import pandas as pd #just compress Anton Petrov's code in 6 lines
d1 = pd.read_csv('../input/tuning_labels.csv', names=['id', 'labels'], index_col=['id'])
d2 = pd.read_csv('../input/stage_1_sample_submission.csv', index_col='image_id')
d2['labels'] = ' '.join(d1['labels'].str.split().apply(pd.Series).stack().value_counts().head(3).index.tolist())
d2.update(d1)
d2.to_csv('last_day_6_lines_baseline.csv')