The update function now accepts the following new parameters:
- alter_operation: Specifies the type of operation to perform for update (
ADD/DROP). - update_style: Specifies the style to be used with
alter_operation(MAJOR/MINOR). - object_names: Specifies the table name/
teradatamlDataFrame to be indexed for vector store. - ignore_embedding_errors: Specifies whether to ignore errors during embedding generation.
- database_name: Specifies the database name where vector is created. User should have create table permission on this database.
- batch: Specifies whether to use batch processing for embedding generation. Applicable only for
aws.
- Batch support is enabled for
similarity_search,ask, andprepare_responseAPIs. - A new API
get_batch_resultis added to retrieve results after triggering the APIs in batch mode.
- Content-based Vector Store can now be created on multiple tables/views.
- The
object_namesparameter can be provided with a list of strings or list ofDataFramewhile creating a Vector Store.
- Creating, updating, and deleting Vector Stores is now asynchronous.
- The
statusAPI can be used to check the status of these operations.
- Errors that were previously displayed in a DataFrame column are now thrown as
TeradatamlException. - DataFrame displayed to the user, except for
status(), is now a teradataml DataFrame instead of a pandas DataFrame for better control. - Implemented
get_detailsfunction which gets the details of the Vector Store.
- Fixed disconnect issue to prevent "failed to disconnect" errors.
- Issue Reference: ELE-7574: [VS EARLY ACCESS] - teradataml: Inconsistent/erratic messaging behavior when disconnecting from a VS.
- Fixed a bug where the
session_idwas not refreshed when the JWT was refreshed.- Issue Reference: ELE-7624: [VS EARLY ACCESS] - Vector Store initialize or create fails after changing the authentication key and token.
createnow displays an appropriate message if creation fails and is reinitialized.