def test_weighted():
np.random.seed(1234)
t = np.arange(100)
y0 = (+0.4 * (t >= 5) + 0.9 * (t >= 10) - 0.2 * (t >= 20) + 0.2 *
(t >= 50) + 1.1 * (t >= 70))
y = y0 + 0.05 * np.random.rand(y0.size)
y = y.tolist()
w = [1] * len(y)
y[15] = 2
right, values, dists = solve_potts(y, w=w, gamma=0.1)
assert right == [5, 10, 15, 16, 20, 50, 70, 100]
steps = detect_steps(y, w=w)
steps_pos = [s[0] for s in steps]
assert steps_pos == [0, 5, 10, 15, 16, 20, 50, 70]
w[15] = 0.1
right, values, dists = solve_potts(y, w=w, gamma=0.1)
assert right == [5, 10, 20, 50, 70, 100]
steps = detect_steps(y, w=w)
steps_pos = [s[0] for s in steps]
assert steps_pos == [0, 5, 10, 20, 50, 70]
# Missing weights and data should be handled properly
y[35] = None
w[23] = None
steps = detect_steps(y, w=w)
steps_pos = [s[0] for s in steps]
assert steps_pos == [0, 5, 10, 20, 50, 70]
def test_zero_weight():
t = list(range(50))
y = [1.0 * (tt > 25) for tt in t]
w = [0 if tt >= 20 and tt < 30 else 1 for tt in t]
# Due to zero weight, part of the data is not assigned to belong to a step
steps = detect_steps(y, w=w)
steps_pos = [s[:2] for s in steps]
assert steps_pos == [(0, 20), (30, 50)]
# A small weight allows to find the actual step
w = [1e-6 if ww == 0 else 1 for ww in w]
steps = detect_steps(y, w=w)
steps_pos = [s[:2] for s in steps]
assert steps_pos == [(0, 26), (26, 50)]
def test_weighted():
np.random.seed(1234)
t = np.arange(100)
y0 = (+ 0.4 * (t >= 5)
+ 0.9 * (t >= 10)
- 0.2 * (t >= 20)
+ 0.2 * (t >= 50)
+ 1.1 * (t >= 70))
y = y0 + 0.05 * np.random.rand(y0.size)
y = y.tolist()
w = [1]*len(y)
y[15] = 2
right, values, dists = solve_potts(y, w=w, gamma=0.1)
assert right == [5, 10, 15, 16, 20, 50, 70, 100]
steps = detect_steps(y, w=w)
steps_pos = [s[0] for s in steps]
assert steps_pos == [0, 5, 10, 15, 16, 20, 50, 70]
w[15] = 0.1
right, values, dists = solve_potts(y, w=w, gamma=0.1)
assert right == [5, 10, 20, 50, 70, 100]
steps = detect_steps(y, w=w)
steps_pos = [s[0] for s in steps]
assert steps_pos == [0, 5, 10, 20, 50, 70]
# Missing weights and data should be handled properly
y[35] = None
w[23] = None
steps = detect_steps(y, w=w)
steps_pos = [s[0] for s in steps]
assert steps_pos == [0, 5, 10, 20, 50, 70]
def test_detect_regressions(use_rangemedian):
try:
np.random.seed(1234)
except NameError:
# work around a bug in old pypy/pytest
pytest.skip("test needs numpy")
return
for seed in [1234, 5678, 8901, 2345]:
np.random.seed(seed)
t = np.arange(4000)
y = 0.7 * np.random.rand(t.size)
y -= 0.3 * (t >= 1000)
y += 2.0 * (t >= 3234 + (seed % 123))
y += 2.0 * ((t >= 3264 + (seed % 53)) & (t < 3700))
y -= 2.7 * ((t >= 3350) & (t < 3500 + (seed % 71)))
y = y.tolist()
y[123] = None
y[1234] = np.nan
steps = detect_steps(y)
steps_lr = [(l, r) for l, r, _, _, _ in steps]
k = steps[0][1]
assert 990 <= k <= 1010
assert steps_lr == [(0, k), (k, 3234 + (seed % 123)),
(3234 + (seed % 123), 3264 + (seed % 53)),
(3264 + (seed % 53), 3350),
(3350, 3500 + (seed % 71)),
(3500 + (seed % 71), 3700), (3700, 4000)]
steps_v = [x[2] for x in steps]
assert np.allclose(steps_v, [0.35, 0.05, 2.05, 4.05, 1.15, 4.05, 2.05],
rtol=0.3)
# The expected mean error is 0.7 <|U(0,1) - 1/2|> = 0.7/4
steps_err = [x[4] for x in steps]
assert np.allclose(steps_err, [0.7 / 4] * 7, rtol=0.3)
# Check detect_regressions
new_value, best_value, regression_pos = detect_regressions(steps)
assert regression_pos == [
(3233 + (seed % 123), (3233 + (seed % 123) + 1), steps_v[1],
steps_v[2]),
(3499 + (seed % 71), 3499 + (seed % 71) + 1, steps_v[4],
steps_v[5])
]
assert np.allclose(best_value, 0.7 / 2 - 0.3, rtol=0.3, atol=0)
assert np.allclose(new_value, 0.7 / 2 - 0.3 + 2, rtol=0.3, atol=0)
def test_detect_regressions(use_rangemedian):
try:
np.random.seed(1234)
except NameError:
# work around a bug in old pypy/pytest
pytest.skip("test needs numpy")
return
for seed in [1234, 5678, 8901, 2345]:
np.random.seed(seed)
t = np.arange(4000)
y = 0.7 * np.random.rand(t.size)
y -= 0.3 * (t >= 1000)
y += 2.0 * (t >= 3234 + (seed % 123))
y += 2.0 * ((t >= 3264 + (seed % 53)) & (t < 3700))
y -= 2.7 * ((t >= 3350) & (t < 3500 + (seed % 71)))
y = y.tolist()
y[123] = None
y[1234] = np.nan
steps = detect_steps(y)
steps_lr = [(l, r) for l, r, _, _, _ in steps]
k = steps[0][1]
assert 990 <= k <= 1010
assert steps_lr == [(0, k),
(k, 3234 + (seed % 123)),
(3234 + (seed % 123), 3264 + (seed % 53)),
(3264 + (seed % 53), 3350),
(3350, 3500 + (seed % 71)),
(3500 + (seed % 71), 3700),
(3700, 4000)]
steps_v = [x[2] for x in steps]
assert np.allclose(steps_v, [0.35, 0.05, 2.05, 4.05, 1.15, 4.05, 2.05], rtol=0.3)
# The expected mean error is 0.7 <|U(0,1) - 1/2|> = 0.7/4
steps_err = [x[4] for x in steps]
assert np.allclose(steps_err, [0.7/4]*7, rtol=0.3)
# Check detect_regressions
new_value, best_value, regression_pos = detect_regressions(steps)
assert regression_pos == [(3233 + (seed % 123), (3233 + (seed % 123) + 1), steps_v[1], steps_v[2]),
(3499 + (seed % 71), 3499 + (seed % 71) + 1, steps_v[4], steps_v[5])]
assert np.allclose(best_value, 0.7/2 - 0.3, rtol=0.3, atol=0)
assert np.allclose(new_value, 0.7/2 - 0.3 + 2, rtol=0.3, atol=0)
def time_detect_regressions(self):
steps = step_detect.detect_steps(self.y)
step_detect.detect_regressions(steps)
def time_detect_regressions(self):
steps = step_detect.detect_steps(self.y)
step_detect.detect_regressions(steps)