APPROXIMATE INFERENCE IN BAYESIAN NETWORKS

V diplomski nalogi bomo opisali metode direktnega vzorčenja, kamor spadajo algoritem vzorčenja, algoritem vzorčenja zavrnitve in algoritem uteženja verjetnosti. Pojasnili bomo tudi delovanje metode vzorčenja s pomočjo algoritma markovske verige Monte Carlo v eni izmed njenih preprostih oblik, Gibbsovem vzorčevalniku. Algoritme bomo implementirali v predstavitveni aplikaciji, testirali njihovo delovanje na enakih primerih in primerjali dobljene rezultate.In this thesis we will describe methods of direct sampling, namely prior sample, rejection sampling and likelihood weighting. We will also explain how Gibbs sampler, one simple variant of Markov chain Monte Carlo algorithm works. We well implement these algorithms in a representation application, then test how they work on the same examples and compare the results