In a previous post, we shared our approach to developing CDAP into a full-spectrum machine intelligence platform with 150+ plugins, from deep and machine learning, business rule processing to natural language processing and more.
Many ML/AI platforms focus on model building and training, and often operationalization is an afterthought or left to someone else.
In contrast, Google CDAP has all batteries included to implement a complete ML/AI model lifecycle, enclosing a model registry, automated version management, tracking and serving.
Model training on the one hand and serving models in production on the other hand: What is the main difference?
With CDAP, both phases can be supported by configurable data pipelines that have plugins in common for pre- and post-processing. Model specific plugins, however, differ. For training purposes, we use model plugins that work as producers. In production, we leverage consumer plugins.
No longer additional platforms like MLflow or others. It is all in one, and that's what makes machine learning work.
Top comments (0)