Skip to the content.

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

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.