ECC Multivariate Autoregressive model with Exogeneous Inputs

This is the companion repository to a paper accepted to the 2025 European Control Conference entitled

Online Bayesian system identification in multivariate autoregressive models via message passing

An arXiv version can be found here.

This repository contains the code associated with our submitted paper. Please note that the code base is currently undergoing cleanup and improvements. We are actively working to resolve known issues and improve reusability. Updates will be pushed in the coming days.

How do I run your experiments?

To install the required Julia packages, run ./install.sh (OS X and Linux) or julia --project=. -e 'using Pkg; Pkg.activate("."); Pkg.instantiate()' (Windows).

To run the experiments, run run-experiments.sh (OS X and Linux) or JULIA_NUM_THREADS=4 julia --project=. experiments-MARX.jl and JULIA_NUM_THREADS=$NUM_THREADS julia --project=. experiments-dmsds.jl (Windows). Results will be stored in a directory called results.

How do I inspect the results of your experiments?

We use Pluto notebooks to interactively inspect our models.

Follow the instructions in the previous section to install required Julia packages. Currently, the results of our experiments take up considerable storage space (11 GB for one environment and the chosen three models; MARX-UI, MARX-WI, RLS). We are actively working to minimize the storage space required for each monte carlo experiment. This means you have to run the experiments (~10 minutes per model and environment) before you can inspect the results stored in the results directory.

Then, run ./run-pluto.sh (OS X and Linux) or julia --project=. -e 'using Pluto; Pluto.run()' (Windows). A new browser tab should open with the url http://localhost:1234/.

There are two Pluto notebooks that let you inspect the experiments:

• inspect-MARX.jl lets you inspect the results for the MARX system
• inspect-dmsds.jl lets you inspect the results for the double mass-spring-damper system

How to I give feedback?

Please submit an issue.