import json
import os
from datasets import load_dataset


def process(split, output):
    data = load_dataset("relbert/nell", split=split)
    df = data.to_pandas()
    rel_sim_data = [{
        "relation_type": pred,
        "positives": [list(y) for y in zip(g['head'], g['tail'])],
        "negatives": []
    } for pred, g in df.groupby("relation") if len(g) >= 2]
    with open(output, "w") as f_writer:
        f_writer.write('\n'.join([json.dumps(i) for i in rel_sim_data]))


if __name__ == '__main__':
    os.makedirs("data", exist_ok=True)
    for s in ['train', 'validation', 'test']:
        process(split=s, output=f"data/{s}.jsonl")