LDS_EM: Learn LDS model

In ntthung/ldsr: Linear Dynamical System Reconstruction

Description

Estimate the hidden state and model parameters given observations and exogenous inputs using the EM algorithm. This is the key backend routine of this package.

Usage

LDS_EM(y, u, v, theta0, niter = 1000L, tol = 1e-05)

Arguments

y	Observation matrix (may need to be normalized and centered before hand) (q rows, T columns)
u	Input matrix for the state equation (m_u rows, T columns)
v	Input matrix for the output equation (m_v rows, T columns)
theta0	A vector of initial values for the parameters
niter	Maximum number of iterations, default 1000
tol	Tolerance for likelihood convergence, default 1e-5. Note that the log-likelihood is normalized

Value

A list of model results

• theta: model parameters (A, B, C, D, Q, R, mu1, V1) resulted from Mstep

• fit: results of Estep

• liks : vector of loglikelihood over the iteration steps

Note

This code only works on one dimensional state and output at the moment. Therefore, transposing is skipped, and matrix inversion is treated as /, and log(det(Sigma)) is treated as log(Sigma).

ntthung/ldsr documentation built on Aug. 28, 2020, 2:34 a.m.