ReLiNet: Stable and Explainable Multistep Prediction with Recurrent Linear Parameter Varying Networks

This repository contains the necessary scripts to reproduce the results from our paper "ReLiNet: Stable and Explainable Multistep Prediction with Recurrent Linear Parameter Varying Networks".

Clone the repository and move into its directory. Install all dependencies with

pip install .

Make sure to use your preferred virtual environment.

Run the following to download all datasets and set up the required directories:

python scripts/setup_environment.py

All directories and files will be created within the cloned directory.

To run the experiments for the ship dataset run the following two scripts in order:

python scripts/run_experiment_ship_ind.py {device}
python scripts/run_experiment_ship_ood.py {device}
python scripts/explain_best_models_ship_ind.py {device}
python scripts/explain_best_models_ship_ood.py {device}

where device is the identifier (an integer starting at 0) for the GPU to run the experiments on. If you only have one GPU, set the value to 0.

If these scripts are stopped for any reason, you can rerun them without issue. run_experiment_ship_ind.py remembers what models where already trained and validated.

To run the experiments for the industrial robot dataset run the following script:

python scripts/run_experiment_industrial_robot.py {device}
python scripts/explain_best_models_industrial_robot.py {device}

Trained models are found in models, results in results, and datasets in datasets. Environment variables pointing to the models, results, and configuration for each experiment are found in environment.

Finally, to summarize the results in tables run:

python scripts/summarize_results.py

You will find CSV files summarizing the results in results/{dataset_name}, where dataset_name corresponds to the SHIP-IND, SHIP-OOD, and ROBOT datasets as described in the paper.

Hyperparameter choices for gridsearch are documented in the directory configuration.