SELF

Our code is based on the feature selection framework ERASE, please put the selfs.py under models/fs path

Datasets

You can download the datasets from movielens, aliccp, kuairand

Running

1. Feature Importance Extraction stage

python predict.py --model=gpt-4-turbo --dataset=movielens

2. Feature Importance Refinement stage

python fs_run.py --dataset=movielens-1m --fs=selfs --retrain=false

3. Retraining stage

python llm_selection.py
python fs_run.py --dataset=movielens-1m --model=dcn --train_or_search=false --rank_path=selfs_movielens-1m --k={k}