benching elkan k-means implementation

elkan_bench.py

from sklearn.cluster import KMeans
from time import time
from sklearn.datasets import load_digits, fetch_mldata, load_iris, fetch_20newsgroups_vectorized

def bench_kmeans(data, n_clusters=5, init='random', n_init=1):
    start = time()
    km1 = KMeans(algorithm='lloyd', n_clusters=n_clusters, random_state=0, init=init, n_init=n_init).fit(X)
    print("lloyd time: %f inertia: %f" % (time() - start, km1.inertia_))
    start = time()
    km2 = KMeans(algorithm='elkan', n_clusters=n_clusters, random_state=0, init=init, n_init=n_init).fit(X)
    print("elkan time: %3.4f inertia: %f" % (time() - start, km2.inertia_))


mnist = fetch_mldata("MNIST original")
digits = load_digits()
iris = load_iris()

for n_clusters in [3, 10, 100]:
    print("\nn_clusters = %d" % n_clusters)
    for dataset, X in [("digits", digits.data), ("iris", iris.data), ("mnist", mnist.data), ("mnist.T", mnist.data.T)]:
        print(dataset)
        bench_kmeans(X, init='random', n_clusters=n_clusters)