$ show dssg/triage
Triage is a Python framework for end-to-end machine learning workflows focused on temporal prediction problems, particularly in social good domains. It provides modular components for feature generation (architect), model training/evaluation (catwalk), experiment orchestration, model selection (audition), and post-modeling analysis. The system uses PostgreSQL for storing experiment metadata, feature matrices, and model predictions, with SQLAlchemy ORM and Alembic migrations managing the schema. Key entry points include the CLI (`triage experiment`, `triage audition`) and `manage.py` for development tasks. The architecture emphasizes reproducibility through configuration-driven experiments, temporal validation to prevent lookahead bias, and metadata tracking. While powerful, the system has complexity from multiple interdependent components and external service dependencies (PostgreSQL, Redis for RQ).
The LLM identified 5 files worth highlighting for reuse:
Provides SerializableDbEngine class that wraps SQLAlchemy Engine to make it picklable for multiprocessing. Solves the common problem of sharing database connections across processes in a self-contained, reusable way. Demonstrates non-obvious skill in handling SQLAlchemy's connection pooling and pickle compatibility.
Implements a storage abstraction layer with Store base class and S3Store/FSStore implementations. Uses factory pattern for storage medium selection and context managers for caching mechanisms. Cleanly separates local filesystem and S3 storage concerns, making it easily extractable for projects needing configurable storage backends.
Defines FeatureGenerator class that orchestrates feature generation from multiple sources using strategy pattern for different feature types. Modular design allows easy extension with new feature generators. Self-contained component that could be adapted for other feature engineering pipelines with minimal modification.
Implements Timechop class for generating temporal validation time windows. Core component for temporal cross-validation that outputs standardized time window definitions consumable by downstream components. Low coupling (only depends on datetime operations) makes it highly reusable in any temporal modeling project.
Defines registry of selection rules (strategy pattern) for model auditioning. Clean, extensible pattern for defining model selection criteria based on temporal performance metrics. Easy to extend with new rules without modifying core audition logic, demonstrating good plugin architecture.