Course Wiki

Experiment Tracking

Recording every training run's parameters, metrics, and artifacts so results can be compared and reproduced.

Experiment tracking solves the ‘which of the fifty training runs produced this model?’ problem. Module 2 instruments the NY Taxi training code with MLflow: every run records its parameters (features, hyperparameters), metrics (RMSE on validation), and artifacts (the pickled model), all browsable in the tracking UI.

With runs logged, model selection becomes a query instead of memory: sort runs by metric, compare parameter sets side by side, and promote the winner. The module treats this as the first non-negotiable MLOps practice — before pipelines, before deployment, you need to know what you trained and how it performed.

Taught in

• LLM Zoomcamp

  • Module 4: Evaluation
    • Generating Ground Truth Data
    • Next Steps

• MLOps Zoomcamp

  • Module 2: Experiment Tracking & Model Management
    • Experiment tracking intro
    • Experiment tracking with MLflow
    • Homework
  • Module 07: Course Project
    • Course Project ## Related concepts

• MLflow

• Model Registry

• MLOps Maturity Model

• Workflow Orchestration

• Model Monitoring