Graph Construction Script
Bridges raw telecom data → GNN-ready graph tensors.
Source File
What It Produces
| Tensor | Shape | Description |
|---|---|---|
x |
[12, 29] | Node features: 19 KPIs + 7 type one-hot + 3 domain one-hot |
edge_index |
[2, 28] | 14 undirected edges → 28 directed edges |
edge_attr |
[28, 10] | Edge features: 9 KPIs + 1 edge type |
y |
[12] | Multi-class labels (5 classes) |
y_binary |
[12] | Binary root cause labels |
y_ranking |
[12] | Root cause ranking scores |
Dependencies
pandas,numpy(required)torch,torch-geometric(optional — runs in demo mode without)
Usage
source .venv/bin/activate
python scripts/07_pyg_graph_construction.py
Also Includes
A minimal SimpleRCAGNN model (GCN, 2 layers, 5-class output) for demonstrating a forward pass. Not trained — requires 500+ labeled incidents for meaningful training.