har_online_sources: Streaming data sources for Harbinger

In harbinger: A Unified Time Series Event Detection Framework

Streaming data sources for Harbinger

Description

The online layer of Harbinger consumes observations through source objects. A source abstracts how new observations are collected, allowing the online session to work with simulated vectors, data frames, callback-based feeds, and external stream collectors.

The source API is intentionally small:

• next_observation() requests the next observation in pull mode;

• source_info() exposes source metadata.

An observation consumed by the online layer is normalized to a list with:

• idx: logical position in the stream when known;

• value: the payload used by the detector;

• timestamp: optional temporal marker;

• payload: original payload preserved for adapters that need it.

Kafka support in this package is intentionally limited to a stub interface. Actual collection is expected to be delegated to Python code integrated via reticulate.

Value

Source object.

harbinger documentation built on July 10, 2026, 5:07 p.m.