Supplementary Material for HyDI

Supplementary Material for Flügel, Mossakowski: HyDI: A hybrid Deep Learning-Inductive Logic Programming ensemble for multi-label classification (IJCLR, 2026, accepted, https://arxiv.org/abs/2609.37740).

The accompanying GitHub repository is https://github.com/ChEB-AI/chebILP/

The files include

• dataset splits and labels for the ChEBI25-3STAR dataset
• ILP runs with learned programs, validation and test set predictions
• DL model weights, validation and test set predictions
• Ensemble test set predictions and model-class mapping

The files do not include the ILP problems used for the ILP runs. Those can be found on [Hugging Face](https://huggingface.co/datasets/chebai/ChEBI25-3STAR-ILP).