def test_base():
num_categ = 8
clf = LatestWindowAnomalyDetector(alpha=0.00001)
history = dummy_data(num_days=10,
num_categories=num_categ,
min_val=100,
max_val=1000)
new_day = dummy_data(num_days=1,
num_categories=num_categ,
min_val=100,
max_val=1000)
clf.fit(history)
prediction = clf.predict(new_day)
assert len(prediction) == num_categ
for x in prediction.values:
assert x in [-1, 0, 1]
num_new_days = 30
many_days = dummy_data(num_days=num_new_days,
num_categories=num_categ,
min_val=100,
max_val=1000)
predictions = clf.predict(many_days)
assert len(predictions) == num_categ * num_new_days
for x in predictions.values:
assert x in [-1, 0, 1]
def test_partial_fit():
num_categ = 8
clf = LastNWindowsAnomalyDetector(
n_windows=4,
weights=uniform_weighter(),
alpha=0.001,
)
history = dummy_data(num_days=10,
num_categories=num_categ,
min_val=1000,
max_val=1200)
recent_history = dummy_data(num_days=6,
num_categories=num_categ,
min_val=1000,
max_val=1200)
new_day = dummy_data(num_days=1,
num_categories=num_categ,
min_val=1000,
max_val=1200)
clf.fit(history)
clf.partial_fit(recent_history)
prediction = clf.predict(new_day)
assert len(prediction) == num_categ
for x in prediction.values:
assert x in [-1, 0, 1]
def test_exp_comp_weighter():
num_categ = 8
n_windows = 4
clf = LastNWindowsAnomalyDetector(
n_windows=n_windows,
weights=exp_comp_weighter(n=n_windows, concave_factor=1),
alpha=0.001,
)
history = dummy_data(num_days=10,
num_categories=num_categ,
min_val=1000,
max_val=1200)
new_day = dummy_data(num_days=1,
num_categories=num_categ,
min_val=1000,
max_val=1200)
clf.fit(history)
prediction = clf.predict(new_day)
assert len(prediction) == num_categ
for x in prediction.values:
assert x in [-1, 0, 1]
clf = LastNWindowsAnomalyDetector(
n_windows=n_windows,
weights=exp_comp_weighter(n=n_windows, concave_factor=3),
alpha=0.001,
)
clf.fit(history)
prediction = clf.predict(new_day)
assert len(prediction) == num_categ
for x in prediction.values:
assert x in [-1, 0, 1]
clf = LastNWindowsAnomalyDetector(
n_windows=n_windows,
weights=exp_comp_weighter(n=n_windows, concave_factor=30),
alpha=0.001,
)
clf.fit(history)
prediction = clf.predict(new_day)
assert len(prediction) == num_categ
for x in prediction.values:
assert x in [-1, 0, 1]
clf = LastNWindowsAnomalyDetector(
n_windows=n_windows,
weights=exp_comp_weighter(n=n_windows, concave_factor=None),
alpha=0.001,
)
clf.fit(history)
prediction = clf.predict(new_day)
assert len(prediction) == num_categ
for x in prediction.values:
assert x in [-1, 0, 1]
def test_non_def_ddof():
num_categ = 8
clf = LatestWindowAnomalyDetector(p_threshold=0.00001, power=-2, ddof=4)
history = dummy_data(
num_days=10, num_categories=num_categ, min_val=100, max_val=1000)
new_day = dummy_data(
num_days=1, num_categories=num_categ, min_val=100, max_val=1000)
clf.fit(history)
prediction = clf.predict(new_day)
assert len(prediction) == num_categ
for x in prediction.values:
assert x in [-1, 0, 1]
def test_diff_categ():
num_categ_1 = 8
num_categ_2 = 7
clf = LatestWindowAnomalyDetector(p_threshold=0.00001)
history = dummy_data(
num_days=10, num_categories=num_categ_1, min_val=100, max_val=1000)
new_day = dummy_data(
num_days=1, num_categories=num_categ_2, min_val=100, max_val=1000)
clf.fit(history)
prediction = clf.predict(new_day)
assert len(prediction) == max(num_categ_1, num_categ_2)
for x in prediction.values:
assert x in [-1, 0, 1]
def test_partial_fit():
num_categ = 8
clf = LatestWindowAnomalyDetector(p_threshold=0.00001)
history = dummy_data(
num_days=10, num_categories=num_categ, min_val=100, max_val=1000)
recent_history = dummy_data(
num_days=6, num_categories=num_categ, min_val=100, max_val=1000)
new_day = dummy_data(
num_days=1, num_categories=num_categ, min_val=100, max_val=1000)
clf.fit(history)
clf.partial_fit(recent_history)
prediction = clf.predict(new_day)
assert len(prediction) == num_categ
for x in prediction.values:
assert x in [-1, 0, 1]
def test_errors():
# bad p thresholds
with pytest.raises(ValueError):
LastNWindowsAnomalyDetector(
n_windows=4,
weights=uniform_weighter(),
alpha=2,
)
# bad p thresholds
with pytest.raises(ValueError):
LastNWindowsAnomalyDetector(
n_windows=4,
weights=uniform_weighter(),
alpha=-1,
)
clf = LastNWindowsAnomalyDetector(
n_windows=4,
weights=uniform_weighter(),
alpha=0.001,
)
new_day = dummy_data(num_days=1,
num_categories=8,
min_val=1000,
max_val=1200)
with pytest.raises(NotFittedError):
clf.predict(new_day)
def test_weights_function():
num_categ = 8
clf = LastNWindowsAnomalyDetector(
n_windows=6,
weights=_weight_by_inverse_days_delta,
alpha=0.001,
)
history = dummy_data(num_days=10,
num_categories=num_categ,
min_val=1000,
max_val=1200)
new_day = dummy_data(num_days=1,
num_categories=num_categ,
min_val=1000,
max_val=1200)
clf.fit(history)
prediction = clf.predict(new_day)
assert len(prediction) == num_categ
for x in prediction.values:
assert x in [-1, 0, 1]
def test_errors():
# bad p thresholds
with pytest.raises(ValueError):
LatestWindowAnomalyDetector(p_threshold=2)
# bad p thresholds
with pytest.raises(ValueError):
LatestWindowAnomalyDetector(p_threshold=-1)
clf = LatestWindowAnomalyDetector(p_threshold=0.00001)
new_day = dummy_data(
num_days=1, num_categories=8, min_val=100, max_val=1000)
with pytest.raises(NotFittedError):
clf.predict(new_day)
def test_diff_categ():
num_categ_1 = 8
num_categ_2 = 7
clf = LastNWindowsAnomalyDetector(
n_windows=4,
weights=uniform_weighter(),
alpha=0.001,
)
history = dummy_data(num_days=10,
num_categories=num_categ_1,
min_val=1000,
max_val=1200)
new_day = dummy_data(num_days=1,
num_categories=num_categ_2,
min_val=1000,
max_val=1200)
clf.fit(history)
prediction = clf.predict(new_day)
assert len(prediction) == max(num_categ_1, num_categ_2)
for x in prediction.values:
assert x in [-1, 0, 1]
