resemble (version 3.0.1)

diss_correlation: Correlation dissimilarity method constructor

Description

Creates a configuration object that fully specifies a correlation (or moving correlation) dissimilarity method. Pass the result to dissimilarity() to compute the dissimilarity matrix.

Usage

diss_correlation(ws = NULL, center = TRUE, scale = FALSE)

Value

An object of class c("diss_correlation", "diss_method") — a list holding the validated method parameters. Intended to be passed to dissimilarity(), not used directly.

Arguments

ws

Either NULL (default) or an odd integer greater than 2. When NULL, standard Pearson correlation dissimilarity is used. When an odd integer is provided, a moving (rolling) correlation dissimilarity is computed using a window of that size.

center

Logical. Should the data be mean-centered before computing dissimilarities? Centering is applied jointly to Xr and Xu (if provided) based on their combined column means. Default is TRUE.

scale

Logical. Should the data be scaled (divided by column standard deviations) before computing dissimilarities? Scaling is applied jointly to Xr and Xu (if provided). Default is FALSE.

Parallel execution

The underlying C++ implementation uses OpenMP for parallel computation. Thread count is controlled by the OMP_NUM_THREADS environment variable. To limit threads (e.g., when calling from within a parallel backend):

Sys.setenv(OMP_NUM_THREADS = 1)

Author

Leonardo Ramirez-Lopez

Details

The correlation dissimilarity between two observations x_ix_i and x_jx_j is:

d(x_i, x_j) = 12(1 - (x_i, x_j))d(x_i, x_j) = (1/2)(1 - rho(x_i, x_j))

where rho is the Pearson correlation coefficient. This is used when ws = NULL.

When ws is specified, the moving correlation dissimilarity is:

d(x_i, x_j; ws) = 12\,ws _k=1^p - ws (1 - (x_i,(k:k+ws),\, x_j,(k:k+ws)))d(x_i, x_j; ws) = 1/(2*ws) * sum_k=1^p-ws (1 - rho(x_i,(k:k+ws), x_j,(k:k+ws)))

where wsws is the window size and pp is the number of variables.

See Also

dissimilarity, diss_euclidean, diss_mahalanobis, diss_cosine

Examples

# Standard correlation dissimilarity
m <- diss_correlation()

# Moving correlation with window size 41
m <- diss_correlation(ws = 41)

# Without centering
m <- diss_correlation(center = FALSE)