MLOps vs DevOps

2022, September, 15

Data ScienceExplainer

MLOps is DevOps for Machine Learning Projects

It has similarity with DevOps but also differs in the following ways:

Deployment

  • DevOps ensures that apps are tested, built and deployed.
  • MLOps has to take a trained ML model and create an API for it and deploy the API. The datasets used to create the ML model also need to be managed. As updated datasets become available, check if models need to be re-trained.

Monitoring

  • DevOps needs to ensure app is up and available. Periodically run sanity checks to ensure that app is returning expected results.
  • MLOps needs to ensure app availability. Also need to check for any data drifts.

Versioning

  • DevOps needs to ensure that app is versioned.
  • ML model, hyper-parameterms used during training and datasets are closely tied together. MLOps need to version the models while managing their dependencies. Model versions could also be used to compare model results during training phase.