Skip to main content

2.0.0 release notes

This release represents AI-Link's second major release of 2022. It introduces new functionality, bridging the gap between BI and AI to make AI predictions consumable, relevant, and explainable to drive business value.

Spark Support (beta)​

Native support for Spark's parallel compute power and objects, like PySpark dataframes (no longer just pandas). Get more ML predictions in the hands of decision-makers faster: anyone utilizing Databricks or other Spark-based environments in their ML workflows can take advantage of AI-Link functionality for large scale datasets.

Consider the following:

  • This is available for WriteBack and WriteTo, however this is limited to writes via JDBC. While a user can read in data into a PySpark dataframe and writeback predictions via JDBC, the user is not able to write a PySpark dataframe into a locally configured Spark metastore. All customers using Databricks notebooks and any other database will not have this issue.
  • This is not yet available for UDFs.

WriteTo (beta)​

Automatically create a semantic layer to enable analysis of ML predictions in BI Tools. This allows you to move away from screenshots and static tables of ML output to dynamic representation of ML predictions. This way you can empower business stakeholders, move more ML models into production, and facilitate additional data modeling to uncover new descriptive and predictive insights.

Semantic inference​

Embed ML models in AtScale to make predictions on-demand through BI Tools with no data movement, extracts or copies. Save time, costs, and compute resources associated with traditional ML deployment by registering ML models in AtScale to take advantage of BI connections to the query optimization engine and correlation of ML predictions with business context.

  • UDF Support for Snowpark (beta):

    Integration support to capture predictions on-demand from ML models created in Snowpark UDFs into BI Tools without moving data from Snowflake. For details, see /using-ai-link/ml-deployment-in-bi/semantic-inference-udf-integration/index.

  • Non-UDF Support (beta)

    Transpose linear regression models into the semantic layer on any relational database (like Postgres, MySQL, and so on) to capture ML predictions in BI Tools on-demand. For details, see /using-ai-link/ml-deployment-in-bi/semantic-inference-non-udf-integration/index.

New documentation Web sites​

Refreshed Web sites with new documentation to help users search and consume for content related to AI-Link.

Syntax changes​

A number of functions were to improve readability and consistency across the code base. These include the following (old -> new):

  • enums.TableExistsActionType -> enums.PandasTableExistsActionType

  • Connection.get_tables -> Connection.get_connected_tables

  • Data_Model.list_fact_datasets -> Data_Model.get_fact_dataset_names

  • Data_Model.list_columns -> Data_Model.get_column_names

  • Data_Model.add_dataset -> function replaced with orchestrators in client

  • SQLConnection.dispose_engine -> SQLAlchemyConnection.dispose_engine

    Moved as not included for all SQLConnection object implementations.

  • utils.eda.pca -> eda.pca

    eda directory created and pca converted from function to module

  • feature_utils.create_calculated_column -> project_utils.create_calculated_column

    Moved and removed project publish parameter

  • metadata_utils.list_all_numeric_features -> metadata_utils.get_all_numeric_feature_names

  • metadata_utils.list_all_categorical_features -> metadata_utils.get_all_categorical_feature_names

Directory structure changes​

The underlying directory structure of the AI-Link package has been updated to improve scalability. Notable changes are the creation of the 'connections' sub-folder in the 'db' directory. And the creation of the eda and base directories.