ML Models
Model configurations allow central listing and management of all data science models.
Model definitions allow reuse of data science libraries and model classes, running algorithms with different set of inputs and parameters.
Initial definition of a model includes:
Name: A descriptive name
Description: Detailed description of the model
Tags: Descriptive tags for the model
Status: Whether this model should be deployed or not
Version: Current version of the model, which is used for deciding whether real-time model handlers need to update their current assets or not
Domain: Categorization of the model based on business domain
Assets Root: Root path in file system for storing and retrieving saved model assets
Assets Directory: Directory under root for saved model assets
Scheduler: Scheduler to use for regular training or execution of the model (e.g. Airflow)
Schedule Status: Whether schedule is currently active or paused
Comments: Historical list of comments, which typically includes information on model update reasons or findings
Parameters: Model level parameters which can be for training or inference purposes
Each model consists of a series of steps, which allows building sequential model pipelines, although most models may consist of a single step. Initial definition of a step includes:
Id: Unique identifier for the step
Name: A descriptive name
Description: Detailed description of the step
Has Assets: Whether step has stored assets or not
Assets Root: Root path override for the model assets root
Assets Identifier: File name for the step assets
Steps also have parameters that are defined in a very dynamic manner to allow configuration of all model types, and may include input, output, training details as well as the class and package names for Python libraries.
Last updated