Skip to main content

CI/CD/CT/CM

Updated May 15, 2023 ·

Overview​

DevOps focuses on collaboration between developers and IT operations to improve the software development process.

  • Combines software development and IT operations
  • Enhances communication, optimizing delivery speed

These DevOps practices, including automation, extend to machine learning (ML) applications, creating MLOps for ML development and deployment.

  • Automates ML model training and deployment
  • Enables faster, more efficient model management

Decaying Performance​

ML model performance degrades over time due to changes in data. Constant training (CT) helps keep models updated.

  • Models retrained regularly
  • Ensures models remain accurate
  • Adapts to new data patterns

CT helps maintain model effectiveness by continuously retraining models as new data becomes available.

CI/CD/CT/CM in MLOps​

CI/CD and ML-specific tests form the backbone of an automated MLOps system, enabling smooth integration and deployment.

  • CI/CD ensures seamless updates
  • Integrates code and fixes into production
  • ML-specific tests included in CI
  • Checks model accuracy and validates performance

CI/CD/CT/CM integrates the ML lifecycle, ensuring that models remain up-to-date and high-performing.

CI/CD​

Continuous integration (CI) and continuous deployment (CD) help improve software quality and speed up delivery.

  • Integrates code from multiple developers
  • Runs automated tests on each commit
  • Deploys automatically after successful tests

For more information, please see CICD Overview.

Continuous Training (CT)​

Continuous training (CT) allows models to stay accurate and effective as new data is added.

  • Retrains models regularly
  • Adapts to new data
  • Ensures up-to-date performance

Continuous Monitoring (CM)​

Continuous monitoring (CM) helps track model and data quality to ensure everything works as expected.

  • Monitors data and model performance
    • Detects issues like data drift
    • Triggers fixes when needed

CM ensures that issues like data drift or model performance degradation are quickly identified and addressed.

Automation First​

A focus on automation throughout the ML lifecycle ensures models are updated quickly and easily.

  • Automates model updates
  • Simplifies maintenance
  • Speeds up model deployment

Automated Incident Response​

An automated response pattern addresses issues that arise in ML systems and ensures minimal disruption.

  • Detects system issues automatically
  • Reduces manual intervention
  • Improves system reliability

CI/CD/CT/CM: Full MLOps Integration​

CI/CD, CT, and CM together creates a fully integrated MLOps system.

  • Continuous integration and deployment

    • Ensures smooth development
    • Keeps models up-to-date

  • Continuous monitoring and training

    • Tracks and improves model performance
    • Ensures accuracy

Together, CI/CD/CT/CM make MLOps systems robust, automating the entire process from model creation to deployment and maintenance.