SIRUS.jl: Interpretable Machine Learning via Rule Extraction
Rik Huijzer, Frank J. Blaauw, Ruud J. R. Den Hartigh · The Journal of Open Source Software · 2023
SIRUS.jl 1 is an implementation of the original Stable and Interpretable RUle Sets (SIRUS) algorithm in the Julia programming language (Bezanson et al., 2017).The SIRUS algorithm is a fully interpretable version of random forests, that is, it reduces thousands of trees in the forest to a much lower number of interpretable rules (e.g., 10 or 20).With our Julia implementation, we aimed to reproduce the original C++ and R implementation in a high-level language to verify the algorithm as well as making the code easier to read.We show that the model performs well on classification tasks while retaining interpretability and stability.Furthermore, we made the code available under the permissive MIT license.In turn, this allows others to research the algorithm further or easily port it to production systems.