Types of Models - Teradata Vantage

ClearScape Analytics™ ModelOps User Guide

Deployment
VantageCloud
VantageCore
Edition
Enterprise
IntelliFlex
VMware
Product
Teradata Vantage
Release Number
7.1
Published
December 2024
ft:locale
en-US
ft:lastEdition
2024-12-13
dita:mapPath
zdn1704469623418.ditamap
dita:ditavalPath
azq1671041405318.ditaval
dita:id
rgn1654191066978
lifecycle
latest
Product Category
ClearScape
ModelOps supports the following types of models:
  • Git model
  • Bring your own model (BYOM)

Git model

The code behind this type of model is stored in a Git repository.

ModelOps requires a specific Git repository structure that will automatically detect and register the models in a given repository.

Use the Teradata modelops (tmo) command-line-tool published in https://pypi.org/project/teradatamodelops/ to generate the required folder structure. An example of this is published in https://github.com/Teradata/modelops-demo-models.

An example folder structure follows.

model_definitions/
     <model_name>
          model_modules
               __init__.py
               requirements.txt
               training.py
               evaluation.py
               scoring.py
          config.json
          model.json
Where:
  • model_definitions contains all the models from a Git repository associated to a project.
  • model_modules is the subfolder associated to a specific user-defined model_name.
  • __init__.py is a python file that is used to load any specific library or binary when the python environment for a model job is created. By default, it is empty.
  • requirements.txt contains the names and versions of the python/r libraries.
  • training.py is the script that the training job will run. You must define the train function in this script.
  • evaluation.py is the script that the evaluation job will run. You must define the evaluation function in this script.
  • scoring.py is the script that the scoring job will run. You must define a score function in this script for batch and for RESTful deployments.
  • config.json contains hyper parameters that the trained model uses with the default values.
  • model.json contains an id (uuid) generated by the tmo command line or by user, name, description and language of the model.

All changes to the model code and metadata occur through Git. ModelOps only ever reads this information. Git models can use the Python, R and SQL languages.

See Supported Languages for supported versions of these languages.

Bring your own model (BYOM)

Use this model type when you train a model on a third-party platform or in an external application, and want to evaluate, score, or monitor it with ModelOps.

ModelOps supports importing Python, R and SAS models, in addition to the model types supported in the Vantage BYOM product (Teradata Vantage™ - Bring Your Own Model User Guide).