dlm-js
    Preparing search index...

    Function dlmPrior

    • Create a DlmLossFn that adds Bayesian prior penalties to the Kalman deviance (−2·logL).

      The returned function computes: objective(θ) = deviance(θ) + Σ [−2 · log prior(paramᵢ)]

      Prior penalties (dropping constants that don't affect optimisation):

      • Inverse-Gamma(α, β) on variance σ² (params provides σ, the std dev): penalty = 4(α+1)·log(σ) + 2β/σ²
      • Normal(μ, σ_p) on coefficient φ: penalty = (φ − μ)² / σ_p²

      All operations are AD-safe (np.split, np.log, np.multiply, etc.) and compose inside jit() with zero overhead.

      Parameters

      Returns DlmLossFn

      import { dlmMLE, dlmPrior } from 'dlm-js';

      const prior = dlmPrior({
        obsVar:     { shape: 2, rate: 100 },
        processVar: { shape: 2, rate: 10 },
      });
      const result = await dlmMLE(y, { order: 1, loss: prior });
      // result.priorPenalty > 0