- Single model generation in VantageCloud Enterprise and VantageCloud Lake might reach memory limits when trained on very large data.
- split() function of model_selection module returns generator of train and test data when used locally with scikit-learn, but with teradataml OpenSourceML module, it returns generator of train and test as teradataml DataFrames.
- Functions like predict, transform, and so on, return only labels/predictions in scikit-learn. They return the features data along with labels/predictions in teradataml OpenSourceML.
- list and delete model in Model Operations are not supported.
- To list the deployed models, run DataFrame(“opensourceml_models”) or SELECT * FROM opensourceml_models.
- To delete the models, run SQL query to delete the rows from the table “opensourceml_models”.
- Classes that take generators, iterators, functions, and callable objects as arguments are not supported.
- feature_extraction module is not supported.
- Performance issues can result during single model generation when trained on very large data.
- On an AI on-prem system, if the Apply table operator is not available and you attempts to execute OpenML functions, it will raise a runtime error.You can resolve the issue by setting the table_operator configuration option to "script":
from teradataml import configure configure.table_operator = "script"
See attached Supportability Matrix for teradataml OpenSourceML - scikit-learn classes and functions support.