pyanom

This library is Python projects for anomaly detection. This contains these techniques.

• Kullback-Leibler desity estimation
• Singular spectrum analysis
• Graphical lasso
• CUMSUM anomaly detection
• Hoteling T2
• Directional data anomaly detection

REQUIREMENTS

• numpy
• pandas
• scikit-learn
• scipy

INSTALLATION

pip install pyanom

USAGE

We have posted a usage example in the demo folder.

Kullback-Leibler desity estimation

import numpy as np
from pyanom.density_ratio_estimation import KLDensityRatioEstimator

X_normal = np.loadtxt("./data/normal_data.csv", delimiter=",")
X_error = np.loadtxt("./data/error_data.csv", delimiter=",")

model = KLDensityRatioEstimator(
    band_width=h, lr=0.001, max_iter=100000)
model.fit(X_normal, X_error)
anomaly_score = model.score(X_normal, X_error)

Singular spectrum analysis

import numpy as np
from pyanom.subspace_methods import SSA

y_error = np.loadtxt("./data/timeseries_error2.csv", delimiter=",")

model = SSA(window_size=50, trajectory_n=25, trajectory_pattern=3, test_n=25, test_pattern=2, lag=25)
model.fit(y_error)
anomaly_score = model.score()

Graphical lasso

import numpy as np
from pyanom.structure_learning import GraphicalLasso

X_normal = np.loadtxt("./data/normal_data.csv", delimiter=",")
X_error = np.loadtxt("./data/error_data.csv", delimiter=",")

model = GraphicalLasso(rho=0.1)
model.fit(X_normal)
anomaly_score = model.score(X_error)

Direct learning sparse changes

from pyanom.structure_learning import DirectLearningSparseChanges

model = DirectLearningSparseChanges(
    lambda1=0.1, lambda2=0.3, max_iter=10000)
model.fit(X_normal, X_error)
pmatrix_diff = model.get_sparse_changes()

CUSUM anomaly detection

import numpy as np
from pyanom.outlier_detection import CAD

y_normal = np.loadtxt(
    "./data/timeseries_normal.csv", delimiter=",").reshape(-1, 1)
y_error = np.loadtxt(
    "./data/timeseries_error.csv", delimiter=",").reshape(-1, 1)

model = CAD(threshold=1.0)
model.fit(y_normal)
anomaly_score = model.score(y_error)

Hoteling T2

import numpy as np
from pyanom.outlier_detection import HotelingT2

X_normal = np.loadtxt("./data/normal_data.csv", delimiter=",")
X_error = np.loadtxt("./data/error_data.csv", delimiter=",")

model = HotelingT2()
model.fit(X_normal)
anomaly_score = model.score(X_error)

Anomaly detection of directional data

import numpy as np
from pyanom.outlier_detection import AD3

X_normal = np.loadtxt(
    "./data/normal_direction_data.csv", delimiter=",")
X_error = np.loadtxt("./data/error_direction_data.csv", delimiter=",")

model = AD3()
model.fit(X_normal, normalize=True)
anomaly_score = model.score(X_error)

Test

python -m unittest discover