import pandas as pd from datasets import load_dataset data = load_dataset('relbert/semeval2012_relational_similarity_v4') stats = [] for k in data.keys(): for i in data[k]: stats.append( { 'relation_type': i['relation_type'], 'split': k, 'positives': len(i['positives']), 'negatives': len(i['negatives']), 'level': i['level'] }) df = pd.DataFrame(stats) g = df.groupby(['relation_type', 'level', 'split']).sum() g.to_csv('stats.csv') with open('stats.md', 'w') as f: f.write(g.to_markdown())