tahamajs/Sysmem2_in_AI/ComputerAssignments/CA10_knowledge_graphs/notebooks - Storage Bucket

CA10 Knowledge Graphs - Notebooks

📚 Available Notebooks

1. 01_Advanced_KG_Embeddings.ipynb

Advanced knowledge graph embedding techniques:

• SimplE, RotatE, ConvE
• Graph Attention Networks
• Transformer-based embeddings

2. 02_Graph_Neural_Networks.ipynb ✨ NEW & COMPLETE

Comprehensive GNN tutorial for Knowledge Graphs:

• GCN: Graph Convolutional Networks
• GAT: Graph Attention Networks
• GraphSAGE: Sample and Aggregate
• R-GCN: Relational GCN for multi-relational graphs

Features:

• ✅ 25+ cells with code and markdown
• ✅ Complete implementations of 4 GNN architectures
• ✅ Training pipeline for link prediction
• ✅ Attention weight visualization
• ✅ t-SNE embedding visualization
• ✅ Model comparison and benchmarking
• ✅ Comprehensive mathematical formulas
• ✅ Production-ready code

3. CA10.ipynb

Main project notebook with all CA10 implementations

---

📖 Documentation

Main Documents:

• GNN_NOTEBOOK_GUIDE.md: Complete 300+ line guide covering:

  • Detailed theory for each GNN architecture
  • Mathematical formulas and intuitions
  • Implementation details
  • Usage instructions
  • Troubleshooting
  • Research applications
  • References

• NOTEBOOK_COMPLETION_REPORT.md: Technical report with:

  • Implementation statistics
  • Feature checklist
  • Code metrics
  • Testing status

---

🚀 Quick Start

Running 02_Graph_Neural_Networks.ipynb:

# 1. Install dependencies
pip install torch torch-geometric scikit-learn matplotlib seaborn

# 2. Launch Jupyter
cd /path/to/CA10_knowledge_graphs/notebooks
jupyter notebook 02_Graph_Neural_Networks.ipynb

# 3. Run all cells or execute step by step

Prerequisites:

# Core dependencies
torch>=2.0.0
torch-geometric>=2.4.0
scikit-learn>=1.3.0
matplotlib>=3.7.0
seaborn>=0.12.0
numpy>=1.24.0

---

📊 Notebook Structure

02_Graph_Neural_Networks.ipynb Sections:

1. Introduction - Overview and objectives
2. Setup - Imports and configuration
3. Knowledge Graph - Create and visualize KG
4. GCN - Theory + Implementation
5. GAT - Attention mechanism + Visualization
6. GraphSAGE - Sampling and aggregation
7. R-GCN - Multi-relational reasoning
8. Training - Link prediction pipeline
9. Comparison - Model benchmarking
10. Visualization - t-SNE embeddings
11. Summary - Key takeaways

---

🎯 Learning Objectives

After completing these notebooks, you will understand:

• ✅ Graph Neural Network fundamentals
• ✅ Message passing framework
• ✅ Attention mechanisms in graphs
• ✅ Aggregation strategies
• ✅ Multi-relational graph reasoning
• ✅ Knowledge graph embedding techniques
• ✅ Link prediction tasks
• ✅ Model comparison and selection

---

🔬 What's Implemented

GNN Architectures:

Model	Layers	Parameters	Best For
GCN	2	~50K	Simple graphs
GAT	2 (4 heads)	~75K	Varying importance
GraphSAGE	2	~55K	Large-scale
R-GCN	2	~60K	Multi-relational

Features:

• Forward propagation
• Training loops
• Loss computation
• Accuracy metrics
• Visualization tools
• Model comparison

---

📈 Visualizations Included

1. Knowledge Graph Structure: NetworkX visualizations
2. Attention Weights: Histogram + bar charts
3. Training Curves: Loss over epochs
4. Embeddings: t-SNE 2D projections (2x2 grid)
5. Model Comparison: Parameter and performance tables

---

🎓 Educational Value

Target Audience:

• Students learning GNNs
• Researchers working with Knowledge Graphs
• Engineers building graph-based systems
• Anyone interested in graph neural networks

Difficulty Level:

• Prerequisite: Basic Python, PyTorch, Graph theory
• Level: Intermediate to Advanced
• Time: 2-3 hours to complete

---

💡 Key Concepts Covered

Theory:

• Graph convolutions
• Message passing
• Attention mechanisms
• Aggregation functions
• Basis decomposition
• Over-smoothing
• Inductive learning

Practice:

• PyTorch implementation
• PyTorch Geometric usage
• Training on Knowledge Graphs
• Link prediction
• Embedding visualization
• Model selection

---

🔧 Customization

You can easily modify:

• Embedding dimensions
• Number of layers
• Hidden dimensions
• Dropout rates
• Learning rates
• Aggregation functions
• Number of attention heads
• Number of bases (R-GCN)

---

📚 References

Papers:

1. Kipf & Welling - "Semi-Supervised Classification with GCNs" (ICLR 2017)
2. Veličković et al. - "Graph Attention Networks" (ICLR 2018)
3. Hamilton et al. - "Inductive Representation Learning" (NIPS 2017)
4. Schlichtkrull et al. - "Modeling Relational Data" (ESWC 2018)

Resources:

• PyTorch Geometric: https://pytorch-geometric.readthedocs.io/
• Stanford CS224W: http://web.stanford.edu/class/cs224w/
• Distill.pub: https://distill.pub/

---

🐛 Troubleshooting

Common Issues:

ImportError: torch_geometric not found

pip install torch-geometric torch-scatter torch-sparse

CUDA out of memory

• Reduce batch size or embedding dimension
• Use CPU: device = 'cpu'

Slow training

• Use GPU if available
• Reduce number of layers
• Use GraphSAGE with sampling

See GNN_NOTEBOOK_GUIDE.md for more troubleshooting tips.

---

🎉 Status

• ✅ Complete: All notebooks functional
• ✅ Documented: Comprehensive guides
• ✅ Tested: Code verified
• ✅ Production-Ready: Professional quality

---

📞 Support

For questions:

1. Review GNN_NOTEBOOK_GUIDE.md
2. Check main CA10 README
3. Consult PyTorch Geometric docs

---

Happy Learning! 🚀📚

Part of the System2_in_AI CA10 project
Last updated: October 10, 2025