Upload README.md with huggingface_hub

README.md

@@ -1,3 +1,102 @@
----
-license: cc
----
+---
+license: cc-by-4.0
+task_categories:
+  - graph-ml
+tags:
+  - physics
+  - high-energy-physics
+  - graph-classification
+  - gnn
+pretty_name: Graphical Bootstrap Correlator Dataset
+size_categories:
+  - 100G<n<1T
+---
+
+# Graphical Bootstrap Correlator Dataset
+
+Graph-structured dataset for binary classification of multi-loop Feynman integrals using Graph Neural Networks (GNNs). Each graph represents a denominator structure of a Feynman integral at loop orders 6–12, with the label indicating whether the integral satisfies a specific reduction criterion.
+
+## Dataset Structure
+
+```
+graphical-bootstrap-correlator-dataset/
+├── data/
+│   ├── den_graph_data_6.npz          # Loop 6 graph data
+│   ├── den_graph_data_7.npz          # Loop 7 graph data
+│   ├── den_graph_data_8.npz          # Loop 8 graph data
+│   ├── den_graph_data_9.npz          # Loop 9 graph data
+│   ├── den_graph_data_10.npz         # Loop 10 graph data
+│   ├── den_graph_data_11.npz         # Loop 11 graph data
+│   ├── den_graph_data_12.npz         # Loop 12 graph data (full)
+│   ├── den_graph_data_12_1.npz       # Loop 12 graph data (split 1/20)
+│   │   ...
+│   ├── den_graph_data_12_20.npz      # Loop 12 graph data (split 20/20)
+│   ├── features_loop_6/              # Pre-computed node features, loop 6
+│   ├── features_loop_7/              # Pre-computed node features, loop 7
+│   ├── features_loop_8/              # Pre-computed node features, loop 8
+│   ├── features_loop_9/              # Pre-computed node features, loop 9
+│   ├── features_loop_10/             # Pre-computed node features, loop 10
+│   └── features_loop_11/             # Pre-computed node features, loop 11
+├── embeddings/                       # Learned graph embeddings from trained models
+├── models/                           # Trained GNN model checkpoints (.pt files)
+└── rung_rule/                        # Rung rule analysis data (JSON/Parquet)
+```
+
+## Data Format
+
+### Graph data (`.npz` files)
+
+Each `.npz` file contains the full dataset for a given loop order and can be loaded with NumPy:
+
+```python
+import numpy as np
+
+data = np.load("data/den_graph_data_10.npz", allow_pickle=True)
+# Keys: depends on the split — typically includes edge arrays and labels
+```
+
+Labels are binary (0 or 1).
+
+### Pre-computed node features (`features_loop_N/`)
+
+Each directory contains `.npy` files with pre-computed node-level features for the corresponding loop order:
+
+| Category   | Features                                                            |
+|------------|---------------------------------------------------------------------|
+| Spectral   | `eigen_1`–`eigen_3`, `low_eigen_1`–`low_eigen_3`                   |
+| Structural | `degree`, `closeness`, `betweenness`, `clustering`, `pagerank`      |
+| Graphlet   | `graphlet_3`, `graphlet_4`                                          |
+| Distance   | `spd` (shortest path distances)                                     |
+
+```python
+import numpy as np
+
+features = np.load("data/features_loop_10/degree.npy")
+```
+
+## Usage with ml-correlator
+
+This dataset is designed for use with the [ML-Correlator](https://github.com/gdian/ML-correlator) training framework:
+
+```python
+from ml_correlator.graph_builder import create_simple_dataset
+
+dataset, scaler, feat_dim = create_simple_dataset(
+    file_ext='10',
+    selected_features=['low_eigen_1', 'degree', 'clustering'],
+    normalize=True,
+    data_dir='data',
+)
+```
+
+## Models
+
+Trained GNN checkpoints in `models/` are compatible with the `ml_correlator.architectures` module. Supported architectures: GIN, GAT, GraphTransformer (hybrid, simple, planar), GAT-GraphTransformer.
+
+## Rung Rule
+
+The `rung_rule/` directory contains analysis files used to study the rung rule structure across loop transitions (6→7, 7→8, 8→9, 9→10, 10→11, 11→12).
+
+## License
+
+This dataset is released under [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/).