PD Échantillonner chaque classe
df.groupby('classes').apply(lambda x: x.sample(sample_size))
Attractive Alpaca
df.groupby('classes').apply(lambda x: x.sample(sample_size))