EDDI

EDDI: Explainable Drift Detection using Influence

Installation Instructions

Setting up the virtual environment and packages

Create a virtual environment using conda with Python 3.10.

conda create -n eddi python=3.10
conda activate eddi

Install poetry for dependency management and conflict resolution. To download poetry into a specific folder run

curl -sSL https://install.python-poetry.org | POETRY_HOME=/hri/storage/user/<USER>/etc/my_poetry python3 -

To install the necessary Python libs, go to the project's main directory, i.e., where the .toml file is on the same dir level and run

poetry install

Download the datasets

python download_ts_classification.py 

Train the models. For example to train TST run:

python models/model_dev.py --run_only_on datasets/Symbols

To simulate all drift types run: