ForML is a framework for researching, implementing and operating data science projects.
Use ForML to formally describe a data science problem as a composition of high-level operators. ForML expands your project into a task dependency graph specific to a given life-cycle phase and executes it using any of its supported runners.
Solutions built on ForML are naturally easy to reuse, extend, reproduce, or share and collaborate on.
Not Just Another DAG¶
Despite DAG (directed acyclic graph) being at the heart of ForML operations, it stands out amongst the many other task dependency processing systems due to:
Its specialization on machine learning problems, that is wired right into the flow topology.
Concept of high-level operator composition which helps to wrap complex ML techniques into simple reusable units.
An abstraction of runtime dependencies allowing to run the same project using different technologies.
ForML started as an open-source project in response to ever painful transitions of datascience research into production. While there are other projects trying to solve this problem, they are typically either generic data processing systems too low-level to provide out-of-the-box ML lifecycle routines or special scientific frameworks that are on the other end too high-level to allow for robust operations.
- Input & Output
- Data Source DSL
- Interactive Mode
- Operator Architecture
- Operator Unit Testing
- Flow Library