mcmc-pymc-101

[smu095]

Todo-list

The following todo-list is meant to give an overview over the most important tasks going forward. Leave a comment if you want to be assigned to a particular task or if you feel others should be added, and I will update the issue accordingly.

Date for presentation: 17.02.2020

Presentation

• [X] Decide on format for slides. I suggest using LaTeX as it is easy to keep under source control.

• [x] Delegate responsibilities for parts 1 and 2 of presentation.

  • [x] Part 1a: Theory (Markov chains, Monte Carlo methods and Markov Chain Monte Carlo) [@smu095 @solveigm].

  • [x] Part 1b: A couple of extra slides on priors. With example (linear regression?) [@solveigm]

  • [x] Part 1b: "But how do I calculate the likelihood?" Probability density function. (By Wednesday: [@smu095 @solveigm])

  • [ ] Part 2: Practice (PyMC3 workshop). Explain API and usage as a part of the example [Undetermined].

• [ ] Add slide motivating use of Bayes in business

Presentation order

• @smu095 : 1-5, 12-24, 27 --> + example 2 (Golf)
• @solveigm : 6-11, 25-26 + example 1 (AR(1))

Practical examples

• [x] By Wednesday, agree on:
  • [x] Find a relevant use case where uncertainty/transparency is important? I.e. where black box methods are not suitable/preferable [@solveigm].
  • [x] Maybe try to find a neat example where we minimise an expected loss function? [@smu095]

Setup

• [X] Write brief setup description and add to README.md.
• [x] Enable master branch protection.
• [X] Install necessary packages and pip freeze > requirements.txt for reproducibility.
• [x] Add .pdf to .gitignore [@smu095]

[smu095]

I'd like to take responsibility for part 1a, concerning the theory of MCMC. My plan is to write a brief overview that can supplement the presentation for those who might be interested in delving deeper.