Wrapper for Models

In this platform version, in collaboration with the onesait data team, support for ML models has been further developed. Improvements have been included in the platform Python client in model management, so that it is homogeneous and allows data scientists to focus on the creation of the model itself.

For use, both in platform notebooks, and from outside (for example, a Python or Jupter shell, or in production in a microservice), the Python library has evolved with new features that allow to:

Transparently store/retrieve the files of the models generated by the trainings in the file repository.
Use ontology data as data input for models.
Manage a meta-information ontology of the different associated models and the recovery of the most accurate model.
Log the entire process centralized in the platform audit.

This library will be, by default, in the notebooks of the next platform version, being published in version 1.4.0 of the pypi repository. Like the previous versions, it will be installable anywhere through the pip command:

In the future, continuing to work together with Onesait data, it will evolve even more and this new piece will serve as a seed for an automated management of the life cycle of models from development to deployment in production.