sabrinacalcina/risk-ai-workshop

Lecture material and sample code for the workshop "Risk, Artificial Intelligence and Discrete Geometry" at the University of Ljubljana

Risk, Artificial Intelligence and Discrete Geometry

Quick-start

• git clone the repository and cd into the root directory.
• For the exercise Jupyter notebooks
  • create a virtual environment (recommended) e.g. with venv, conda or other.
  • install python dependencies with pip install -r requirements.txt in a virtual environment. Note: the fake-data-for-learning package has some non-python dependencies; see its installation instructions.

Workshop topics

Slides are built as part of the repo's ci-cd pipeline, and can be accessed by clicking on the Artifacts section of a workflow run, as in the below screenshot

Artificial intelligence for risk management

Examples and exercises

Discrete geometry for risk

Examples and exercises: graphical models, probability polytopes, and simpson's paradox

Correlation and causality

Examples and exercises

Risk and Artificial Intelligence in practice

• Model Selection Examples
• Risk and AI in Practice Exercises

Relate python packages

In the exercise notebooks and requirements.txt you see which python packages I used in creating and solving the exercises, though this list is far from exhaustive. Below are some (additional) python packages that may be useful

Graph visualization

• graphviz
• networkx, plus many graph operations

Bayesian networks, causal inference

• pgmpy
• brent
• causalgraphicalmodels
• causality
• dowhy
• pyro

Releases

I follow a semantic-versioning-like convention for releases of <workshop-year>.<minor>.<patch>. For now, the minor value will be an incrementing integer; as the module being developed in this repo is not planned to be pushed to pypi, I won't be very strict, and will likely keep minor at 0 even if the api changes. If it looks like I would ever give more than one workshop a year, the minor value could be used for the month of the workshop.

Process of creating a new workshop release

If the release is for a new workshop year, then first manually change the version in the code-base to <new-year>.0.0. This release need not be a tagged release, as it is the same as the final release of the previous workshop in an earlier year.

Once an initial release has been created for a new workshop, create subsequent tagged releases by using bump2version.

Release history

Note: I do not follow Semantic Versioning for this project. For the first digit (in semver, major), I use the year of the target workshop, and for the last (in semver patch), I increment when a chunk of work is done towards giving the workshop. The middle digit (in semver, minor) stays on 0 until I give the workshop, when it bumps to 1. Fixes to the given workshop get reflected in the patch versions yyyy.1.<patch-version>.

Latest

Fix typos in discrete geometry slides, including definition of d-separation

v2022.1.0

Refactor introduction and concluding lectures, as well as their exercises

Add simpson's paradox content and exercises to the discrete geometry lecture

Add ci tests of package and notebooks for mac and windows (latest) operating systems

v2022.0.2

Add github actions workflow for automated testing, with unit, notebook-run and latex-slides-build

Add github actions job for continuous delivery of slide artifacts

Add example of model selection pipeline for artificial credit scoring data

v2022.0.1

Add initial exercises and methods for Simpson's paradox.

v2020-02-uni-lj

Created prior to the above versioning scheme, workshop at University of Ljublana in February, 2020.