GCN on Facebok Large Page-Page Network dataset (s4722208)

[jiwhan5]

Overview

This pull request implements a GNN model for semi-supervised, multi-class node classification on the Facebook Large Page-Page Network dataset.

Data

he dataset, preprocessed into facebook.npz by UQ, includes:

Nodes: 22,470
Edges: 171,002 (forming a directed graph)
Node Features: 128-dimensional vectors, divided into four classes—politicians, government organizations, TV shows, and companies

Dataset was split into training (70%), validation (20%), and test (10%)

Model Architecture

The GCN model utilizes convolution on graph-structured data, aggregating features from neighboring nodes.
Each node updates its feature representation based on information from adjacent nodes, similar to convolution operations in CNNs.

The model is implemented in PyTorch with:

• Graph Convolution Layers: Four layers with ReLU activation after the first three.
• Input, Hidden, and Output Channels: 128 input channels (matching feature dimensions), 64 hidden channels, and 4 output channels (number of classes).

Performance:

The model achieves an average test accuracy of 93.5% across all classes.

Hyperparameters and Functions

• Optimizer: Adam optimizer
• loss: CrossEntropyLoss
• num_epochs = 350
• hidden_layer = 64
• learning_rate = 5e-4

Figures

• TSNE plot (pre - training)
• TSNE plot (post - training)
• Test, validation accuracy plot
• Test, validation loss plot
• Model structure example

Files

• modules.py : GCN Model class
• predict.py : Run the saved model on the dataset
• train.py : training and testing functions and run them after loading data
• dataset.py : function for loading data from facebook.npz
• utiles.py : functions for plotting

Depedencies

• torch
• sklearn
• matplotlib
• numpy

[gayanku]

This is an initial inspection, no action is required at this point
GNN FB Page-Page Network dataset ---- Normal Difficulty

Category		Marks	Comments
Algorithm solves the problem	5	4.5	Some overfitting seen. No early stop
Implementation functions as intended	3	3	4 layer GCN, torch_geometric/GCNConv
Good design	1	1	Modular, Reusable
Commenting	1	1	Meaningful docstrings and comments
Algorithm above Normal Difficulty	5	5
Algorithm is Hard difficulty	5	0	Normal Difficulty
Section IV : Max mark 15 from 20		14.5

TSNE / UMAP:
Good.

• Discussion: Good, with a bit more depth useful.

Suggestions:
Setup/Reproducibility/Requirements would be good to have.
Code can have some early stop, ex: train loop should save best model by keeping track of current best accuracy.
Could add a hyper parameter search or discuss how the values were obtained.

[gayanku]

Marking

Good/OK/Fair Practice (Design/Commenting, TF/Torch Usage)
	Good design and implementation.
	Spacing and comments.
	Header blocks.
Recognition Problem
	OK solution to problem. Some overfitting seen. No early stop.	-1
	Driver Script present.
	File structure present.
	Good Usage & Demo & Visualisation & Data usage.
	Module present.
	Commenting present.
	No Data leakage found.
	Difficulty : Normal. GNN Task.	-5
Commit Log
	Good Meaningful commit messages.
	Good Progressive commits.
Documentation
	Readme :Acceptable. Usage missing.	-1
	Model/technical explanation :Good.
	Description and Comments :Good.
	Markdown used and PDF submitted.
Pull Request
	Successful Pull Request (Working Algorithm Delivered on Time in Correct Branch).
	No Feedback required.
	Request Description is good.
TOTAL		-7

Marked as per the due date and changes after which aren't necessarily allowed to contribute to grade for fairness.
Subject to approval from Shakes

[hanemma7moud]

Jiwhan Oh

[shakes76]

Model files present, please remove to receive feedback marks -2

[jiwhan5]

Model files present, please remove to receive feedback marks -2

Good afternnon Shakes! Thanks for marking and teaching throughout the course. However, I am a little curious about the model file. I have talked to one of the tutors before I upload this and he said I could just update the model file as my model file is very small. (Model file not recommended to be uploaded as they are noramally very big). But I will remove the GCN.pth. Its been removed now. Please reply for this once you have checked and please tell me if this is not what u have aksed for!! Thanks!