Covariate matrix X [n × q] (empty array when no covariates). In MATLAB DLM, this is XX.
Deviance: -2 · log-likelihood. In MATLAB DLM, this is lik.
Observation matrix F. p=1: [m] row vector (backward compat). p>1: [p, m].
Filtered state means [n, m]. In MATLAB DLM, this is xf.
Filtered state covariances [n, m, m]. In MATLAB DLM, this is Cf.
State transition matrix G (m × m)
Initial state covariance (scaled). In MATLAB DLM, this is C0.
Initial state mean (after first smoother pass). In MATLAB DLM, this is x0.
Innovations (one-step-ahead prediction errors). In MATLAB DLM, this is v.
Innovation variances. In MATLAB DLM, this is Cp.
State dimension (m_base + q for covariates)
Mean absolute percentage error
Mean squared error
Number of observations
Number of non-NaN observations
Observation noise standard deviations. In MATLAB DLM, this is V.
Observation dimension (1 for univariate, >1 for multivariate)
Raw residuals: y - yhat. In MATLAB DLM, this is resid0.
Residual variance. In MATLAB DLM, this is s2.
Residual sum of squares. In MATLAB DLM, this is ssy.
Scaled residuals: (y - yhat) / V. In MATLAB DLM, this is resid.
Smoothed state means [n, m]. In MATLAB DLM, this is x.
Smoothed state covariances [n, m, m]. In MATLAB DLM, this is C.
Smoothed state standard deviations [n, m] = sqrt(diag(smoothedCov)). In MATLAB DLM, this is xstd.
Standardized residuals: innovation / sqrt(innovationVar). In MATLAB DLM, this is resid2.
State noise covariance W (m × m)
Observations
Fitted values: yhat = F · filtered state.
Prediction standard deviations: sqrt(F·C·F' + V²).
Result from dlmFit — materialized TypedArrays with JS-idiomatic names.
State estimates use StateMatrix and CovMatrix wrappers over contiguous [n, m] / [n, m, m] row-major buffers — zero-copy from the JIT output with no transpose.
MATLAB DLM users: call toMatlab to get the familiar
x[state][time]layout and single-letter field names.