def ng_dist(arr, supergaussian):
NN = arr.size
nong_mask = ng.nongaussian_filter(arr)
lmean = arr.mean()
lvar = arr.var()
gauss_resid = arr[~nong_mask] - lmean
nongauss_resid = arr[nong_mask] - lmean
gauss_kurt = np.sum(gauss_resid**4) / (lvar**2 * NN)
nongauss_kurt = np.sum(nongauss_resid**4) / (lvar**2 * NN)
true_kurt = stats.kurtosis(arr)
assert_almost_equal(gauss_kurt + nongauss_kurt - 3.0, true_kurt, places=6)
if supergaussian:
assert_almost_equal(gauss_kurt - 3.0, 0.0, places=2)
assert_almost_equal(nongauss_kurt, true_kurt, places=2)
else:
assert_almost_equal(gauss_kurt - 3.0, true_kurt, places=2)
assert_almost_equal(nongauss_kurt, 0.0, places=2)
def test_ng_zeros():
all_zeros = np.zeros((100,100))
nong_mask = ng.nongaussian_filter(all_zeros)
eq_(nong_mask.size, all_zeros.size)
ok_(not np.any(nong_mask))
