| Title: | Wrapper to the 'Emd-L1' Library by Haibin Ling and Kazunori Okada |
|---|---|
| Description: | The 'Emd-L1' distance metric from the paper by H. Ling and K. Okada, An Efficient Earth Mover's Distance Algorithm for Robust Histogram Comparison, IEEE Transaction on Pattern Analysis and Machine Intelligence (PAMI), 29(5):840-853, 2007, is provided for use by R. It is based on code downloaded from the URL <http://www.ist.temple.edu/~hbling/code/EmdL1_v3.zip>. |
| Authors: | Rainer M Krug <[email protected]> and Dirk Eddelbuettel <[email protected]> |
| Maintainer: | Dirk Eddelbuettel <[email protected]> |
| License: | file LICENSE |
| Version: | 0.1.3 |
| Built: | 2024-10-05 03:40:24 UTC |
| Source: | https://github.com/eddelbuettel/earthmovdist |
The emdL1 provides a wrapper to the emd-L1 distance norm by
Haibin Ling and Kazunori Okada.
emdL1(x, y, dims=NULL, verbose=FALSE)emdL1(x, y, dims=NULL, verbose=FALSE)
x |
vector, matrix or 3D array containing signature 1 |
y |
vector, matrix or 3D array containing signature 2 |
dims |
optional vector of dimensions of input data, see Details |
verbose |
optional boolean flag indicating verbose operation |
This function provides an implementation of the Earth Movers Distance (EMD), using the L1 (taxicab) metric.
To quote wikipedia (http://en.wikipedia.org/wiki/Earth\_Mover\'s\_Distance):
'The earth mover's distance (EMD) is a mathematical measure of the distance between two distributions over some region D. Informally, if the distributions are interpreted as two different ways of piling up a certain amount of dirt over the region D, the EMD is the minimum cost of turning one pile into the other; where the cost is assumed to be amount of dirt moved times the distance by which its is moved'
(For a detailed discussion of the EMD, see http://homepages.inf.ed.ac.uk/rbf/CVonline/LOCAL\_COPIES/RUBNER/emd.htm)
The distance 'the dirt' has to be moved is in the implementation of emd-L1, the L1 distance: if we have two points (x1, y1) and (x2, y2), the L1 distance (or taxicab metric) is abs(x2-x1)+abs(y2-y1).
The above definition only covers the case, where the 'amount of dirt' in both distributionssignatures is the same. In this case, the EMD is identical to the 1st Mallows distance or 1st Wasserstein distance.
For details on how the emd-L1 distance is computed in this algorithm, see
@ARTICLE{
author = {Ling, Haibin and Okada, K.},
title = {An efficient earth mover's distance algorithm for robust histogram
comparison},
journal = {IEEE Transactions on Pattern Analysis and Machine Intelligence},
year = {2007},
volume = {29},
pages = {840-53},
number = {5},
month = {05},
doi = {10.1109/TPAMI.2007.1058},
publication-type = {J},
publisher = {IEEE Comput. Soc},
type = {Journal Paper}
}
The function has to be called by providing two signatures (x and y). these can be vectors (1D signatures), matrix (2D signatures) or 3D arrays (3D signatures). The dimensions of x and y have to be the same, and x and y should not contain any missing or infinite values.
If the vector dims is supplied, x and y have to be vectors, and dims will be used to interpet those as vectors (length(dims)==1), matrix (length(dims)==2) or 3D array (length(dims)==3).
If the vector dims is not supplied and dim(x)==dim(y), dims is set internally to dim(x).
A real number with the EMD-L1 distance metric between the two histograms
Dirk Eddelbuettel and Rainer M Krug ([email protected]) for the R package; Haibin Ling and Kazunori Okada for emd-L1.
set.seed(42) x <- rnorm(100) y <- x + rnorm(100)/10 emdL1(x, y, verbose=TRUE)set.seed(42) x <- rnorm(100) y <- x + rnorm(100)/10 emdL1(x, y, verbose=TRUE)