Architecture

CapyMOA is organised into research domains. Each research domain are maintained semi-independently by domain experts (see CODEOWNERS) with shared cross domain interoperability, continuous integration, and documentation. Domains can develop independently. This reflects how CapyMOA is built, allowing researchers to own and contribute to their specific research domains. Reduces the cognitive load required for a user to get started, while also driving the discovery of related features within a domain.

capymoa.anomaly

Streaming anomaly detection.

capymoa.automl

Streaming automated machine learning.

capymoa.classifier

Streaming classification.

capymoa.cluster

Streaming clustering.

capymoa.drift

Streaming concept and data drift detection.

capymoa.feature

Streaming feature importance estimation.

capymoa.ocl

Online (/streaming) continual learning.

capymoa.regressor

Streaming regression.

capymoa.ssl

Streaming semi-supervised learning

capymoa.uncertainty

Streaming prediction intervals and uncertainty estimation.

Warning

At the moment this parts of the documentation is aspirational rather than reflecting the current state of the project.

Each domain shall implement its own:

capymoa.{{domain}}                      # Domain-specific modules (listed above)
- {{Domain}}Metrics          (class)    # Serializable metrics
- evaluate_{{domain}}        (function) # Evaluation logic
- *Algorithm                 (classes)  # Algorithm implementations

capymoa.{{domain}}.base      (optional) # Abstract base classes for the module
capymoa.{{domain}}.datasets  (optional) # Domain-specific datasets
capymoa.{{domain}}.evaluate  (optional) # Public evaluation code
capymoa.{{domain}}.plot      (optional) # Public plotting code

In addition to the research domain CapyMOA maintains some common features to simplify implementation and facilitate interoperability: