si-clad 0.0.1

This package provides a Statistical Inference framework for testing the anomaly results obtained by DBSCAN algorithm.

Statistical Inference for Cluster-based Anomaly Detection

This package provides a Statistical Inference framework for Cluster-based Anomaly Detection, with controllable FPR. In particular, we propose a valid p-value for testing the anomaly results obtained by DBSCAN algorithm. Basically, the problem is decomposed into multiple tractable sub-problems to enable an efficient test with the highest TPR while controlling the FPR through the divide-and-conquer approach.

This package has the following requirements:

mpmath
numpy>=1.23.0
scipy>=1.4.1

Installization

You can install this package from PyPI using:

pip install si-clad

Example

from si_clad import SI_CLAD, generate, DBSCAN_AD

n = 50
d = 10
delta =  0 #no true outliers
X, Sigma, _ = generate(n, d, delta)

minpts =  10
eps = 3
O, _ = DBSCAN_AD(eps, minpts).fit(X)
p_value = SI_CLAD(X, Sigma, minpts, eps, O, j = None) #randomly choose an outlier j for testing

print(p_value)