Question

Pr. 1.4 Consider the latent Dirichlet allocation (LDA) or topic model. Write out how one would use Gibbs sampling to estimate the model parameters.

Answer 1

for LDA model, barameter e stimation by Gibb's samplim technia We construct a Maukov chain whick is sed to sample rom a desired joint diut ie -One dimensional conditonaldiat

r K-,2, do do ack tcraton of outer loa is a th of Gibbs ampleりand ~(k), afer a sweep is a sample Chain te desived Joirt diit Next, we assum that the spread ofe data can be modeled a Namaldist with mean μ and variance 2 So lie take . Batesian, a#road &Hace -this uncertainti rer se ramna We wou ditt

ampling direct can be dayticult. Plaxver, we can carily hanple fon he fella Conditional dist ; 20 In articular,地have ^^っ where 2.