calculate, calculateDiff, simulate, or get the current log probabilities (densities) of one or more nodes of a NIMBLE model and (for calculate and getLogProb) return the sum of their log probabilities (or densities). Part of R and NIMBLE.
calculate(model, nodes, nodeFxnVector, nodeFunctionIndex)calculateDiff(model, nodes, nodeFxnVector, nodeFunctionIndex)
getLogProb(model, nodes, nodeFxnVector, nodeFunctionIndex)
simulate(model, nodes, includeData = FALSE, nodeFxnVector, nodeFunctionIndex)
A NIMBLE model, either the compiled or uncompiled version
A character vector of node names, with index blocks allowed, such as 'x', 'y[2]', or 'z[1:3, 2:4]'
An optional vector of nodeFunctions on which to operate, in lieu of model and nodes
For internal NIMBLE use only
A logical argument specifying whether data nodes should be simulated into (only relevant for simulate)
calculate and getLogProb return the sum of the log probabilities (densities) of the calculated nodes, with a contribution of 0 from any deterministic nodes
calculateDiff returns the sum of the difference between the new and old log probabilities (densities) of the calculated nodes, with a contribution of 0 from any deterministic nodes.
simulate returns NULL.
These functions expands the nodes and then process them in the model in the order provided. Expanding nodes means turning 'y[1:2]' into c('y[1]','y[2]') if y is a vector of scalar nodes. Calculation is defined for a stochastic node as executing the log probability (density) calculation and for a deterministic node as calculating whatever function was provided on the right-hand side of the model declaration.
Difference calculation (calculateDiff) executes the operation(s) on the model as calculate, but it returns the sum of the difference between the new log probabilities and the previous ones.
Simulation is defined for a stochastic node as drawing a random value from its distribution, and for deterministic node as equivalent to calculate.
getLogProb collects and returns the sum of the log probabilities of nodes, using the log probability values currently stored in the model (as generated from the most recent call to calculate on each node)
These functions can be used from R or in NIMBLE run-time functions that will be compiled. When executed in R (including when an uncompiled nimbleFunction is executed), they can be slow because the nodes are expanded each time. When compiled in NIMBLE, the nodes are expanded only once during compilation, so execution will be much faster.
It is common to want the nodes to be provided in topologically sorted order, so that they will be calculated or simulated following the order of the model graph. Functions such as model$getDependencies(nodes, ...) return nodes in topologically sorted order. They can be directly sorted by model$topologicallySortNodes(nodes), but if so it is a good idea to expand names first by model$topologicallySortNodes(model$expandNodeNames(nodes))