CompareOneToMany-function {SpatialKWD}

Compare a reference spatial histogram to other histograms

Description

This function computes the Kantorovich-Wasserstein among a single reference histogram and a given list of other spatial histograms. All the histograms are defined over the same grid map.

The grid map is described by the two lists of N coordinates Xs and Ys, which specify the coordinates of the centroid of each tile of the map. For each tile i with coordinates Xs[i], Ys[i], we have a positive weight for each histogram.

The two lists of coordinates are passed to compareOneToMany as a matrix with N rows and two columns. The weights of the histograms are passed as a single matrix with N rows and M columns, where the first column is the reference histogram.

Usage

compareOneToMany(Coordinates, Weights, L = 3, recode = TRUE,
           method = "approx",    algorithm = "colgen",
           model="mincostflow",  verbosity = "silent",
           timelimit = 14400,    opt_tolerance = 1e-06,
           unbalanced = FALSE, unbal_cost = 1e+09, convex = TRUE)

Arguments

Coordinates

A Matrix with N rows and two columns:

• Coordinates[,1]: (First Column) Vector of horizontal coordinates of the centroids of each tile of the map (Xs). Data type: vector of positive integers.

• Coordinates[,2]: (Second Column) Vector of vertical coordinates of the centroids of each tile of the map (Ys). Data type: vector of positive integers.

Weights

A Matrix of positive weights of the tiles specified by Coordinates.

• Weights[,1]: (First Column) Weights of the reference spatial histogram, a weight for each tile located at position Xs[i], Ys[i] for i=1,...N. Data type: vector of positive doubles.

• Weights[,2:M]: (Remaining Columns) Weights of the spatial histograms to be compared with the reference histogram. Data type: vector of positive doubles.

L

Approximation parameter. Higher values of L give a more accurate solution, but they require a longer running time. Data type: positive integer.

recode

If equal to True, recode the input coordinates as consecutive integers.

method

Method for computing the KW distances: exact or approx.

algorithm

Algorithm for computing the KW distances: fullmodel or colgen.

model

Model for building the underlying network: bipartite or mincostflow.

verbosity

Level of verbosity of the log: silent, info, or debug.

timelimit

Time limit in second for running the solver.

opt_tolerance

Numerical tolerance on the negative reduced cost for the optimal solution.

unbalanced

If equal to True, solve the problem with unbalanced masses.

unbal_cost

Cost for the arcs going from each point to the extra artificial bin.

convex

If equal to True, compute the convex hull of the input points.

Details

The function compareOneToMany(Coordinates, Weights, ...) computes the distances among a reference spatial histogram and a given set of other histograms. All the histograms are specified by the M columns of matrix Weights, and where the support points (i.e., centroids of each tile of the map) are defined by the coordinates given in Xs and Ys in the two columns of matrix Coordinates. The algorithm used to compute such distance depends on the parameters specified as optional arguments of the function.

The most important is the parameter L, which by default is equal to 3 (see compareOneToOne).

Value

Return an R List with the following named attributes:

• distances: An array of M-1 KW-distances among the input histograms.

• status: Status of the solver used to compute the distances.

• runtime: Overall runtime in seconds to compute all the distances.

• iterations: Overall number of iterations of the Network Simplex algorithm.

• nodes: Number of nodes in the network model used to compute the distances.

• arcs: Number of arcs in the network model used to compute the distances.

See Also

See also compareOneToOne, compareAll, focusArea, Histogram2D, and Solver.

Examples

# Define a simple example
library(SpatialKWD)

# Random coordinates
N = 90
Xs <- as.integer(runif(N, 0, 31))
Ys <- as.integer(runif(N, 0, 31))
coordinates <- matrix(c(Xs, Ys), ncol=2, nrow=N)

# Random weights
m <- 3
test2 <- matrix(runif((m+1)*N, 0, 1), ncol=(m+1))

# Compute distance
print("Compare one-to-many with approximate algorithm:")
d <- compareOneToMany(coordinates, Weights=test2, L=3, method="approx")
cat("L: 3, runtime:", d$runtime, " distances:", d$distance, "\n")