🐸 FROG: Full-Resolution and Optimizable Graph Structure Learning Framework for RDL

Official implementation of the paper:

[ICML 2026] Is Fixing Schema Graphs Necessary? Full-Resolution Graph Structure Learning for Relational Deep Learning

---

📖 Overview

FROG is a Full-Resolution and Optimizable Graph Structure Learning Framework for Relational Deep Learning (RDL).

Instead of relying on fixed schema graphs, FROG enables full-resolution graph structure learning and jointly optimizes relational graph construction with downstream GNN training.

---

🚀 Installation

FROG uses the same environment as RelBench.

Install dependencies with:

pip install relbench[full]

---

⚡ Quick Start

1. Entity Classification / Regression

Run entity classification or regression tasks:

python -u gnn_entity.py \
    --dataset {DATASET} \
    --task {TASK} \
    --load_para

2. Entity Recommendation

Run recommendation tasks:

python -u gnn_recommendation.py \
    --dataset {DATASET} \
    --task {TASK} \
    --load_para

---

📂 Project Structure

FROG/
├── log/                           # Training logs
│   └── rel-comp.py               # Statistical analysis of Table-as-Node/Edge behaviors
├── exp_model.py                  # REG structure + GNN optimization framework
├── gnn_entity.py                 # Entity classification & regression training
├── gnn_recommendation.py         # Entity recommendation training
├── graph.py                      # Graph construction utilities
├── hop_mode.py                   # Co-occurrence & completion implementation
├── load_train_args.py            # Hyperparameter configuration
├── model.py                      # Core FROG model
├── text_embedder.py              # Text embedding module
├── utils.py                      # Utility functions
└── README.md

---

🧪 Supported Tasks

• Entity Classification
• Entity Regression
• Entity Recommendation

---

🔗 Resources

• Paper: https://arxiv.org/pdf/2605.21475
• Code: https://github.com/RingBDStack/FROG