This repository provides source code for the paper DecoR: Deconfounding Time Series with Robust Regression. In particular, the robust-deconfounding package implements robust deconfounding method DecoR and the robust linear regression methods Torrent and BFS.

The experiments directory holds code for synthetic experiments and provides the necessary scripts to reproduce the experimental results reported in the paper.

To install the minimal dependencies needed to use the meta-learning algorithms, run in the main directory of this repository

pip install -e .

For full support of all scripts in the repository, for instance to reproduce the experiments, further dependencies need to be installed. To do so, please run in the main directory of this repository

pip install -r requirements.txt

The following code snippet demonstrates the core functionality of the meta-learners provided in this repository. The code in demo.ipynb demonstrates the core functionality of the DecoR provided in this repository.

Below we point to the experiment scripts that were used to generate the results reported in the paper.

To run the experiments:

python experiments/experiments.py

To run the ablation experiments:

python experiments/experiments_ablation.py

You can adjust method number of datapoints and the data generating process in the files themselves. Note that the experiments may take a long time to run, depending on the number of repetitions and the size of the datasets especially if BFS is used.