benchmarking sklearn.metrics.precision_recall_fscore_support

gistfile1.txt

import numpy as np, timeit, sklearn.metrics, sklearn.preprocessing, functools
from joblib import Memory
N_TRIALS = 50

memory = Memory('/tmp', verbose=0)

@memory.cache
def gen_mc(s, N=1000000, K=10):
    return np.random.randint(K, size=N)

@memory.cache
def gen_ml(s, N=1000, K=10):
    r = np.arange(K)
    return [r[np.random.randint(2, size=K).astype(bool)].tolist() for i in xrange(N)]

@memory.cache
def gen_ml_li(s, N=1000000, K=10):
    lb = sklearn.preprocessing.LabelBinarizer().fit([list(range(K))])
    return lb.transform(gen_ml(N, K))

def go(gen):
    return timeit.timeit(functools.partial(sklearn.metrics.f1_score, gen('true'), gen('pred')), number=N_TRIALS)

for gen in [gen_mc, gen_ml, gen_ml_li]:
    print '%0.3g' % (go(gen) / N_TRIALS)