Mandala is a powerful tool that automates ML experiment tracking with the @op decorator and ComputationFrame data structure in Python. Unlike other popular tools, Mandala synchronizes persistence, versioning, and querying at the function level for efficient development. It can transparently track individual elements across collections and calls, simplifying storage management. The @op cache is invalidated based on compatible versions of dependencies, providing flexibility when modifying code. Mandala is still in alpha, with known performance limitations. The ultimate goal is to seamlessly integrate storage logic with Python code, creating a simple and reliable way to manage computational artifacts. It combines memoization, computation frames, and versioning for a holistic approach to computational artifact management.
https://github.com/amakelov/mandala