Anomaly Detection Framework Documentation¶
Warning
Anomaly Detection Framework is under an early development phase. API, architecture and implementations could suffer important changes. For this, Bluekiri do not offer any type of warranty. Use at your own responsibility.
Overview¶
This project born from the need of detect anomalies on multiple and completely different signals, and react to it rapidly. To achieve this, Bluekiri decided to implement its own system to manage multiple signals at the same time in a easy and scalable way.
This project is not focused on Machine Learning models, but in an effective Framework to put those models in production.
Content¶
Features¶
- It has an abstraction layer to implement engines, streaming sources and sinks, repository sources and sinks.
- It supports Kafka and PubSub by default.
- It has a default implementation for that supports tumbling window aggregation on Spark, on Kafka and PubSub sources using an specific JSON schema messages.
- Configuration file in YAML format.
- It also includes a dashboard to visualize the signal anomalies and play with signals in a sandbox to try models, tune parameters an see which parameters fits better into your signal.
Note
The Sandbox is limited to use the included message handlers and models only. Custom models will be available to use after we implement the plugin system.
Content
- Project
- License
- Getting started
- Configuration
- Deploy
- Plugins
- API Reference
- anomalydetection
- anomalydetection.common
- anomalydetection.backend
- backend.core
- backend.entities
- backend.engine
- backend.interactor
- backend.repository
- backend.repository.builder
- backend.repository.observable
- backend.repository.sqlite
- backend.sink
- backend.sink.repository
- backend.sink.stream
- backend.sink.websocket
- backend.stream
- backend.stream.agg
- backend.stream.builder
- backend.stream.kafka
- backend.stream.pubsub