Week 4 (Probabilistic models) Flashcards
How to expand this probability distribution
What is a Bayesian network
How does a Bayesian network factor based on the parent notation
What does a DAG represent
A DAG represents a certain factoring of a probability distribution
Do you have to define a prior probability distribution for Bayesian networks
How is polynomial regression factorised in a bayesian network
Plate notation
What parameters does the bayesian network model for polynomial regression contain
What is the bayesian model for polynomial regression
What is Naive Bayes
What is hierarchical regression
Separate regressors where the parameters come from the same underlying distribution
What makes 2 variables conditionally independent
What are the 4 rules of independence (triplets) within a bayesian network
What is a collider in a bayesian network
What are the requirements for d-separation
d-separation = independence
What are the 2 requirements for a path to be blocked
What is the Bayesian posterior distribution, likelihood function, and conjugate prior
(We want to know the probability of these being the correct parameters given the data)
What are the problems with the Bayesian approach
What is univariate sampling
What is ancestral sampling
How to sample from marginal and conditional distributions (and what do these mean)
How to approximate an expected value wrt a posterior distribution
What are Markov chains
What is MCMC
What is the Metropolis-Hastings (and how it works)
What is the Metropolis-Hastings acceptance probability formula
Does Metropolis-Hastings (always) work?
Typically
How is Metropolis-Hastings typically run (ie what 2 things are done to run it well)