def test_statistics_1():
a, w = get_data()
_means = np.zeros((3, 1, 1, 6))
_weights = np.zeros((3, 1, 1, 6))
for i, j, k, l in itertools.product(range(a.shape[0]), range(a.shape[1]),
range(a.shape[2]), range(a.shape[3])):
_means[i, 0, 0, l] += w[i, j, k, l] * a[i, j, k, l]
_weights[i, 0, 0, l] += w[i, j, k, l]
_means /= _weights
np.testing.assert_allclose(_means,
mean(a, w, axis=(1, 2), keepdims=True),
rtol=1e-07)
_stds = np.zeros((3, 1, 1, 6))
for i, j, k, l in itertools.product(range(a.shape[0]), range(a.shape[1]),
range(a.shape[2]), range(a.shape[3])):
_stds[i, 0, 0,
l] += w[i, j, k, l] * (a[i, j, k, l] - _means[i, 0, 0, l])**2
_stds = np.sqrt(_stds / _weights)
np.testing.assert_allclose(_stds,
std(a, w, axis=(1, 2), keepdims=True),
rtol=1e-07)
_values = np.zeros((3, 6, 4 * 5))
_weights = np.zeros((3, 6, 4 * 5))
for i, j, k, l in itertools.product(range(a.shape[0]), range(a.shape[1]),
range(a.shape[2]), range(a.shape[3])):
_values[i, l, j * a.shape[2] + k] = a[i, j, k, l]
_weights[i, l, j * a.shape[2] + k] = w[i, j, k, l]
def get_quantiles(q):
_quantiles = np.zeros((3, 6))
for i in range(a.shape[0]):
for l in range(a.shape[3]):
isort = np.argsort(_values[i, l])
v = _values[i, l][isort]
u = _weights[i, l][isort]
U = u.cumsum()
r = (U - 0.5 * u) / U[-1]
for m in range(1, len(u)):
if r[m - 1] <= q and r[m] > q:
_quantiles[i, l] = v[m - 1] + (q - r[m - 1]) / (
r[m] - r[m - 1]) * (v[m] - v[m - 1])
break
return _quantiles
np.testing.assert_allclose(get_quantiles(0.1),
quantile(a, 0.1, w, axis=(1, 2),
keepdims=False),
rtol=1e-07)
np.testing.assert_allclose(get_quantiles(0.5),
quantile(a, 0.5, w, axis=(1, 2),
keepdims=False),
rtol=1e-07)
np.testing.assert_allclose(get_quantiles(0.9),
quantile(a, 0.9, w, axis=(1, 2),
keepdims=False),
rtol=1e-07)
def test_mean_shapes():
a, w = get_data()
out = mean(a)
assert out.ndim == 0
out = mean(a, w)
assert out.ndim == 0
out = mean(a, keepdims=True)
assert out.shape == (1, 1, 1, 1)
np.testing.assert_allclose(out.ravel()[0], mean(a), rtol=1e-07)
out = mean(a, w, keepdims=True)
assert out.shape == (1, 1, 1, 1)
np.testing.assert_allclose(out.ravel()[0], mean(a, w), rtol=1e-07)
out = mean(a, axis=1)
assert out.shape == (3, 5, 6)
out = mean(a, w, axis=1)
assert out.shape == (3, 5, 6)
out = mean(a, axis=(1, 2))
assert out.shape == (3, 6)
np.testing.assert_allclose(out,
mean(a, np.ones_like(a), axis=(1, 2)),
rtol=1e-07)
out = mean(a, w, axis=(1, 2))
assert out.shape == (3, 6)
out = mean(a, axis=(1, 2), keepdims=True)
assert out.shape == (3, 1, 1, 6)
np.testing.assert_allclose(out.ravel(),
mean(a, axis=(1, 2)).ravel(),
rtol=1e-07)
out = mean(a, w, axis=(1, 2), keepdims=True)
assert out.shape == (3, 1, 1, 6)
np.testing.assert_allclose(out.ravel(),
mean(a, w, axis=(1, 2)).ravel(),
rtol=1e-07)
