Bachelor's Thesis in Explainable AI

This repository comprises all dependencies of Dieterich Lab Heidelberg for the Bachelor's thesis in Computational Linguistics (B.A., 2023) at Heidelberg University, titled:

"Interpretability in NLP

A comparative analysis of two interpretability methods on medical text in light of the faithfulness metric"

🔍 Contents

• Thesis presentation at Dieterich Lab
• Notebooks
  • Faithfulness - Sufficiency
  • Faithfulness - Comprehensiveness
  • Training Data Statistics
• Supplementary materials
  • Python code for training and testing BERT model
  • Test Results as plain text
  • Bash script for running GPU intensive job on Dieterich Lab SLURM cluster

To run the Faithfulness notebooks, please beforehand contact the author to obtain the necessary confidential data. After cloning and changing directory to the repository, you may then proceed as follows:

🧰 Prerequisites

1. Create and activate a Python virtual environment: python3 -m venv xai
2. Install all required packages in requirements.txt: pip install -r requirements.txt
3. Create the according IPython kernel for Jupyter: python3 -m ipykernel install --user --name=xai

⚙️ Run Faithfulness Notebooks

1. Save the obtained confidential pickle data in the repository folder structure
2. Open Jupyter Notebook: jupyter notebook and select the created kernel xai
3. Both Comprehensiveness & Sufficiency notebooks can be executed

✏️ Maintenance

Author: Raziye Sari - sari@cl.uni-heidelberg.de
Last updated: June 20th 2023