Crate modewise_vlmd 

Variational Latent Mode Decomposition.

Morante and Rehman (2025), “Latent Mode Decomposition”, arXiv:2505.17797. Built from the paper’s equations; where the paper and its reference code (github.com/Dmocrito/vlmd) disagree, docs/vlmd.md records which one is followed and why.

Structs

Vlmd

VlmdOutput

Everything VLMD estimates. modes are the channel modes U⁽ᵏ⁾ = Θ⁽ᵏ⁾A (Eq. 7); they sum to the latent fit ZA, not exactly to the signal.

VlmdParams

Defaults follow the reference implementation.

Functions

lasso

Sparse loadings for every channel: column c of loadings (latent × channel) minimises (1/2T)‖x_c − Zᵀa‖² + penalty·‖a‖₁, where latents is Z (latent × sample) and signal is X (channel × sample). This is scikit-learn’s Lasso(alpha=penalty, fit_intercept=False).