MISS

| Paper |

This is a repository containing the code to generate Multiclass Interpretable Scoring Systems (MISS) as the one below:

Installation

To use MISS, clone the repository and install all the required libraries:

pip install -r requirements.txt

Then, install risk-slim with multiclass extensions:

cd risk-slim
pip install -e .

Usage

You can run the example MISS training with:

cd miss
python miss_example.py

This will create a multiclass scoring system based on the iris dataset.

You can train your own scoring systems with scikit-learn compatible api:

from miss.models import MISSClassifier

mcrsc = MISSClassifier(
    mc_l0_min=0,
    mc_l0_max=3,
    max_coefficient=5,
    max_intercept=10
)

x_train = #... load dataset with binary features
y_train = #... pandas dataframe with 0, ..., K-1 values

mcrsc.fit(x_train, y_train)

References

Please consider citing our paper:

@inproceedings{grzeszczyk2024miss,
  title={MISS: Multiclass Interpretable Scoring Systems},
  author={Grzeszczyk, Michal K and Trzci{\'n}ski, Tomasz and Sitek, Arkadiusz},
  booktitle={Proceedings of the 2024 SIAM International Conference on Data Mining (SDM)},
  pages={55--63},
  year={2024},
  organization={SIAM}
}

The implementation of risk-slim is taken from the original risk-slim repository. We have broadened the implementation to enable Multiclass (mc) scoring systems generation.

Among the most important papers that helped during the implemenation of this project we have to name:

Ustun, Berk, and Cynthia Rudin. "Learning optimized risk scores." Journal of Machine Learning Research 20.150 (2019): 1-75.

Pajor, Arkadiusz, et al. "Effect of feature discretization on classification performance of explainable scoring-based machine learning model." International Conference on Computational Science. Cham: Springer International Publishing, 2022.