How algpy/conceptgcd works
(minimal) file structure
• data/ (raw, processed)
• src/ (model definitions, training loops, evaluation)
• notebooks/ (exploratory code)
• configs/ (experiment settings)
• scripts/ (job submission, inference scripts)
• tests/ (unit and integration tests)
README for additional insights
• The README promises a “concept-driven approach” to class discovery, implying components for concept extraction, embedding, clustering, and open-set evaluation.
• No technologies are named yet, but typical choices might include PyTorch/TensorFlow, Hydra/argparse for configs, MLFlow or TensorBoard for logging, and possibly a CLI or Jupyter interface.