ICIC Data Analysis Workshop 2016
ICIC DATA ANALYSIS WORKSHOP, SEPTEMBER 58 2016
Principled statistical methods for researchers
Venue:
Imperial Centre for Inference and Cosmology (ICIC), Imperial College, South Kensington, London.
Dates:
Starts: 2 p.m. 5 September 2016.
Ends: 4 p.m. 8 September 2016.
SUMMARY
We will run a 4day course/workshop on statistical methods and tools for data analysis, aimed at PhD students, postdocs and any researchers interested in understanding Bayesian statistics and numerical techniques of data analysis. The course plan is to combine lectures with handson computational work in the afternoons. It will concentrate on setting down firm foundations of principled Bayesian data analysis, but a feature of the workshop will be a substantial element of handson classes where participants will learn how to apply the ideas in practice. It will be hosted by the Imperial Centre for Inference and Cosmology (ICIC) at Imperial College.
BACKGROUND
Most researchers will at some point be required to perform some form of data analysis. This may be anything from simple linefitting, through parameter estimation, to complex and computationallydemanding sampling for model selection on large datasets. Anecdotal evidence suggests that many researchers are not well prepared for this, often doing the right thing incorrectly, or picking up an inappropriate statistical tool. The purpose of this course is to provide understanding of principled data analysis, and experience of applying appropriate methods to data.
PREPARATION
We expect all participants to bring their own laptop, and to do a simple computational exercise in advance (in whatever language suits) to ensure they have appropriate software in place before the workshop starts.
COSTS
The workshop has sponsorship from STFC's Education, Training and Careers Committee, and also a generous donation from Winton Capital.
STFCfunded students will have accommodation, meals, refreshments and social activities paid for, and can also claim travel from STFC. Other participants may need to pay accommodation costs and a £75 course fee  see the registration page for more details of eligibility for cost reductions.
ACCOMMODATION
We have reserved rooms in Imperial College Student Accommodation, in Beit Hall, within easy walking distance of the workshop.
Social events
There will be a drinks reception on the roof terrace on Monday evening, and a barbecue at Princes Gate on Wednesday. Tuesday evening will be free to allow participants to explore London.
PROVISIONAL PROGRAMME
Day 1 (Monday 5 September 2016). Clore Lecture Theatre, Huxley Building
 From 1.30 p.m. Registration.
 Start of Workshop 2 p.m.
 Bayesian Foundations:
 What is probability?
 The Laws of Probability and Bayes’ Theorem
 Priors
 Parameter inference
 Marginalization
 Confidence intervals, credibility intervals
 Problem class: Simple problems
 Tutorial: day summary
 Talk by Dr Geraint Harker (Winton Capital and UCL Astrophysics)
 End: 5.30 p.m.
 Drinks reception. Roof Terrace, Level 8, Blackett Lab.
Day 2 (Tues 6 September 2016). Skempton Building Lecture Theatre 201
 Bayesian Computation: Parameter Estimation and Sampling
 Gridbased methods
 Markov Chain Monte Carlo
 MetropolisHastings algorithm
 Convergence tests – RubinGelman

 Hands on: MCMC code from scratch. Cosmology from the Supernova Hubble Diagram.

 Day summary
 End: 5 p.m.
 Evening: free
Day 3 (Weds 7 September 2016) Clore Lecture Theatre, Huxley Building
 Gibbs Sampling
 Hamiltonian Monte Carlo
 Day summary

