Groups a signature set into cosine-threshold connected components:
signatures with pairwise cosine >= `threshold` share a cluster. Each
cluster reports its size and its medoid (the member with the highest
mean within-cluster cosine). Deterministic.
Usage
ms_cluster_signatures(signatures, threshold = 0.9)
Arguments
- signatures
An m x k numeric matrix (columns sum to 1 after
normalization; canonical rownames enforced as everywhere), or an
[MsSignature].
- threshold
Single number in (0, 1]: the cosine join threshold
(default 0.90, the Islam TP line).
Value
A data.frame with one row per signature: `signature`,
`cluster` (label `C1`, `C2`, ... by descending size),
`is_medoid`, `cluster_size`; the pairwise cosine matrix rides as
attribute `cosines`.