pystream

An MOA-based implementation for data stream classification in Python/Cython integrated with Stream Based Active Learning Techniques.

Includes:

Base learners:

• Very Fast Decision Tree (https://homes.cs.washington.edu/~pedrod/papers/kdd00.pdf)
• Strict Very Fast Decision Tree (https://www.sciencedirect.com/science/article/pii/S0167865518305580)
• OLBoost (to be upload on arxiv)

Ensembles:

• OzaBag/OzaBoost (https://ti.arc.nasa.gov/m/profile/oza/files/ozru01a.pdf)
• OAUE (https://www.sciencedirect.com/science/article/pii/S0020025513008554)
• LevBag (https://core.ac.uk/download/pdf/129931682.pdf)
• ARF (https://link.springer.com/article/10.1007/s10994-017-5642-8)

Util and evaluation classes.

To run:

• pip install -r requirements.txt --user
• chmod +x compile.sh
• ./compile.sh (builds Cython extensions, creates .so files and installs the library inplace)
• Follow tests/test.py file

TODO:

• Fully document code
• Improve Cython implementation
• Add more algorithms
• Provide a better usage manual
• Cleanup code better

Notes

• Needs cython to compile code when installing
• The datasets files can be found in here, just download the datasets.tar.gz file and extract the entire datasets directory inside the tests directory before running test.py