How lijfrank-open/graphsmile works
Project type and purpose
• This is a research‐oriented machine‐learning pipeline implementing a graph‐based multimodal emotion‐recognition model (“GraphSmile”) from the referenced paper.
• It is not a user‐facing web app or microservices suite but a standalone Python experiment suite: you feed it datasets, it trains a model, and you collect outputs (models, logs, metrics).
File structure
Top‐level files and their responsibilities:
• run.py – CLI entrypoint; parses arguments, sets up environment, invokes training/inference.
• dataloader.py – dataset readers (IEMOCAP, MELD, CMU-MOSEI), batching logic.
• utils.py – shared helpers (path management, logging, metric calculators).
• module.py – graph construction & feature‐engineering modules, multimodal fusion logic.
• model.py – PyTorch model definition (neural layers, forward pass).
• trainer.py – training loop, optimization, checkpoint saving, evaluation.
• requirements.txt – external dependencies (PyTorch, numpy, etc).
README for additional insights
• Describes datasets, CLI flags (epochs, learning rate, batch size, graph hyperparams).