Mechanism abstract base class — Mechanism • kagu

A mechanism owns the full lifecycle of a node's conditional model, decoupled from any particular inference backend. Subclasses implement:

$fit(node, parents, data, ...) - fit the local model, returning an opaque fit object.
$predict_mean(node, parents, parent_values, fit) - the conditional mean E[node | parents] for each posterior draw, as a [n_chains, n_draws] matrix (Gaussian processes use a single chain, so n_chains = 1).
$log_marglik(node, parents, data) - the log marginal likelihood log P(node | parents), used by structure discovery.
$posterior_shape(fit) - c(n_chains, n_draws) for the fit.
$node_terms(node, parents, data, fit) - summary rows for model$summary().
$simulate_prior(node, parents, parent_values, data, ndraws) - replicated node values drawn from the prior predictive distribution.
$simulate_posterior(node, parents, parent_values, fit, draws, n) - replicated node values drawn from the posterior predictive distribution.
$function_draws(node, parents, grid, data, fit, ndraws) - prior or posterior draws of the node's conditional-mean function on a grid.

Methods

`Mechanism$fit()`

Fit the node's local model.

Usage

Mechanism$fit(node, parents, data, ...)

Arguments

node: Character scalar - the node name (response).
parents: Character vector of parent node names.
data: A data.frame with columns for the node and its parents.
...: Backend-specific arguments.

Returns

An opaque fit object.

`Mechanism$predict_mean()`

Conditional mean for each posterior draw.

Usage

Mechanism$predict_mean(node, parents, parent_values, fit)

Arguments

node: Character scalar - the node name.
parents: Character vector of parent node names.
parent_values: Named list of [n_chains, n_draws] matrices, one per parent, giving the parent values for each posterior draw.
fit: A fit object from $fit().

Returns

A [n_chains, n_draws] numeric matrix.

`Mechanism$log_marglik()`

Log marginal likelihood log P(node | parents).

Usage

Mechanism$log_marglik(node, parents, data)

Arguments

node: Character scalar - the node name.
parents: Character vector of parent node names.
data: A data.frame.

Returns

A single numeric.

`Mechanism$posterior_shape()`

Posterior shape of a fit.

Usage

Mechanism$posterior_shape(fit)

Arguments

fit: A fit object from $fit().

Returns

Named integer vector c(n_chains, n_draws).

`Mechanism$node_terms()`

Summary rows for this node (used by model$summary()).

Usage

Mechanism$node_terms(node, parents, data, fit)

Arguments

node: Character scalar - the node name.
parents: Character vector of parent node names.
data: A data.frame.
fit: A fit object from $fit().

Returns

A tibble with columns node, term, mean, sd, hdi_lower, hdi_upper.

`Mechanism$simulate_prior()`

Simulate node values from the prior predictive distribution.

Usage

Mechanism$simulate_prior(node, parents, parent_values, data, ndraws)

Arguments

node: Character scalar - the node name.
parents: Character vector of parent node names.
parent_values: Named list of [ndraws, n] matrices of simulated parent values, one per parent (empty for a root node).
data: A data.frame - sets the node's location/scale and n.
ndraws: Integer - number of prior predictive draws.

Returns

A [ndraws, n] numeric matrix.

`Mechanism$simulate_posterior()`

Simulate node values from the posterior predictive distribution (conditional mean plus observation noise).

Usage

Mechanism$simulate_posterior(node, parents, parent_values, fit, draws, n)

Arguments

node: Character scalar - the node name.
parents: Character vector of parent node names.
parent_values: Named list, one entry per parent: either a length-n vector shared by every draw (e.g. the observed parents) or a [length(draws), n] matrix with one row per draw.
fit: A fit object from $fit().
draws: Integer vector - which posterior draws to use.
n: Integer - number of observations to simulate per draw.

Returns

A [length(draws), n] numeric matrix.

`Mechanism$function_draws()`

Prior (fit = NULL) or posterior draws of the node's conditional-mean function, evaluated on a grid of parent values.

Usage

Mechanism$function_draws(node, parents, grid, data, fit = NULL, ndraws = 50L)

Arguments

node: Character scalar - the node name.
parents: Character vector of parent node names.
grid: A data.frame with one column per parent.
data: A data.frame - sets the node's location/scale.
fit: A fit object from $fit(), or NULL for prior draws.
ndraws: Integer - number of function draws.

Returns

A [ndraws, nrow(grid)] numeric matrix.

`Mechanism$clone()`

The objects of this class are cloneable with this method.

Usage

Mechanism$clone(deep = FALSE)

Arguments

deep: Whether to make a deep clone.