MCMC Result

MCMC Result

MCMC Result

Details

R6 class containing posterior samples from mcmc(). Access samples via the accessor methods below.

Methods

Public methods

• SrfResult$new()

• SrfResult$z()

• SrfResult$alloc()

• SrfResult$sufficient_stat()

• SrfResult$z_map()

• SrfResult$psi()

• SrfResult$emission_params()

• SrfResult$dag()

• SrfResult$edge_inclusion_probs()

• SrfResult$acceptance_rates()

• SrfResult$diagnostics()

• SrfResult$summary()

• SrfResult$plot()

---

Method new()

Create a new SrfResult. Use mcmc() instead.

Usage

SrfResult$new(raw, emission_type = NULL, nug = NULL, model_type = "mdgm")

Arguments

raw

List of raw MCMC output from C++.

emission_type

Character string or NULL.

nug

A NaturalUndirectedGraph or NULL.

model_type

Character string: "mdgm" or "mrf".

---

Method z()

Get the latent field samples.

Usage

SrfResult$z(iteration = NULL)

Arguments

iteration

Optional integer iteration to extract (1-indexed). If NULL, returns the full n x J matrix.

Returns

Integer matrix (n x J) or integer vector (length n).

---

Method alloc()

Get allocation counts (always available).

Usage

SrfResult$alloc()

Returns

Integer matrix (n x k) of per-site class counts across all iterations.

---

Method sufficient_stat()

Get per-iteration sufficient statistic T(z) (same-color edge count).

Usage

SrfResult$sufficient_stat()

Returns

Numeric vector of length J.

---

Method z_map()

Get the joint MAP configuration.

Usage

SrfResult$z_map()

Returns

Named list with z (integer vector), log_posterior (scalar), and iteration (1-indexed integer).

---

Method psi()

Get the psi (dependence parameter) samples.

Usage

SrfResult$psi()

Returns

Numeric vector of length J.

---

Method emission_params()

Get the emission parameters.

Usage

SrfResult$emission_params(iteration = NULL)

Arguments

iteration

Optional integer iteration to extract (1-indexed).

Returns

Named list appropriate to emission type, or NULL for standalone models.

---

Method dag()

Get the DAG parent vector samples.

Usage

SrfResult$dag(iteration = NULL)

Arguments

iteration

Optional integer iteration to extract (1-indexed). If NULL, returns the full n x J matrix.

Returns

Integer matrix (n x J) or integer vector (length n), or NULL for MRF models (which have no DAG structure). Values are 1-indexed parent vertex IDs; NA indicates a root.

---

Method edge_inclusion_probs()

Compute edge inclusion probabilities.

Usage

SrfResult$edge_inclusion_probs(nug = NULL, burnin = 0L)

Arguments

nug

A NaturalUndirectedGraph object. If NULL, uses the graph stored at construction time.

burnin

Number of initial iterations to discard (default 0).

Returns

A data frame with columns vertex1, vertex2, and prob.

---

Method acceptance_rates()

Get acceptance rates for MH steps.

Usage

SrfResult$acceptance_rates()

Returns

A named numeric vector with psi and graph rates.

---

Method diagnostics()

Compute MCMC diagnostics.

Usage

SrfResult$diagnostics(burnin = 0L)

Arguments

burnin

Number of initial iterations to discard (default 0).

Returns

A named list with rhat and ess (effective sample size) for psi and emission parameters.

---

Method summary()

Print a summary of the MCMC results.

Usage

SrfResult$summary(burnin = 0L)

Arguments

burnin

Number of initial iterations to discard (default 0).

Returns

Invisible named list with summary statistics.

---

Method plot()

Plot MCMC trace plots and diagnostics.

Usage

SrfResult$plot(burnin = 0L, which = "trace")

Arguments

burnin

Number of initial iterations to discard (default 0).

which

Character vector of what to plot: "trace" for trace plots of psi and emission params, "edge_inclusion" for edge inclusion probabilities (requires stored graph and igraph), "posterior_field" for posterior mode of the latent field (requires stored graph).

Returns

Invisible list of ggplot/igraph plot objects.