Course Objective & Scope
When money transfers, accounts sharing devices or IPs, supply-chain dependencies or user-item interactions are flattened into isolated table rows, the most valuable signal—network structure—is lost. This programme covers constructing homogeneous and heterogeneous graphs from relational tables, classical graph features such as PageRank, betweenness and Louvain/Leiden communities, embeddings like Node2Vec, and GraphSAGE, GCN and GAT models in PyTorch Geometric (PyG).
It compares when graph features fed into GBDTs are sufficient versus when end-to-end Graph Neural Networks (GNNs) add real lift. Participants build and evaluate a graph model on a fraud-ring, risk-contagion or link-prediction scenario.
Who Should Attend?
- Data scientists and ML engineers
- Fraud, AML, cybersecurity and risk intelligence analytics teams
- Developers working on supply networks, telecom topologies and recommendation graphs
- Prerequisite: Python, machine learning and basic PyTorch knowledge
Key Takeaways
- Design homogeneous and heterogeneous temporal graph schemas from relational tables
- Extract centrality, k-core, motif and community features to enrich tabular GBDT models
- Train GraphSAGE and GAT models in PyTorch Geometric for node classification and link prediction
- Scale training with neighbour sampling (NeighborLoader) while preventing temporal edge leakage
- Benchmark tabular+graph-feature baselines against end-to-end GNNs on business metrics
Curriculum
From Tables to Graphs: Schema and Network Analytics
- Bipartite, multi-edge and heterogeneous graph design
- Degree, PageRank, betweenness and Louvain/Leiden community detection
Graph Feature Engineering and Embeddings
- Neighbourhood aggregations, ego-net statistics and risk propagation
- Random-walk embeddings with Node2Vec and DeepWalk
Message Passing and Graph Neural Networks (GNNs)
- GCN, GraphSAGE and Graph Attention Networks (GAT)
- Over-smoothing and choosing neighbourhood depth
Node and Link Prediction with PyTorch Geometric
- Modelling multi-type nodes and relations with HeteroData
- Link prediction, negative edge sampling and temporal splits
Scaling and Enterprise Case Study
- Mini-batch training with NeighborLoader and inference latency
- Case benchmark: GBDT + graph features vs end-to-end GNN
Other Courses in This Category
-
In-context learning on tabular data with TabPFN v2, plus zero-shot and fine-tuned time-series forecasting with Chronos, TimesFM, Moirai and Lag-Llama, benchmarked against GBDTs and classical models.
- Live Course
- 5 modules
-
Supervised and unsupervised learning, train/test splits, overfitting, core metrics and first models in scikit-learn, so technical teams can frame a business problem as an ML task and build a sound baseline.
- Live Course
- 5 modules
-
Gradient boosting, ensembles and stacking, hyperparameter optimisation with Optuna, imbalanced data, probability calibration and custom loss functions for experienced data scientists building production-grade models.
- Live Course
- 5 modules
-
Takes analysts beyond classic reporting with segmentation, driver analysis, anomaly flagging and simple predictive models in Python, scikit-learn and SHAP, producing analyses that tie insight to business decisions.
- Live Course
- 5 modules
-
Categorical encoding, time and window-based features, target encoding, interaction and text features, feature selection and feature stores: a systematic approach to the step that most improves model performance on tabular data.
- Live Course
- 5 modules
-
Choosing the right metrics, evaluation tied to business KPIs, error analysis, fairness checks and model explanation with SHAP, LIME and PDP, so models can be presented to stakeholders reliably and understandably.
- Live Course
- 5 modules
Get in touch
