KAML-D is mainly useful in the area of collaboratively developing and running cloud native Machine Learning (ML) apps, for example:
Before an ML feature can be integrated into an app, a data scientist needs to, for example:
In each of the steps the data scientist might perform many iterations, trying out a number of different approaches and/or using different test and training datasets. With KAML-D, rather than manually creating new versions of datasets and models, you can quickly generate new versions and also go back in time.
As we’ve discussed in the motivation, there’s a double divide (data scientists/developers/ops) and KAML-D is written with mitigating this in mind. By providing a unified way of handling both datasets and models across different roles, you can expect to more smoothly transition an app to production. With KAML-D, everyone involved can focus on their role while benefiting from a common language and environment, from experimentation to prod.
Datasets and models are often handled separately, by different people and/or roles and using different methods, from no versioning to purely manual versioning to proprietary methods. With KAML-D, you get a unified way for different roles such as data scientists or developers to handle datasets and models in a unified manner, based on industry-leader Dotmesh.