When predictions are made on a variable with evidence, the prediction simply returns the evidence.
When predictions are made on a variable with evidence, the variable's own evidence is ignored.
For example, consider a network with variables {A,B,C}, with evidence set on all variables. If you query {P(A),P(B),P(C)}, P(A) will only use the evidence on B and C, P(B) will only use the evidence on A and C, and P(C) will only use the evidence on A and B.
This allows the prediction of variables, without having to perform multiple calls to {@link com.bayesserver.inference.IInference#query}, each time setting a different variable to null (missing).
Determines how predictions on variables with evidence are performed.