The bottleneck to deploying machine learning, and therefore its use more broadly in a great range of problems in business and society, is a lack of interpretability. This is the sentiment "we can't trust this black-box". This repository aims to replicate various interpretable machine learning algorithms for their broad use and dissemination.
It is highly recommended to use the provided conda environment:
conda env create -f environment.yml
conda activate pynterp
To install, run:
python setup.py install
The basic interface is:
from pynterp.rules.decision_set import DecisionSet
ds = DecisionSet()
ds.fit(X_train, y_train)
Rules-based methods are models that learn a list or set of IF-THEN-ELSE rules;
they are highly interpretable and appealing to humans. The current implementation
of DecisionSet
follows the paper by Lakkaraju et. al.
October 23, 2019:
- Implement test suite and continuous integration framework
- Incorporate Bayesian decision lists (due to Rudin et. al)