Continuous delivery for ML

ML code, unlike traditional software code, is dynamic and is constantly affected by changes to the model code itself, the underlying training data, or the model parameters. Thus, ML model performance needs to be continuously monitored, and models need to be retrained and redeployed periodically to maintain the desired level of model performance. This process can be daunting and time-consuming and prone to mistakes when performed manually. However Continuous Delivery for ML (CD4ML) can help streamline and automate this process.

CD4ML is derived from the software engineering principles of continuous integration and continuous delivery (CI/CD), which were developed to promote automation, quality, and discipline and help create a reliable and repeatable process that can release software into production. CD4ML builds on and adapts this CI/CD process to ML, where data teams produce artifacts related to the ML process, such as code data and models, in safe and...