https://github.com/takuti/anompy

A Python library for anomaly detection
https://github.com/takuti/anompy

anomaly-detection data-science forecasting machine-learning python

Last synced: about 1 month ago
JSON representation

A Python library for anomaly detection

• Host: GitHub
• URL: https://github.com/takuti/anompy
• Owner: takuti
• License: mit
• Created: 2016-10-15T03:50:31.000Z (about 8 years ago)
• Default Branch: master
• Last Pushed: 2017-08-28T08:04:00.000Z (about 7 years ago)
• Last Synced: 2024-10-04T15:59:24.569Z (about 2 months ago)
• Topics: anomaly-detection, data-science, forecasting, machine-learning, python
• Language: Python
• Size: 34.2 KB
• Stars: 13
• Watchers: 5
• Forks: 3
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

README

        
anompy

===

**anompy** is a Python package of forecasting and anomaly detection algorithms.

## Installation

```

$ pip install git+https://github.com/takuti/anompy.git

```

## Usage

Generate dummy time-series:

```py

>>> import random

>>> series = [random.random() for i in range(10)]

>>> series

[0.29749066250070444, 0.17992724665541393, 0.24201406949661697, 0.3467356134915024, 0.45318143064943217, 0.20825014566859423, 0.597497516445304, 0.5442072127508967, 0.1920841531842088, 0.2711214524302953]

```

Import `BaseDetector` which simply returns the last observed data point as a forecasted value, and create a detector with initial data point (i.e., training sample) and threshold:

```py

>>> from anompy.detector.base import BaseDetector

>>> detector = BaseDetector(series[0], threshold=0.5)

```

Get forecasted time-series and their anomaly labels by calling `detect()` method:

```py

>>> detector.detect(series[1:])

[(0.29749066250070444, False), (0.17992724665541393, False), (0.24201406949661697, False), (0.3467356134915024, False), (0.45318143064943217, False), (0.20825014566859423, False), (0.597497516445304, True), (0.5442072127508967, True), (0.1920841531842088, False)]

```

See [this notebook](https://gist.github.com/takuti/36d54e432a49424bb31d948926cd49b4) for more examples.

## Algorithm

anompy currently supports following algorithms:

- `BaseDetector`

    - Directly use the last observation as a forecasted value, and detect anomaly based on threshold.

- `AverageDetector`

    - Forecast either global average, simple moving average or weighted moving average.

- `ExponentialSmoothing`, `DoubleExponentialSmoothing`, and `TripleExponentialSmoothing`

    - See ["Exponential smoothing" on Wikipedia](https://en.wikipedia.org/wiki/Exponential_smoothing).

    - Triple exponential smoothing is also known as Holt-Winters method.

- Experimental

    - `ChangeFinder`

    - `SingularSpectrumTransform`

    - `StreamAnomalyDetector`