1. Clone the repo
   
2. Create a conda environment conda create -n SUNAR python=3.10.3
   conda activate SUNAR
3. pip install -e .
   

curl https://gitlab.tudelft.nl/anonymous_arr/bcqa_data/-/raw/main/2wikimultihopQA.zip -o 2wikimultihopQA.zip

To download the Document graph of the corpus employed in neighborhood adaptive retrieval for MusiqueQa and WikimultihopQA use this link https://drive.google.com/drive/folders/1zyWtCyhQzxaMQpM6uXT5oqvtqFMg7nYl?usp=sharing

If you want to create your own corpus graph run

python evaluation/form_corpus_graph.py

Download wiki_musique_corpus.json from https://drive.google.com/drive/folders/1zyWtCyhQzxaMQpM6uXT5oqvtqFMg7nYl?usp=drive_link and save it in data folder

If you do not want to run splade download wqa_splade_docs.json from https://drive.google.com/drive/folders/1zyWtCyhQzxaMQpM6uXT5oqvtqFMg7nYl?usp=drive_link and drop it inside data/intermediate_outputs/

In evaluation/config.ini configure the corresponding paths to downloaded files configure project root directory to PYTHONPATH variable

Additionally set the following environment variables in the terminal

export PYTHONPATH=/path

export OPENAI_KEY=<your openai key>

export huggingface_token = <your huggingface token to access models  >

To run first-stage retrieval run

python evaluation/wikimultihop/run_splade_inference.py

To directly reproduce the best results in paper (Searchain+SUNAR) in Table 2 run

python evaluation/wikimultihop/llms/run_searchain_sunar.py

Note the above script runs using evidences saved froma run of SUNAR algorithm and does inference to enable easy reproduction of the results in paper

If interested in running SUNAR end-end for reproducing SUNAR numbers in Table 1 run

python evaluation/wikimultihop/llms/run_sunar.py