import numpy as np
from sklearn.cluster import KMeans
from sklearn.metrics import adjusted_rand_score

def stabilite(X, k, paires=12, graine=7):
    r = np.random.default_rng(graine)
    aris = []
    for _ in range(paires):
        A = r.choice(len(X), int(0.8 * len(X)), replace=False)
        B = r.choice(len(X), int(0.8 * len(X)), replace=False)
        ka = KMeans(k, n_init=3, random_state=int(r.integers(10**6))).fit(X[A])
        kb = KMeans(k, n_init=3, random_state=int(r.integers(10**6))).fit(X[B])
        commun = np.intersect1d(A, B)               # clients présents dans les deux
        aris.append(adjusted_rand_score(ka.predict(X[commun]), kb.predict(X[commun])))
    return np.mean(aris), np.min(aris)