DataTerminology.comNEW
Systematically logging parameters, metrics, artifacts, and code versions to reproduce and compare model experiments.
DESCRIPTION
Experiment tracking records the who, what, and how of ML runs to enable reproducibility, comparison, and auditability. Typical systems capture parameters, metrics, artifacts, environment, and code version, with a UI for visualization. It supports hyperparameter optimization, ablation studies, and performance baselining. Integrations with registries and CI/CD streamline deployment. Good tracking reduces experiment debt and accelerates model iteration.
ALTERNATIVE DEFINITIONS
MLflow
MLflow Tracking is an API and UI for logging parameters, code versions, metrics, and artifacts when running machine learning code and for later visualizing results.
Explicitly defines tracking scope: parameters, code versions, metrics, artifacts and visualization.
Weights & Biases
Track experiments by logging hyperparameters, metrics, and outputs to visualize performance and compare runs.
Focuses on comparison and visualization benefits across runs.
RELATED TERMS
MLflow Weights & Biases Metadata Repositories Reproduction Version Control Observability Artifact Management Experiment Management
COMPETING TERMS
KNOWLEDGE AREAS
MLOps Platform & Tooling Observability & Telemetry Testing & Validation
TARGET AUDIENCE
MLOps Engineer AI Engineer Data Scientist Software Developer
EXTERNAL SOURCES
- Wikipedia - MLflow
- MLflow - MLflow Tracking docs
- Weights & Biases - Experiment tracking guide
- Microsoft Learn - MLOps practices overview
TOP QUESTIONS
What is experiment tracking in ML?
It’s systematic logging of parameters, metrics, artifacts, and code to reproduce and compare experiments.
Category: Definition
Why track experiments?
It ensures reproducibility, accelerates iteration, and supports auditing and governance.
Category: Relevancy
What should be logged?
Hyperparameters, metrics, artifacts, environment details, code versions, dataset lineage, and notes.
Category: Requirements
What tools enable experiment tracking?
MLflow, Weights & Biases, Vertex ML Metadata, and custom platforms.
Category: Resources
How does tracking integrate with CI/CD?
Runs are logged during training; results inform promotion decisions and deployment pipelines.
Category: Collaboration
What are common mistakes in tracking?
Not logging dataset versions, random seeds, or environment; inconsistent naming hampers comparison.
Category: Mistakes
How do you visualize tracked experiments?
Use dashboards to compare metrics, plot hyperparameters, and review artifacts and logs.
Category: Best practices
What’s the relationship to a model registry?
Tracking records experiments; registries manage promoted models, stages, and deployments.
Category: Difference
How to ensure governance in tracking?
Implement access controls, retention policies, and audit trails for experiment records.
Category: Management