Soft evidence (also known as virtual evidence), can be specified when the exact state of a discrete variable is unknown, yet information exists about the chances of the variable being in particular states.
Soft evidence can be set on discrete variables, including discrete temporal (time series) variables.
The Soft Evidence window is launched by right clicking a node and selecting the Edit soft evidence menu item, or from the Evidence window, when the soft evidence option is selected for a discrete variable.
To set soft evidence, enter a probability distribution across the discrete variable states. A value of 0 indicates that a state is not currently possible.
The soft evidence entered can be adjusted by checking the Divide by option and selecting a distribution. This can be useful when you want a query on the node to equal the soft evidence, taking into account the node's original distribution or based on queries on the network.