 End: 5 p.m.
 6 p.m. Workshop Barbecue. 58 Princes Gate.
Day 4 (Thurs 8 September 2016) Clore Lecture Theatre, Huxley Building
 Why not pvalues and reduced chisquared?
 Model Comparison with Bayesian Evidence
 Hands on: Bayesian evidence: the SavageDickey Density Ratio
 Bayesian Hierarchical Models
 Workshop summary
 4 p.m. End of Workshop
LEARNING OUTCOMES
At the end of the Workshop, the participants should be able to (nonexhaustive list):
 Express stochastic problems in terms of fundamental probability and Bayes’ theorem.
 Demonstrate by application to real data understanding of probability, inference, priors, posteriors, marginalisation, parameter estimation, hypothesis testing, model selection, sampling.
 Code and apply a simple MCMC program to physical data.
 Formulate model selection problems in a principled statistical framework, and be capable of executing some methods of solution.
COURSE TEAM
Prof Alan Heavens, Prof Andrew Jaffe, Dr Jonathan Pritchard, Dr Daniel Mortlock (ICIC Physics and Mathematics), Dr Elena Sellentin (University of Geneva/ICIC Physics)
Point of contact: Professor Alan Heavens, Director, Imperial Centre for Inference and Cosmology, Blackett Laboratory, Prince Consort Road, London SW7 2AZ. Email a.heavens@ null imperial.ac.uk Tel. 0207 594 2930, or Louise Hayward (l.hayward@ null imperial.ac.uk) 0207 594 7679.
Registration
Registration is now closed. Please contact Louise Hayward for adminstrative matters l.hayward@ null imperial.ac.uk.
Lecturers and Demonstrators
Prof Alan Heavens (ICIC Physics)
Prof Andrew Jaffe (ICIC Physics)
Dr Daniel Mortlock (ICIC Physics and Mathematics)
Dr Jonathan Pritchard (ICIC Physics)
Dr Elena Sellentin (University of Geneva and ICIC Physics)
List of Participants
A list of registered participants will be available later.
Code of conduct
The meeting has a code of conduct: Meeting Code of Conduct.pdf
STFC claim form
For STFC and selffunded students, travel and extra subsistence can be claimed using STFC Claim form.pdf. Rerturn to Studentships, STFC, Polaris House, North Star Avenue, Swindon SN2 1SZ
Practical info
Travel
The Huxley Building is very close to the Royal Albert Hall in South Kensington. You enter Huxley (No. 13 on the map) from the Queen's Gate road side. Nearest tube stops are South Kensington and Gloucester Road (10 mins walk), and there are buses which pass close by  see the map and the TfL website for details ( http://www.tfl.gov.uk/ ). You may find the Journey Planner facility useful. Note that it is worth buying a payasyougo Oyster card if you are going to use the system at all  the fares are much cheaper than buying individual tickets. You can also use a (UK only) contactless payment card, which has the same fares as Oyster.
From South Kensington Underground station: if it's raining, when you come through the barriers, you can turn right below ground and take the long tunnel from the station, which emerges pretty much at the bottom of the main map (attached). Otherwise, it's preferable to go straight ahead and head up for the daylight, exit to street level immediately, turning right to get into Thurloe St (this is mostly pedestrianised now)  it's a more pleasant walk above ground past the museums.
Handouts:
Course materials will appear here.
Preliminary Exercise:
Please do this BEFORE the workshop!
Instructions and background: Preliminary Exercise 2016.pdf
Supernova data file: jla_mub.txt
Covariance matrix: jla_mub_covmatrix.txt
Workshop handouts:
Programme and logistics: Welcome Pack a.pdf
Day 1 exercises: HandsOn Day 1 2016.pdf
Day 1 solutions: HandsOn Day 1 solutions 2016.pdf
Day 1 lecture  Alan Heavens: ICIC Workshop Lecture 1 2016 final.pdf
Day 1 lecture  Jonathan Pritchard: Marginalisation_JRP.pdf
Day 2: SN MCMC exercise instructions (updated from paper copy): SN MCMC project 2016.pdf
Day 2: GelmanRubin formula: GelmanRubin_fromSAS_STAT_usersguide.pdf
Day 2 lecture  Daniel Mortlock: Metropolis MCMC notes metropolis_DM.pdf
Day 2 lecture  Alan Heavens: CLT and the Lighthouse problem (also a better answer to Day 1 problem) CLT and Lighthouse Problem.pdf
Day 3 lecture  Andrew Jaffe: Gibbs, HMC, Linear Models Gibbs_HMC_GLM_AJ.pdf
Day 3 lecture  Elena Sellentin: Model comparison ModelSelection_ES.pdf
Day 4  Jonathan Pritchard: Nested sampling NestedSampling_JRP.pdf
Day 4  Elena Sellentin: Frequentist and Bayesian NoiseVsSignal_ES.pdf
Day 4  Alan Heavens: Bayesian Hierarchical Models BHM_AH.pdf
Day 4  Alan Heavens: SDDR slide SDDR slide.pdf
Day 4  Roberto Trotta: Public Engagement Lunch PE_Lunch RT.pdf
Python code
Alan's python 3 code for the 2parameter case: SNcodePython.txt
Inverse covariance matrix: Inverse covariance.txt
Updated programme: ICIC 2016 Programme.pdf