How whu-aise/tvdiag works
Main Components
• Core Failure Diagnosis Framework: The heart of the project resides in the “core” folder. Here you find the primary diagnosis module (TVDiag.py), data augmentation (aug.py), task-oriented modules (ita.py), and various loss functions (in the “loss” subdirectory). The “model” folder holds the different submodules (Classifier, Encoder, MainModel, Voter, and backbone implementations like FC and sage), which together constitute the machine learning models for failure identification.
• Configuration: The “config” directory (with exp_config.py) defines the experiment and model parameters. These configurations drive how experiments run (e.g., which dataset to use, whether to dynamically weigh losses, etc.).
• Data Management: The “data” folder covers two datasets (gaia & sockshop). Notice the consistent structure (label.csv, raw, and tmp folders) indicating raw data ingestion, processing, and temporary storage for intermediate results.
• Extractor Module: In the “extractor” folder, you have scripts for extracting and processing raw logs, metrics, and traces. This includes specialized modules such as the “drain” folder and event-specific extractors.
• Helper Utilities