CML or Continuous Machine Learning is a novel open-source library for continuous integration and continuous delivery (CI/CD) specifically tailored for machine learning projects.
The new CI library was developed by the engineers behind DVC – an open-source version control system for machine learning projects. Together with DVC, Tensorboard, and Cloud computing, the novel CI system is set to simplify the work of researchers and engineers working in the field of Machine Learning.
CML was designed to help ML practitioners and allow them to automate parts of the project lifecycle including training experiments, evaluations, datasets and their evolution, etc. The idea behind CML was to build a library that can support GitFlow for data science (keep track of models, data, and experiments), allow generation of automated reports for experiments, and hide complex setup of external services, such as AWS, Azure, GCP, etc.
The CML library is quite flexible and offers a wide range of features, from sending reports and publishing data all the way to allocating cloud resources necessary for your ML project. Together with DVC, docker, and Terraform, MLOps can be defined as a complete ecosystem and it brings DevOps to machine learning projects.
The developers of CML have provided a few example ML projects that are implemented using CML, in order to showcase the features of the new library but also help beginners to get started with it. Detailed documentation for the CML library can be found on Github. The implementation of the library was open-sourced and can be also accessed on Github.