type of metric to use
d1, d2	distance object(s).
p	if type="lp" the value of $p$. If not supplied, $p=2$ is used.
leavesonly	compute the distances between the leaves of the graph/tree only?
...	further parameters

Details

type="l1" yields $p=1$

type="l2" yields $p=2$

type="linfinity" yields $p=\infty$

type="lp" is for (not so efficient) computation using constrOptim for arbitrary $1\le p<\infty$

If d1 is a list, the distance matrix between all elements of the list is computed. It is represented as an object of class dissimiliarity, see dissimilarity.object.

The distance is only computed for that part of the objects where the labels are present in both objects. If there are no labels the elements are numbered consecutively.

Value

The distance (one numeric) or a distance matrix for the list and multiPhylo methods

Author(s)

Volkmar Liebscher

References

V.Liebscher, Gromov meets Phylogenetics - new Animals for the Zoo of Metrics on Tree Space. preprint 2015 arXiv:1504.05795

See Also

dist, dissimilarity.object,phangorn-package, igraph-package,constrOptim.

Examples

library("ape")
tr1<-rtree(n=10)
tr2<-rtree(n=10)
gromovdist(tr1,tr2,"l1")
gromovdist(tr1,tr2,"l2")
#thesame, but slower
gromovdist(d1=tr1,d2=tr2,type="lp",p=2)
gromovdist(tr1,tr2,"linf")

---