A Monte Carlo Murder Mystery: A Holmes-Watson-PyMC3 adventure!9:10am - 9:40am on Friday, October 2 in Online
(The following is intended to be in the style of a movie poster for a 1920’s silent thriller.)
On a beautiful summer’s eve in the coast of Monte Carlo, Count Frick Von Twist is found dead in his luxurious hotel suite on the heels of a dinner party celebrating his 90th birthday. In shock and anguish, his adopted daughter, Viscountess Bae Zhun, calls upon famed detective Shellakybooky Holmes and his trusted associate Jahnvi Watson to deduce her father’s killer.
A meticulous Watson interviews the eleven dinner guests separately and examines the Count’s pet python to find probable cause. Holmes quietly enters his Hamiltonian Mind Castle (HMC) to run simulations of possible versions of the fated evening’s events and derive posterior distributions of the probability of each of the guests’ murderous intent. Together, will they find the Count’s killer with a degree of confidence that would convince a jury? Will Bae Zhun find peace, convergence and justice amidst all the uncertainty? Tune in on October 2nd at 9:30 p.m. EST on PyGotham TV to watch: A Monte Carlo Murder Mystery!*</b>
*A PyMC3 production directed by an ardent pythonista (who might have read one too many detective novels) with a little help from friends!
Motivation behind this Proposal: How to Solve a Murder using Bayesian Inference!
For some time now, I have been very interested in giving an introductory talk on Probabilistic Programming and Bayesian Data Analysis that is geared towards a general audience of Python programmers. There are, however, a series of intimidating topics (such as likelihood functions, posterior distributions, Monte Carlo simulations, Generalized Linear Models, etc) en route to understanding Bayesian methods in programming that can often dampen the spirit of a curious learner. These topics have been covered wonderfully well by experts on the subject (Hugo Bowne-Andersen, Cameron Davidson-Pilon and Mitzi Morris to name a few) over many hours of conference tutorials and courses. Deriving inspiration from these as well as drawing upon PyGotham TV 2020’s encouragement to experiment with the television theme, I propose to give a high level overview of these concepts in an accessible and entertaining manner using the medium of a silent film with a murder mystery plotline.
To this end, I would like to rope in one of my favorite fictional detectives who also happens to be a true Bayesian**, Sherlock Holmes, to take us through the steps involved in using Bayesian methods in answering a question about a dataset (here, for instance, “Who is the murderer?”) and in updating the inference in the light of new information. The classical detective novel arc gives a very tangible parallel to making deductions via Bayesian methods and the visual medium offers exciting possibilities in terms of conveying complex concepts (detailed further in the Outline), making the combination very effective for the purpose of reaching a broad audience.
The talk will thus be in the form of a 1920’s style silent movie (not unlike another Conan Doyle adaptation such as this) interspersed with relevant infographics. There will be a small amount of narration/voiceover wherever it may be necessary to convey the plot or underlying concepts. The amount of code and math depicted on screen will be minimal and an accompanying Github repo will in turn cover that in further depth. The repo will contain: the dataset and problem statement; a Jupyter notebook (or two) that demonstrates how Holmes arrives at the conclusion; a requirements.txt file for relevant installations; and sufficient resources and corollary questions that the enterprising audience can play around with.
**Some quotes from that illustrate Holmes’ Bayesian thinking :
“How often have I said to you that when you have eliminated the impossible, whatever remains, however improbable, must be the truth?” (The Sign of the Four)
“… we balance probabilities and choose the most likely. It is the scientific use of the imagination … ” (The Hound of the Baskervilles)
“Data! Data! Data!” he cried impatiently, “I can’t make bricks without clay.” (The Adventure of the Copper Beeches)