use of IncrementalAnytimeExactBeliefPropagation.Model.Node.VariableNode in project aic-expresso by aic-sri-international.
the class PartitionTree method createPartitionTreeWithBFS.
/*------------------------------------------------------------------------------------------------------------------------*/
/**
* This class creates a partition based on a BFS process.
*
* There are various ways to create the partition tree.
*
* In fact any partition tree will provide an exact inference. But some of them will
* generate so many separator variables (or cutset variables) that the inference time
* can be close to the Naive computation.
*
* The study of the complexity of choosing the optimal tree (or a good heuristic)
* still needs to be done in the project. (TODO)
*
* The choice of a BFS seems intuitively good because it distributes the nodes more or less
* evenly among the partitions
*
* @param model
*/
public void createPartitionTreeWithBFS(Model model) {
HashMap<Node, PartitionTree> hashTable = new HashMap<>();
Set<Node> visited = new HashSet<>();
Queue<Node> queue = new LinkedList<>();
visited.add(node);
queue.add(node);
hashTable.put(this.node, this);
while (!queue.isEmpty()) {
Node n = queue.remove();
PartitionTree parentPartition = hashTable.get(n);
Set<Node> neighbors = new HashSet<>();
if (n.isFactor()) {
Collection<VariableNode> variableNeighbors = model.getExploredGraph().getAsOfB((FactorNode) n);
neighbors.addAll(variableNeighbors);
} else {
Collection<FactorNode> factorNeighbors = model.getExploredGraph().getBsOfA((VariableNode) n);
neighbors.addAll(factorNeighbors);
}
for (Node neighbor : neighbors) {
if (!visited.contains(neighbor)) {
queue.add(neighbor);
visited.add(neighbor);
PartitionTree pChild = new PartitionTree(neighbor);
pChild.parent = parentPartition;
hashTable.put(neighbor, pChild);
parentPartition.children.add(pChild);
}
}
}
}
Aggregations