Logging, Configuration, and Reproducibility

Under development

This lesson is part of the course scaffold and is being actively written. The learning objectives and outline below define its final scope.

Learning objectives

By the end of this lesson you will be able to:

• Instrument a research pipeline with structured logging, choosing appropriate levels and context fields for each event.
• Externalize run parameters into YAML or TOML configuration validated with pydantic-settings, with environment-specific overrides.
• Make a research run reproducible end to end: pinned seeds, recorded configuration, and captured run metadata.
• Audit an existing script for hidden state — global RNGs, hardcoded paths, mutable defaults — that breaks reproducibility.

Outline

1. Structured logging — JSON logs, context fields, correlation across pipeline stages
2. Log levels — what belongs at DEBUG, INFO, WARNING, and ERROR in a research pipeline
3. Configuration formats — YAML vs TOML, layering, per-environment overrides
4. pydantic-settings — typed validation, environment variables, secrets handling
5. Reproducible runs — seeds, RNG discipline, recording config and code version per run
6. Putting it together — a configurable, logged, reproducible research script

Prerequisites

• Typing, Dataclasses, and Code Structure