Bobby's Teaching Page
This academic year, 2009-10, my teaching duties are split between UC Santa Barbara and the University of Cambridge. At UCSB I am teaching a PSTAT 215A Bayesian Inference, a graduate course, in the Fall quarter. In Cambridge I am teaching Part III/Mphil Time series and Part III/Mphil Monte Carlo Inference, which are taught in serial in Lent as part of the Mathematical Tripos.
2009 Fall PSTAT 215A Bayesian Inference
PSTAT 215A is a graduate course in Bayesian Inference. The course will focus on understanding the principles underlying Bayesian modeling and on building experience in the use of Bayesian analysis for making inference about real world problems. Particular attention will be paid to the computational techniques (e.g., MCMC) needed for most problems and their implementation in the R language for statistical computing.
Course syllabus (including required and recommended texts)
Notices:
- Final project on data due 11 Dec 2009
- The "midterm" exam will take place on 19 Nov
- Planned date of "third week" quiz: 20 Oct
- Starting Thursday 1 Oct the lectures will be held in Phelps 1160
- Lecture canceled on 15 Oct A make-up lecture will be held on Monday 19 October, 3:45-5pm in GIRV 2124
- The 12 Nov Lecture will be given by Jarad Niemi
Homeworks: (due at 3pm on the date indicated in my mailbox, 5524 South Hall)
- Homework 1 covering Parts 0-2, due 15 Oct 2009
- Homework 2 covering Parts 3-4.5, due 27 Oct 2009. You will need the following data: male-bachelors, male-none for question 1, and schools one, two, and three for question 4.
- Homework 3 covering Parts 4.5-5, due 10 Nov 2009. You will need the following data on clouds.
- Homework 4 covering Parts 6-7, due 24 Nov 2009. You will need the following data on ages, swimming, studdying, and bicycling.
Lecture slides:
- Parts 0 & 1: Introduction and fundamentals
- Part 2: One-parameter models
- Part 3: Monte Carlo inference
- Part 4: Multi-parameter and normal models
- Part 5: MCMC: Metropolis and Gibbs samplers
- Part 6: Multivariate normal and linear models
- Part 7: Hierarchical models
- Part 8: Model criticism, selection, and averaging
- Part 9: GLMs, Hierarchical LMs & GLMs
- Part 10: Latent variables and missing data
Demos:
- Parts 0 & 1: Rare event and binomal, Poisson, and normal examples
- Part 2: beta binomal, and poisson examples on education v. fertility and heart transplants
- Part 3: Monte Carlo inference, rejection sampling and importance sampling; also see the education v. fertility (datafile) example
- Part 4: the midge and IQ examples for Normal models, and the CBS poll for multinomials
- Part 5: the Gibbs sampler for a bivariate normal; the MH sampler(s) for a bivariate normal; the 1-d normal example; and the ACF and ESS examples; also see the midge example
- Part 6: the bivariate normal, reading comprehension (datafile), Wishart deviates, and oxygen uptake (datafile) examples
- Part 7: the math testing (datafile) and rat tumor (datafile) examples
- Part 8: the diabetes exammple
- Part 9: the Poisson regression (datafile) example, a plot of binomial links, the math testing/SES (same data as above) and mice intestine (datafile) examples
- Part 10: the data augmentation example
2010 Lent Part III/Mphil Time Series
Time Series is a graduate level course taught in serial with Monte Carlo Inference (first 8 lectures). This year it is being offered in the Lent term. It covers topics like ARMA processes, spectral methods, the Kalman Filter, etc.
Course syllabus (including recommended texts)
Notices:
- Examples class Tuesday 9 February, 4-6pm in MR13
- Two lectures canceled: 27 and 29 Jan
- Makeup classes on 26 Jan and 2 Feb, 2pm in MR3
Example Sheet:
- Example Sheet 1 requiring data
on salt and
HL temperatures
- Solutions have been kindly prepared by James Lawrence, as well as accompanying R code, which requires the Kalman filtering library.
Supplimentary slides:
Demos:
- Characteristics of time series requiring data on speech, SOI, and stock recruitment
- Basics requiring data on global temperatures
- ARMA models requiring data on stock recruitment and varves
- Periodogram requiring data on SOI and stock recruitment
- State space models requiring library functions and temperature data HL and Folland
2010 Lent Part III/Mphil Monte Carlo Inference
Monte Carlo Inference is a graduate level course taught in serial with Time Series (last 16 lectures). This year it is being offered in the Lent term. It covers topics like random number generation, importance sampling, the bootstrap and jacknife for estimation and hypothesis tests, MCMC, sequential importance sampling, simulated annealing, and the EM algorithm.
Course syllabus (including recommended texts)
Notices:
- The revision examples class will be held in MR4 on 18 May, from 4-6pm. The last three years worth of exam questions (except on uniform pseudo-random number generation) will be covered
Vignettes:
- on pseudo-random number generators
- on generating deviates from common distributions
- on effective sample size
- on power of a Monte Carlo test
Example sheets:
- Example Sheet 1 (12 Feb 2010)
Examples class 24 Feb, 4-6pm in MR9 (CMS) -- Solutions- R code to accompany #3b: rnorm.R
- requires the C code in rnorm.c or, on Windows, the library rnorm.so.dll
- R code to accompany #8c: norm_k.R
- R code to accompany #10b: ctrlvar.R
- R code to accompany #3b: rnorm.R
- Example Sheet 2 (22 Feb 2010),
requires the data file ar3.txt and optionally the code
rmultnorm.R
Examples class 9 March, 4-6pm in MR13 (CMS) -- Solutions - Example Sheet 3 (8 March 2010)
Examples class 4 May, 4-6pm in MR11 (CMS) -- Solutions- R code to accompany #3: boh.R
Demos:
- reject.R: Rejection sampling from Lecture 2
- rou.R: Ratio of Uniforms sampling from Lecture 3
- is.R: Importance Sampling from Lectures 4 and 5
- resample.R: Resampling methods (Jackknife and Bootstrap) from Lecture 8
- gibbs.R which requires gibbs_norm.R: Gibbs sampling from Lecture 9
- da.R: Data augmentation from Lecture 9
- mh.R which requires mh_norm.R: the Metropolis Hastings algorithm from Lecture 10
- ess.R which requires gibbs_norm.R and mh_norm.R: Effective sample size from Lecture 11
- rjmcmc.R: Reversible Jump MCMC from Lecture 12
- sa.R: Simulated Annealing from Lecture 14
- em.R: Expectation Maximisation from Lecture 16
Robert B. Gramacy -- 2010