Skip to main content

COMPEL Glossary / pipeline

Pipeline

An automated execution graph connecting data ingestion, feature engineering, training, evaluation, and deployment stages — parameterized, versioned, and re-runnable.

What this means in practice

Pipelines convert one-off experiments into reproducible, auditable production workflows; pipeline-as-code is the MLOps equivalent of infrastructure-as-code.

Synonyms

ML pipeline , training pipeline , inference pipeline

See also

  • Continuous integration (ML) — Automated test and build of model code, data contracts, and training scripts on every change — extended from software CI with data-schema validation, model-schema validation, and lightweight training smoke tests.
  • Continuous delivery (ML) — Automated, governed promotion of models through lifecycle stages — development, staging, production — with gated checkpoints (evaluation thresholds, bias checks, cost thresholds, human approval where required).
  • Experiment tracking — The infrastructure and practice of recording artifacts, metrics, parameters, environment, and lineage for every experiment run — enabling later reproduction, comparison across runs, and audit.

Related articles in the Body of Knowledge