2.8.0 release notes

This release of AI-Link focuses improving usability and validating stability of the code base. This is the final planned major release of AI-Link for the AtScale Installer product. Versions 3.0 and later will be built to soley work with the AtScale Container product.

In this release, the following was changed:

• A removal of enums from the public facing code base. Only enums that are explicitly meant for customer use are not in the enums.py file.
• Bug fixes and additional testing.
• Additional parameters for our read and write operations.

Please refer to our API documentation for the latest syntax to use with AI-Link. See below for updates associated with this release.

New Package Requirements

• To avoid issues with subdependecies, AI-Link 2.8.0 will only work with pandas versions less than 2.2.0

User Flow Improvements

• get_data and its variants can now leverage calculation groups using the group_by parameter
• writeback and its variants and get_dimensions can now verify the AtScale service account can see write destinations using the check_permissions parameter

Bug Fixes

• writeback and its variants will now correctly handle 0 row table write operations
• the linear_regression function will correctly clean up its dependecy tables in the event of an error

Changelog for Syntax Updates

configuration

NEW PACKAGE REQUIREMENTS:

• pandas
  • version must now be less than 2.2.0

data_model.py::DataModel

UPDATED FUNCTIONS:

• writeback

  • Bug Fix

    • Adjusted behavior in the case of writing 0 row tables

  • Added parameter check_permissions

    • Determines if the function should error or proceed if the AtScale warehouse connection does not have read access to the created table

• write_feature_importance

  • Bug Fix

    • Adjusted behavior in the case of writing 0 row tables

  • Added parameter check_permissions

    • Determines if the function should error or proceed if the AtScale warehouse connection does not have read access to the created table

• get_data

  • Added optional parameter group_by
    • Create group by clause in the query for the given features. Defaults to None to group in the order of the categorical features in the query.

• get_data_direct

  • Added optional parameter group_by
    • Create group by clause in the query for the given features. Defaults to None to group in the order of the categorical features in the query.

• get_data_jdbc

  • Added optional parameter group_by
    • Create group by clause in the query for the given features. Defaults to None to group in the order of the categorical features in the query.

• get_data_spark

  • Added optional parameter group_by
    • Create group by clause in the query for the given features. Defaults to None to group in the order of the categorical features in the query.

• get_data_spark_jdbc

  • Added optional parameter group_by
    • Create group by clause in the query for the given features. Defaults to None to group in the order of the categorical features in the query.

• get_database_query

  • Added optional parameter group_by
    • Create group by clause in the query for the given features. Defaults to None to group in the order of the categorical features in the query.

• writeback_spark_jdbc

  • Added parameter check_permissions
    • Determines if the function should error or proceed if the AtScale warehouse connection does not have read access to the created table

• writeback_spark

  • Added parameter check_permissions
    • Determines if the function should error or proceed if the AtScale warehouse connection does not have read access to the created table

• get_features

  • Added support for parsing calculation groups when the use_published parameter is False

• update_aggregate_feature

  • Added functionality to support updating metrical attributes
  • Added better error handling around roleplaying

• bulk_operater

  • Added optional parameter return_error_dict

    • If the function should return a dictionary of dictionaries when failures are found. Defaults to False to error on failure.

  • Added optional parameter continue_on_errors

    • If the function should should still attempt changes for all valid inputs when failures are found. Defaults to False to not attempt any changes in the event of an error.

feature_engineering.py

UPDATED FUNCTIONS:

• create_one_hot_encoded_features
  • Added optional parameter allow_large_cardinality
    • If the ohe algorithm can generate more than 20 columns, defaults to False. Raises an error if the planned number of resulting columns is higher than 20.

enums.py

REMOVED CLASSES:

• DMVColumnBaseClass

  • Moved to private_enum module as it is not intended for customer use

• Dimension

  • Moved to private_enum module as it is not intended for customer use

• Hierarchy

  • Moved to private_enum module as it is not intended for customer use

• Measure

  • Moved to private_enum module as it is not intended for customer use

• Level

  • Moved to private_enum module as it is not intended for customer use

• DBDataType

  • Moved to private_enum module as it is not intended for customer use

• TimeLevels

  • Moved to private_enum module as it is not intended for customer use

• PlatformType

  • Moved to private_enum module as it is not intended for customer use

• RequestType

  • Moved to private_enum module as it is not intended for customer use

• ScikitLearnModelType

  • Moved to private_enum module as it is not intended for customer use

• CheckFeaturesErrMsg

  • Moved to private_enum module as it is not intended for customer use

client.py::Client

UPDATED FUNCTIONS:

• autogen_semantic_model

  • Bug Fix

    • Adjusted behavior in the case of writing 0 row tables

  • Added parameter check_permissions

    • Determines if the function should error or proceed if the AtScale warehouse connection does not have read access to the created table

linear_regression.py

UPDATED FUNCTIONS:

• linear_regression
  • Fixed bug where a subset of temporary tables could be left in the underlying database in the event of a failure