Skip to contents

Compute a Neural RDM for RSA

Source: R/distcalc.R
rsa_neural_rdm.Rd

Compute the full representational dissimilarity matrix (RDM) among observed neural patterns using the correlation distances used by rsa_model.

Usage

rsa_neural_rdm(
  patterns,
  method = c("spearman", "pearson"),
  pattern_center = c("none", "stimulus_mean")
)

Arguments

patterns

A finite numeric matrix with observations in rows and measured features (for example, voxels) in columns. At least two observations and two features are required. Each row must vary across features after optional pattern centering.

method

Correlation method: "spearman" (the default, matching rsa_model()) or "pearson". Spearman uses average ranks for ties.

pattern_center

Use "none" for uncentered patterns, or "stimulus_mean" to subtract each feature's mean across the supplied observations before computing correlations (and before Spearman ranking).

Value

A symmetric numeric matrix with one row and column per observation, a zero diagonal, and distances 1 - correlation. Observation order and row names are preserved on both axes; unnamed inputs yield unnamed outputs.

Details

These are correlation distances, not noise-whitened or crossvalidated distances. No averaging, feature selection, run/item exclusion, reliability estimate, or noise-ceiling calculation is performed. All pairs are returned; exclusions belong to rsa_design or the caller.

To reproduce an rsa_model() neural RDM, supply the same preprocessed pattern matrix and feature selection, with method matching the model's distmethod. Stimulus-mean centering depends on all supplied observations: centering separate runs or subsets generally changes the distances. The model centers its input before any design-level row selection or pair exclusions; reproduce that order when extracting subsets of this matrix.

Nonfinite inputs and patterns that are constant across features after centering are rejected because their correlations are undefined.

See also

rsa_model, rsa_design, cordist

Examples

patterns <- rbind(item_a = c(1, 2, 4, 3),
                  item_b = c(2, 1, 3, 4),
                  item_c = c(4, 3, 1, 2))
rdm <- rsa_neural_rdm(patterns)
rdm[lower.tri(rdm)]
#> [1] 0.4 2.0 1.6
rsa_neural_rdm(patterns, method = "pearson",
               pattern_center = "stimulus_mean")
#>           item_a    item_b   item_c
#> item_a 0.0000000 0.6837722 1.857493
#> item_b 0.6837722 0.0000000 1.759257
#> item_c 1.8574929 1.7592566 0.000000