import numpy as np
import time
from mne.stats import permutation_cluster_1samp_test

def large_samples_perm(n_perm=100, n_space=100, n_time=100):
    noise_level = 20
    n_time_1 = n_time
    n_time_2 = n_time
    normfactor = np.hanning(20).sum()
    rng = np.random.RandomState(42)
    condition1_1d = rng.randn(n_time_1, n_space) * noise_level
    for c in condition1_1d:
        c[:] = np.convolve(c, np.hanning(20), mode="same") / normfactor
    pseudoekp = 10 * np.hanning(round(n_space / 2))[None, :]
    condition1_1d[:, round(n_space / 2):] += pseudoekp

    thresh_tfce = dict(start=0, step=0.2)
    permutation_cluster_1samp_test(condition1_1d, threshold=thresh_tfce,
                                   n_permutations=n_perm, tail=0, seed=1,
                                   buffer_size=None)

%timeit large_samples_perm()
# master: 436 ms ± 13.6 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
# opt: 413 ms ± 2.04 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

%timeit large_samples_perm(n_perm=1000, n_space=100, n_time=100)
# master: 3.85 s ± 13.6 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
# opt: 3.75 s ± 6.31 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

%timeit large_samples_perm(n_perm=1000, n_space=300, n_time=300)
# master: 5.83 s ± 22.1 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)
# opt: 5.73 s ± 31 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

t0 = time.time()
large_samples_perm(n_perm=1000, n_space=1000, n_time=1000)
t1 = time.time()
elapsed = t1 - t0
# master: 17.224292039871216
# opt: 16.965471029281616