Practical workshop on Bayesian Hierarchical Models

Agenda

1. Theorie

   1. Machine Learning vs. Statistische Modellierung
   2. Grundlagen der Bayesschen Modellierung
   3. Pymc3 – Ein probabilistisches Programmierframework

2. Praxis

   1. Unsicherheit beherrschen: Bayessche Lineare Modelle
   2. Informationen teilen: Bayessche Hierarchische Modelle

Spin up the environment

Install docker on your environment following these instructions.

After having installed docker, build the image as follows (this needs to be done only once):

docker build -t bayes_workshop .

Run the container with the following command:

(NOTE: The following command was tested on osx and linux. For docker on windows, this can be slightly different)

docker run -p 8888:8888 \
           -v $(PWD)/2_praxis/:/home/jovyan/praxis/ \
           -v $(PWD)/data/:/home/jovyan/data/ \
           -v $(PWD)/utils/:/home/jovyan/utils/ \
           bayes_workshop

And then just copy paste the http://127.... (with token) into your browser.

If you want to add additional packages you can do so in two ways:

• Add them permanently by adding it to the requirements.txt.
• Install them in the jupyter notebook by executing !pip install <YOUR PACKAGE> in a cell.

TODOs

• Theorie: Besseres intuitives Beispiel wählen.
• Theorie: Ausbauen von Pymc3 Teil.
• Theorie: Einbauen, wann bayes modellierung besonders sinnvoll ist (ML vs. Stat Modeling).
• Theorie: Übergang von P(H|D) zu real-world Modell glatter machen.
• Praxis: Teil I und II in verschiedenen Sessions.
• Praxis: Teil 0 hinzufügen: Sampling from posterior from scratch (Sampling Idee kennenlernen).