HONAM is an interpretable deep learning model proposed in our paper: Higher-order Neural Additive Models: An Interpretable Machine Learning Model with Feature Interactions.

HONAM consists of two parts: 1) feature networks and 2) a feature interaction module. The feature networks make the representation vectors of the corresponding features, and then the feature interaction module makes high-order feature interactions. Therefore, HONAM can produce accurate and interpretable predictions.

We have implemented the code in the following python environment:

• python 3.8.12 (3.9.16)
• pytorch 1.10.2 (1.11.0)
• pandas 1.2.3 (1.2.4)
• numpy 1.21.2 (1.24.3)
• scikit-learn 1.0.2 (1.2.2)

We provide an example code for the California Housing Prices dataset.

For training:

python run.py --mode=train --dataset=clifornia_housing

HONAM supports sklearn-style interface.

from model import HONAM

model = HONAM(...)
model.fit(x_train, y_train)
prediction = model.predict(x_test) 

@article{kim2022higher,
  title = {Higher-order Neural Additive Models: An Interpretable Machine Learning Model with Feature Interactions},
  author = {Kim, Minkyu and Choi, Hyun-Soo and Kim, Jinho},
  journal = {arXiv preprint arXiv:2209.15409},
  year = {2022}
}