CapyMOA

CapyMOA does efficient machine learning for data streams in Python. A data stream is a sequence of items that arrive one by one and are too large or urgent to process offline. CapyMOA is a toolbox of methods and evaluators for: classification, regression, clustering, anomaly detection, semi-supervised learning, online continual learning, and drift detection for data streams.

To install:

pip install capymoa

The deep-learning parts of CapyMOA (capymoa.ocl, capymoa.core.torch.ann, the Batch* learners) need PyTorch, which is an optional extra:

pip install capymoa[torch]

Refer to the Setup guide for other options, including CPU-only PyTorch and dev dependencies.

from capymoa.datasets import Electricity
from capymoa.classifier import HoeffdingTree
from capymoa.evaluation import prequential_evaluation

# 1. Load a streaming dataset
stream = Electricity()

# 2. Create a machine learning model
model = HoeffdingTree(stream.get_schema())

# 3. Run with test-then-train evaluation
results = prequential_evaluation(stream, model)

# 4. Success!
print(f"Accuracy: {results.accuracy():.2f}%")

Next, we recommend the Tutorials.

⚠️ WARNING

CapyMOA is still in the early stages of development. The API is subject to change until version 1.0.0. If you encounter any issues, please report them in GitHub Issues or talk to us on Discord.

Benchmark comparing CapyMOA against other data stream libraries. The benchmark was performed using an ensemble of 100 ARF learners trained on the capymoa.datasets.RTG_2abrupt dataset containing 100,000 samples and 30 features. You can find the code to reproduce this benchmark in benchmarks/README.md, with the runnable script at benchmarks/benchmarking.py. CapyMOA has the speed of MOA with the flexibility of Python and the richness of Python's data science ecosystem.

Cite Us

If you use CapyMOA in your research, please cite us using the following BibTeX item.

@misc{gomes2025,
  title={{CapyMOA}: Efficient Machine Learning for Data Streams and Online Continual Learning in Python},
  author={Heitor Murilo Gomes and Anton Lee and Nuwan Gunasekara and Yibin Sun and Guilherme Weigert Cassales and Justin Jia Liu and Marco Heyden and Vitor Cerqueira and Maroua Bahri and Yun Sing Koh and Bernhard Pfahringer and Albert Bifet},
  year={2025},
  eprint={2502.07432},
  archivePrefix={arXiv},
  primaryClass={cs.LG},
  url={https://arxiv.org/abs/2502.07432}
}