def test_calibrators():
from pysad.transform.probability_calibration import GaussianTailProbabilityCalibrator
import numpy as np
from pysad.transform.probability_calibration import ConformalProbabilityCalibrator
from pysad.utils import fix_seed
fix_seed(61)
scores = np.random.rand(100)
calibrators = {
GaussianTailProbabilityCalibrator: {},
ConformalProbabilityCalibrator: {}
}
for calibrator_cls, args in calibrators.items():
calibrator = calibrator_cls(**args)
calibrated_scores = calibrator.fit_transform(scores)
assert calibrated_scores.shape == scores.shape
assert not np.isnan(calibrated_scores).any()
calibrator = calibrator_cls(**args).fit(scores)
assert type(calibrator) is calibrator_cls
calibrated_scores = calibrator.fit_transform(scores)
assert calibrated_scores.shape == scores.shape
assert not np.isnan(calibrated_scores).any()
def test_stats_with_batch_numpy():
from pysad.statistics import AverageMeter
from pysad.statistics import CountMeter
from pysad.statistics import MaxMeter
from pysad.statistics import MedianMeter
from pysad.statistics import MinMeter
from pysad.statistics import SumMeter
from pysad.statistics import SumSquaresMeter
from pysad.statistics import VarianceMeter
import numpy as np
from pysad.utils import fix_seed
fix_seed(61)
num_items = 100
stat_classes = {
AverageMeter: np.mean,
CountMeter: len,
MaxMeter: np.max,
MedianMeter: np.median,
MinMeter: np.min,
SumMeter: np.sum,
SumSquaresMeter: lambda x: np.sum(x**2),
VarianceMeter: np.var
}
for stat_cls, val in stat_classes.items():
stat = stat_cls()
abs_stat = AbsStatistic(stat_cls)
window_size = 25
running_stat = RunningStatistic(stat_cls, window_size=window_size)
arr = np.random.rand(num_items)
prev_value = 0.0
for i in range(arr.shape[0]):
num = arr[i]
stat.update(num)
abs_stat.update(num)
running_stat.update(num)
if i > 1: # for variance meter.
assert np.isclose(stat.get(), val(arr[:i + 1]))
assert np.isclose(running_stat.get(),
val(arr[max(0, i - window_size + 1):i + 1]))
assert np.isclose(abs(stat.get()), abs_stat.get())
stat.remove(num)
abs_stat.remove(num)
if i > 1:
assert np.isclose(stat.get(), prev_value)
assert np.isclose(abs_stat.get(), abs(prev_value))
stat.update(num)
abs_stat.update(num)
prev_value = stat.get()
def test_all_zero_stats():
import numpy as np
from pysad.statistics import AbsStatistic
from pysad.statistics import RunningStatistic
from pysad.statistics import AverageMeter
from pysad.statistics import CountMeter
from pysad.statistics import MaxMeter
from pysad.statistics import MedianMeter
from pysad.statistics import MinMeter
from pysad.statistics import SumMeter
from pysad.statistics import SumSquaresMeter
from pysad.statistics import VarianceMeter
from pysad.utils import fix_seed
fix_seed(61)
num_items = 100
stat_classes = {
AverageMeter: 0.0,
CountMeter: "count",
MaxMeter: 0.0,
MedianMeter: 0.0,
MinMeter: 0.0,
SumMeter: 0.0,
SumSquaresMeter: 0.0,
VarianceMeter: 0.0
}
for stat_cls, val in stat_classes.items():
stat = stat_cls()
abs_stat = AbsStatistic(stat_cls)
window_size = 25
running_stat = RunningStatistic(stat_cls, window_size=window_size)
arr = np.zeros(num_items, dtype=np.float)
prev_value = 0.0
for i in range(arr.shape[0]):
num = arr[i]
stat.update(num)
abs_stat.update(num)
running_stat.update(num)
if i > 1: # for variance meter.
assert np.isclose(stat.get(), val if val != "count" else i + 1)
assert np.isclose(abs_stat.get(),
val if val != "count" else i + 1)
assert np.isclose(
running_stat.get(),
val if val != "count" else min(i + 1, window_size))
stat.remove(num)
abs_stat.remove(num)
assert np.isclose(stat.get(), prev_value)
assert np.isclose(abs_stat.get(), abs(prev_value))
stat.update(num)
abs_stat.update(num)
prev_value = stat.get()
def test_fit_and_score_separately():
from pysad.models import xStream
import numpy as np
from pysad.utils import fix_seed
fix_seed(61)
X = np.random.rand(150, 1)
model = xStream()
model = model.fit(X)
y_pred = model.score(X)
assert y_pred.shape == (X.shape[0], )
def test_all_correct():
from pysad.evaluation import PrecisionMetric, AUPRMetric, AUROCMetric, RecallMetric
import numpy as np
from pysad.utils import fix_seed
fix_seed(61)
metric_classes = [PrecisionMetric, RecallMetric, AUPRMetric, AUROCMetric]
metric_classes = {metric_cls: 1.0 for metric_cls in metric_classes}
y_true = np.random.randint(0, 2, size=(25, ), dtype=np.int)
y_true[0] = 1
y_true[1] = 0
y_pred = y_true.copy()
helper_test_all_metrics(metric_classes, y_true, y_pred)
def test_perfect_model():
import numpy as np
from pysad.models import PerfectModel
from pysad.utils import fix_seed
fix_seed(61)
model = PerfectModel()
y1 = np.random.randint(0, 2, 100)
y = np.random.randint(0, 2, 100)
X = np.random.rand(100)
y_pred = model.fit_score(X, y)
print(y_pred)
assert np.all(np.isclose(y, y_pred))
assert not np.all(np.isclose(y1, y_pred))
def test_standard_absolute_deviation():
from pysad.models import StandardAbsoluteDeviation
import numpy as np
from numpy.testing import assert_raises
from pysad.utils import fix_seed
fix_seed(61)
X = np.random.rand(150, 1)
model = StandardAbsoluteDeviation(substracted_statistic="mean")
model = model.fit(X)
y_pred = model.score(X)
assert y_pred.shape == (X.shape[0],)
model = StandardAbsoluteDeviation(substracted_statistic="median")
model = model.fit(X)
y_pred = model.score(X)
assert y_pred.shape == (X.shape[0],)
with assert_raises(ValueError):
StandardAbsoluteDeviation(substracted_statistic="asd")
with assert_raises(ValueError):
StandardAbsoluteDeviation(substracted_statistic=None)
def test_unsupervised_models():
from pysad.models import RobustRandomCutForest
from pysad.models import ExactStorm
from pysad.models import HalfSpaceTrees
from pysad.models import IForestASD
from pysad.models import KitNet
from pysad.models import KNNCAD
from pysad.models import LODA
from pysad.models import LocalOutlierProbability
from pysad.models import MedianAbsoluteDeviation
from pysad.models import NullModel
from pysad.models import RandomModel
from pysad.models import RelativeEntropy
from pysad.models import RSHash
from pysad.models import StandardAbsoluteDeviation
from pysad.models import xStream
import numpy as np
from pysad.utils import fix_seed
fix_seed(61)
X = np.random.rand(150, 1)
model_classes = {
ExactStorm: {},
HalfSpaceTrees: {
"feature_mins": [0.0],
"feature_maxes": [1.0]
},
IForestASD: {},
KitNet: {},
KNNCAD: {
"probationary_period": 50
},
LODA: {},
LocalOutlierProbability: {
"initial_X": True
},
MedianAbsoluteDeviation: [{}, {
"absolute": False
}],
NullModel: {},
RandomModel: {},
RelativeEntropy: {
"min_val": 0.0,
"max_val": 1.0
},
RSHash: {
"feature_mins": [0.0],
"feature_maxes": [1.0]
},
StandardAbsoluteDeviation: [{}, {
"absolute": False
}],
xStream: {},
RobustRandomCutForest: {}
}
for model_cls, params_dict in model_classes.items():
if type(params_dict) is dict:
helper_test_model(X, model_cls, params_dict)
elif type(params_dict) is list:
for params in params_dict:
helper_test_model(X, model_cls, params)