Certified Error Control of Candidate Set Pruning for Two-Stage Relevance Ranking (EMNLP 2022 Oral).
pip install -r requirement.txt
You also need to install torch according to your configuration. This project is also based on the RCPS library.
wget https://vault.cs.uwaterloo.ca/s/8H8gSLFMRSMdz8L/download -O msmarco.zip
unzip msmarco.zip
wget https://vault.cs.uwaterloo.ca/s/WBQ4F6DnHcjmc2g/download -O quora.zip
unzip quora.zip
After downloading and decompressing the data, you could run the following code to calibrate and test your candidate set pruning method.
bash cec_prediction.sh