Experiment Tracking - CI/CD Glossary Definition
Experiment tracking logs each ML runs parameters, data, metrics, and artifacts so runs can be compared and reproduced.
Experiment tracking records the parameters, code version, data version, metrics, and artifacts of each machine-learning run so results can be compared and reproduced.
Tracking turns a folder of ad-hoc runs into a queryable history, letting teams pick the best model and reproduce how it was made.
What it captures
Each run logs hyperparameters, git commit, dataset version, evaluation metrics, and the produced model artifact, usually via tools like MLflow or Weights and Biases.
Related guides
Hyperparameter Tuning - CI/CD Glossary DefinitionHyperparameter Tuning: Hyperparameter tuning searches over model configuration values (learning rate, tree de…
Reproducible Pipeline - CI/CD Glossary DefinitionReproducible Pipeline: A reproducible pipeline produces identical outputs when rerun with the same inputs, pi…
Model Training - CI/CD Glossary DefinitionModel Training: Model training is the process of fitting a machine-learning model to training data by optimiz…