ahmeddeladly/singular-spectrum-transformation

fast implementation of singular spectrum transformation (change point detection algorithm)

SST (Singular Spectrum Transformation)

A fast implementation of Singular Spectrum Transformation for python.

What is SST?

A change point detection algorithm. see more examples

Features

fast computation

• efficient algorithm using Lanczos method
• Numba

Installation

$pip install fastsst

Basic Usage

from fastsst import SingularSpectrumTransformation

sst = SingularSpectrumTransformation(win_length=30)

"""
note:
 - data must be 1d np.ndarray
 - the first run takes a few seconds for jit compling
"""
score = sst.score_offline(data)

Others

• Comparison between conventional method(i.e. SVD) and faster method(i.e. lanczos)

TODO

• online evaluation

References

1. Tsuyoshi Ide, Koji Tsuda, Change-Point Detection using Krylov Subspace Learning, SIAM International Conference on Data Mining, pp.515-520, 2007
2. Tsuyoshi Ide, Speeding up Change-Point Detection using Matrix Compression (Japanse), Workshop on Information-Based Induction Sciences, 2006
3. Tsuyoshi Ide, Masashi Sugiyama, Anomaly Detection and Change Detection (Japanse), Kodansha, 2015