crcf 1.0.0

Combination Robust Cut Forests

Combination Robust Cut Forests

Isolation Forests [Liu+2008] and Robust Random Cut Trees [Guha+2016] are very similar in many ways, as outlined in the supporting overview. Most notably, they are extremes of the same outlier scoring function:

$$\theta \textrm{Depth} + (1 - \theta) \textrm{[Co]Disp}$$

The combination robust cut forest allows you to combine both scores by using an theta other than 0 or 1.

Install

You can install with through pip install crcf. Alternatively, you can download the repository and run python3 setup.py install or pip3 install . Please note that this package uses features from Python 3.7+ and is not compatible with earlier Python versions.

Tasks

• complete basic implementation
• provide clear documentation and usage instructions
• ensure interface allows for fitting and scoring on multiple points at the same time
• implement a better saving method than pickling
• use random tests with hypothesis
• implement tree down in cython
• accelerate forests with multi-threading
• incorporate categorical variable support, including categorical rules
• complete the write-up document with a benchmarking of performance

References

• [Liu+2008]: Liu, Fei Tony, Kai Ming Ting, and Zhi-Hua Zhou. "Isolation forest." In 2008 Eighth IEEE International Conference on Data Mining, pp. 413-422. IEEE, 2008.
• [Guha+2016]: Guha, Sudipto, Nina Mishra, Gourav Roy, and Okke Schrijvers. "Robust random cut forest based anomaly detection on streams." In International conference on machine learning, pp. 2712-2721. 2016.