Module gpfa 

Structs

GpfaResult

GPFA result.

Functions

chol_logdet 🔒

log-determinant of an SPD matrix from its Cholesky factor (2 Σ log L_ii).

gp_kernel 🔒

Squared-exponential GP kernel for n time points.

gpfa

Extract smooth latent trajectories from parallel spike trains via EM.

gpfa_e_step 🔒

E-step: joint Gaussian posterior p(x|y) over all latents and time points.

gpfa_em_from_init

Run the GPFA EM loop from a fixed initialisation (the dispatchable kernel).

gpfa_log_likelihood 🔒

Exact marginal Gaussian log-likelihood via the Woodbury identity.

gpfa_m_step 🔒

M-step: update C, d, R.

gpfa_precision 🔒

Assemble the posterior precision M = blkdiag(K_j⁻¹) + AᵀR⁻¹A (row-major, n_state × n_state with n_state = n_latents · n_bins) and the GP prior log-determinant log|K|.

gpfa_transform

Project new spike trains using learned GPFA parameters.

spd_cholesky 🔒

Cholesky factor of a symmetric positive-definite row-major matrix (LAPACK-grade dense linear algebra via nalgebra). GPFA only factors SPD matrices — the GP prior, the posterior precision and the second-moment matrix — so the decomposition is the numerically stable, ~2x cheaper choice over a general LU.

spd_inverse 🔒

Inverse of an SPD matrix (row-major in, row-major out) via Cholesky.

Type Aliases

GpfaEmOutput

EM output: (trajectories, C, d, R_diag, log_likelihoods).