From: A survey of Monte Carlo methods for parameter estimation
Notation | Description |
---|---|
Dy | Dimension of the data. |
L | Number of data available. |
\(\mathbf {y} \in \mathbb {R}^{LD_{\mathrm {y}}}\) | LDy-dimensional observations vector, y=vec{y1,…,yL} with \(\mathbf {y}_{i} \in \mathbb {R}^{D_{\mathrm {y}}}\) for i=1,…,L. |
Dθ | Dimension of the parameter space. |
\(\Theta = \Theta _{1} \times \cdots \times \Theta _{D_{\theta }}\phantom {\dot {i}\!}\) | Feature space for the parameter vector θ. |
\(\boldsymbol {\theta } \in \mathbb {R}^{D_{\theta }}\) | Dθ-dimensional parameter vector, \(\phantom {\dot {i}\!}\boldsymbol {\theta } = [\theta _{1}, \ldots, \theta _{D_{\theta }}]\) with θd∈Θd for d=1,…,Dθ. |
θ(m) | mth sample of the parameter vector in MC and RS. |
θ(t) | Sample of the parameter vector at the tth iteration in MCMC methods. |
\(\bar {\pi }(\boldsymbol {\theta }|\mathbf {y}) \equiv \bar {\pi }(\boldsymbol {\theta })\) | Target (i.e., posterior) PDF. |
π(θ|y)≡π(θ) | Target function (i.e., non-negative but unnormalized). |
p0(θ) | Prior probability density function. |
ℓ(y|θ) | Likelihood. |
Z(y) | Normalizing constant of the target (a.k.a. partition function, marginal likelihood, or model evidence). |
\(\bar {\pi }(\theta _{d}|\boldsymbol {\theta }_{\neg d})\) | Full conditional PDF for the dth parameter given all the other parameters (used in the Gibbs sampler). |
T | Number of Monte Carlo iterations performed. |
Tb | Number of iterations for the burn-in period in MCMC. |
N | Number of proposals used in multiple IS approaches. |
M | Number of samples drawn in the MC algorithm, RS and IS approaches. Usually M≥N in MIS. |
\(\bar {q}(\boldsymbol {\theta })\), \(\bar {q}_{t}(\boldsymbol {\theta })\), \(\bar {q}_{m,t}(\boldsymbol {\theta })\) | Proposal PDF. |
q(θ), qt(θ), qm,t(θ) | Proposal function (i.e., non-negative but unnormalized) for t=1,…,T and m=1,…,M. |
wm,t(θ) | Unnormalized weight of the mth particle (m=1,…,M) at the tth iteration (t=1,…,T) for AIS approaches. |
wm,t(θ) | Normalized weight of the mth particle (m=1,…,M) at the tth iteration (t=1,…,T) for AIS approaches. |
\(\hat {\pi }(\boldsymbol {\theta })\) | Random measure used to approximate the target at the tth iteration. |
\(\boldsymbol {\mathcal {N}}(\mu,\mathbf {C})\), \(\boldsymbol {\mathcal {N}}(\cdot |\mu,\mathbf {C})\) | Gaussian PDF with mean μ and covariance C. |
\(\boldsymbol {\mathcal {U}}(\mathcal {I})\) | Uniform PDF within the interval \(\mathcal {I}\). |