This repository contains experiments and derivations for the paper entitled

"Efficient Model Evidence Computation in Tree-structured Factor Graphs".

Setting up

Before implementing the experiments, we need to initialize an environment in Julia. This can be done by the following steps:

• In a terminal, navigate to the location where you store the repository after cloning
• type julia
• type using Pkg, or ]
• type Pkg.activate("."), or activate . if we use ] in the previous step. If you clone the repository and keep its name, you should see (SiPS2022-EfficientModeEvidenceComputation) pkg> in the terminal when you press ].

Now you can instantiate the project by Pkg.instantiate(), or instantiate if you press ]. This will install all necessary packages for the experiments.

Experiments

The repository contains 3 experiments located in 3 seperate files Coin_toss.ipynb, HMM.ipynb and LGSSM.ipynb. The experiments can be implemented by executing every code block in the corresponding files.

Supplementary document

We also include a supplement document sips2022_scalefactor_supplement.pdf which contains the derivation for all scale factor update rules in the paper.