Statistics for Data Science II
3 Credit Hours
Focuses are on Bayesian reasoning for Data Science in the first half of the course. Students will learn how to formulate and implement Bayesian inference using the prior-to-posterior paradigm, which represents the core of the Bayesian perspective that one’s beliefs can be updated with latest evidence. Bayesian statistics is increasingly popular due to its flexibility and recent improvements in computational techniques. Topics include Bayes theorem, conjugate priors, posterior distributions, credible intervals, Monte Carlo approximation, MCMC, Gibbs sampling, Metropolis-Hastings algorithm, Bayesian hypothesis testing, and hierarchical modeling. In the second half of the course, students will learn theoretical tools to evaluate statistical evidence from randomized experiments. Topics include optimal sample size determination, A/B testing, factorial and fractional designs, response surface methods, conjoint designs, sequential designs, bandit problems used in online ads, design and modeling of complex computer experiments, etc. Students will perform data analysis using appropriate software tools in the course.