Supplementary Code for LLM-based Decision Tree Induction

More detailed README will be added shortly.

Installation: See pyproject.toml

Running: There are two main notebooks:

1. ./build_tree.ipynb for learning a tree from scratch.
2. ./build_tree_over_tabpfn.ipynb for learning LLM-guided tree that refines TabPFNv2 predictions.

The notebooks rely on an OpenAI-like proxy API by default.

To switch to a public API, replace ProxyAPIModel with smolagents.LiteLLMModel and initialize it according to the documentation. For open-source models, you can use either HuggingFace API or custom deployment as per the smolagents API reference.

---

Human input experiments:

3. ./lost_feature.ipynb for learning a tree with lost feature.

Fairness will be added shortly.