Skip to main content

2.1.0 release notes

This release of AI-Link is meant to improve the core function of the product to make it easier to use across AtScale-supported databases.

In addition to new functional and non-functional aspects of AI-Link being introduced with this release, AI-Link will be renaming existing functionality:

  • WriteBack (joining a dataframe - e.g. ML predictions - to an existing AtScale data model) is now known as /using-ai-link/ml-writeback-to-bi/semantic-predictions/index.
  • WriteTo (programmatically creating a new AtScale data model) is now known as /using-ai-link/auto-semantic-model-creation/index.

Please refer to our API documentation for the latest syntax to use with AI-Link.

For details on the updates that are part of this release see the sections below.

Orchestrator-Based Design​

Functions that have historically lived in feature_utils and metadata_utils should now be accessed through orchestrator methods in the relevant atscale object.

For example, the get_feature_description that used to live in metadata_utils can now be called directly off an instance of the DataModel object like so:

sample_data_model.get_feature_description(<feature_name>)

This removes the need to pass DataModel, Project, and Connection objects to various feature calls, and makes available operations on each object accessible via autocomplete. A full list of the features now accessible in the parent objects can be found in the API documentation.

Feature Impact (Beta)​

Machine Learning (ML) model teams can ascertain feature importance from ML models they build through python to whatever ML tool/framework they are building with. This allows the model builder, evaluator, and consumer to understand the relationship between the ML features used in the model algorithm to the prediction output itself. Feature Impact score tables in this Beta will be added to the same data model that contains the predictions table but will not be connected to any existing tables.

Database Write Function Updates​

Users will now be able to leverage Semantic Predictions to write dataframe objects, to Databricks, GBQ, Iris data warehouses.

Testing Updates​

AI-Link test suite has been expanded to accommodate working with more complex customer data models.

Lab and User Experience Updates​

Introducing new documentation to improve first time interaction, onboarding, and continued usage of AI-Link with the AtScale hosted lab environment and customer production environments.