Trick question? P(A) is just a probability of some event, so depending on the problem it could be calculated in any of the possible ways. "A" can for example correspond to a value of some random variable in a (dynamic) graphical model, taking observations into account, so that its probability value is obtained from belief propagation.