Score Columns of a Matrix Containing NAs by its Values

Description

Helper function for Predict used to score the columns of a matrix according to their values. The scoring of a given column is done by pair-wise comparisons with all other columns. The comparison of columns is done by pair-wise comparisons of the non-missing values. This procedure is robust to missing values, if all columns of the matrix have a similar (potentially shifted) distribution of values.

Usage

Score(mat)

Arguments

Value

Numeric vector of length ncol(mat).

Note

Interfaces a C++ function. The R package Rcpp is used.

Author(s)

Florian Gerber, florian.gerber@math.uzh.ch.

References

F. Gerber, R. Furrer, G. Schaepman-Strub, R. de Jong, M. E. Schaepman, 2016, Predicting missing values in spatio-temporal satellite data. http://arxiv.org/abs/1605.01038.

Examples

mat <- rbind(c( 1,  2, NA),
             c(NA, NA,  1),
             c( 2, NA,  3),
             c( 1,  5, NA),
             c(NA,  2,  5))
s <- Score(mat)

## manual calculation in R
Mean <- function(x) mean(x, na.rm = TRUE)
sByHand <- c(Mean(c(Mean(mat[,1] > mat[,2]),
                    Mean(mat[,1] > mat[,3]))),
             Mean(c(Mean(mat[,2] > mat[,1]),
                    Mean(mat[,2] > mat[,3]))),
             Mean(c(Mean(mat[,3] > mat[,1]),
                    Mean(mat[,3] > mat[,2]))))
stopifnot(identical(s, sByHand))

Results