How ddw2aigroup2cqupt/grig works
Project type and purpose
• This is a research‐style machine‐learning library (“GRIG”) that implements a graph‐based image representation algorithm (“Granular‐ball”) and applies it to two standard benchmarks (MNIST and CIFAR‑10).
• Key indicators: top‐level folders MNIST/ and cifar-10/, each containing data archives, loaders, model definitions, and training scripts. README title “Graph‑based Representation for Image based on Granular‐ball.”
File structure
• Top level
• LICENSE, README.md (project description)
• MNIST/ (data + nets + main script + readme)
• cifar-10/ (data_loader/, data archives, ParserArgument.py, model.py, train.py, utils.py)
• MNIST module
• data/ (zipped image sets) → uncompress → feed to main_mnist_ceng33.py
• nets/ (Net1.py, Net2.py, Net3.py) → different GNN architectures or network variants
• main_mnist_ceng33.py (entry point for training/eval)
• CIFAR‑10 module
• Data_loader/ (ClassificationBase_dataset.py, transforms.py) → PyTorch‐style dataset & transforms
• ParserArgument.py (CLI argument parsing) → config for model, training hyperparams
• model.py (GNN model definition)
• train.py (training loop)
• utils.py (metrics, logging, checkpoint helpers)
• No external microservices; all components live in one codebase. Follows a pipeline pattern: data → preprocessing/graph‐generation → model → train/evaluate.
Insights from README
• Confirms focus on a graph‐based representation algorithm (“Granular‐ball”).