def test_linregress_ufunc():
y = np.random.random(20)
x = np.arange(len(y))
fit = np.array(stats.linregress(x, y))
assert np.allclose(fit, _linregress_ufunc(x, y))
# test with nans
y[0] = np.nan
fit = np.array(stats.linregress(x[1:], y[1:]))
assert np.isnan(_linregress_ufunc(x, y)).all()
assert np.allclose(fit, _linregress_ufunc(x, y, nanmask=True))
# test with all nansum
y[:] = np.nan
assert np.isnan(_linregress_ufunc(x, y, nanmask=True)).all()
def test_xr_linregress(chunks, variant, dtype, nans):
a = xr.DataArray(np.random.rand(3, 13, 5), dims=["x", "time", "y"])
b = xr.DataArray(np.random.rand(3, 5, 13), dims=["x", "y", "time"])
if nans:
# add nans at random positions
a.data[np.unravel_index(np.random.randint(0, 2 * 4 * 12, 10),
a.shape)] = np.nan
b.data[np.unravel_index(np.random.randint(0, 2 * 4 * 12, 10),
b.shape)] = np.nan
if chunks is not None:
if variant == 0:
a = a.chunk(chunks)
elif variant == 1:
b = b.chunk(chunks)
elif variant == 2:
a = a.chunk(chunks)
b = b.chunk(chunks)
reg = xr_linregress(a, b, dtype=dtype)
for xx in range(len(a.x)):
for yy in range(len(a.y)):
pos = dict(x=xx, y=yy)
expected = _linregress_ufunc(a.isel(**pos), b.isel(**pos))
reg_sub = reg.isel(**pos)
for ni, nn in enumerate(
["slope", "intercept", "r_value", "p_value", "std_err"]):
np.testing.assert_allclose(reg_sub[nn].data, expected[ni])
def test_lin_trend_full_legacy():
# This is meant to be a test if the old ufunc for the trend produces
# identical output
y = np.random.random(20)
x = np.arange(len(y))
fit = _linregress_ufunc(x, y)[0:2]
fit_legacy = _lin_trend_legacy(y)
assert np.allclose(fit, fit_legacy)
def test_linregress_ufunc():
y = np.random.random(20)
x = np.arange(len(y))
fit = np.array(stats.linregress(x, y))
assert np.allclose(fit, _linregress_ufunc(x, y))