How keitabroadwater/gnns_in_action works
Project Type and Its Purpose
• This repository serves as the code and data companion for the book “Graph Neural Networks in Action.” It is primarily an educational resource that organizes Jupyter notebooks, datasets, and supplemental files by chapter.
• Each chapter’s folder contains experiments and examples illustrating core concepts of graph neural networks (from basic graph embeddings to advanced topics like spatial-temporal GNNs, graph autoencoders, and even scalability concerns).
File Structure
• At the top level, you see common configuration files (.gitignore, .gitattributes, README.md) and a set of directories named by chapter (chapter_2, chapter_3, …, chapter_8). This organization reflects a modular design where each chapter is largely self-contained.
• Within chapters, you find:
• Jupyter notebooks (".ipynb") that give hands-on examples and implementations.
• Data files (e.g., CSVs, zipped datasets, npy files) that are used as inputs for the experiments.
• Secondary directories (like “archived_files” in chapter_2 or “data” in chapter_4) which hold legacy or large files.
• This reveals that the architecture is divided functionally by chapter/topic – each acting as a module covering a specific aspect of GNNs.
Insights from the README
• The README outlines the book’s structure across three parts: basic concepts, core GNN models, and advanced topics.
• It provides a sequential roadmap of topics (from graph embeddings in chapter 2, moving through various GNN architectures in chapters 3–5, to dynamic/large-scale graphs in chapters 6–8). This sequential progression can be represented in the system design through clearly labeled sections.
• The resource emphasizes that the repository supports both learning and hands-on experimentation—this must be depicted by showing the interplay between “code modules” (the notebooks) and “data sources” (the supplemental datasets).
a. Main Components to Include:
• A “Repository Hub” node representing the entire codebase.
• A “Data Store” node covering datasets (CSV files, zipped data like YelpChi.zip, JSONs, npy files) that are accessed by multiple chapters.
• “External Resources” for datasets that come from sources outside the repository (for instance, the OGB datasets mentioned in the README).
• “Notebook Execution Environment” (e.g., Jupyter Notebooks) indicating that the experiments and code examples are intended to be executed interactively.
b. Relationships and Interactions:
• Link data flows from the “Data Store” to the appropriate chapter modules where data processing, model training, or experimentation occurs.
c. Architectural Patterns and Design Principles:
• The design is modular and loosely coupled as each chapter is self-contained.
• “layered” educational approach: foundational concepts in early chapters leading to advanced methods, which can be depicted as a progressive flow.
• Separation between code (Jupyter notebooks), configuration/data files (in various subdirectories), and external datasets.
d. Technologies and Frameworks: