Skip to contents

Score the Similarity of Two Kruskal Tensors

Source: R/ktensor_methods.R
score.Rd

Computes the factor match score between two KTensor objects with the same number of components, mirroring the MATLAB Tensor Toolbox score. Each candidate component pairing is scored by the product of the absolute cosine similarities of the matched factor columns, optionally discounted by the relative difference of the component weights; components are matched greedily.

Usage

score(x, ...)

# S3 method for class 'KTensor'
score(x, y, lambda_penalty = TRUE, greedy = TRUE, ...)

Arguments

x

A KTensor (the estimate).

...

Additional arguments passed to methods.

y

A KTensor with the same dimensions and number of components (the reference).

lambda_penalty

Logical; if TRUE (default) each pair score is multiplied by 1 - |la - lb| / max(la, lb) computed from the normalized weights.

greedy

Logical; must be TRUE (greedy matching is the only implemented strategy, as it is the MATLAB default).

Value

A list with elements score (mean of the matched component scores), perm (the permutation of y's components matched to x), and x (a normalized copy of x arranged to match y).