<< Click to Display Table of Contents >> Navigation:  Using SMILE Wrappers > Dynamic Bayesian networks > Temporal beliefs

For plate nodes in a dynamic Bayesian network, the inference algorithm computes temporal beliefs, which are marginal posterior probability distributions that vary over time. Temporal beliefs are returned by Network.get_node_value.

Because temporal beliefs depend on time, the resulting belief array is larger than for a static node. If a node has X outcomes and the number of unrolled time slices is Y, the array will contain X * Y elements. The elements corresponding to a single time slice are stored contiguously: indices [0 .. X-1] represent beliefs for t=0, indices [X .. 2X-1] for t = 1, and so on.

Only PLATE nodes contain temporal beliefs. Other temporal node types (such as ANCHOR, TERMINAL, and CONTEMPORAL nodes) have regular belief arrays, whose size equals the number of their outcomes.