Skip to main content

Run an Inference on an ML Model

Run Enrichment Notebook that will create/update a Lakehouse Table.

Category: Machine Learning | Tags: ML

How it works

Run Inference on ML Model '<<ModelSchemaName>>.<<ModelName>>' and save results to Delta Table '<<DeltaSchemaName>>.<<DeltaTableName>>' (using Update Type '<<DeltaTableUpdateType>>')

To use this activity within the API, use an ActivityCode of ML-RUN-INFERENCE.

Example JSON​

An example of what the Task Config would look like for a task using this activity. Some of these variables would be set at the group level to avoid duplication between tasks.

{
  "NotebookPath": "/Users/fred.nurks@example.com/MyRepo/My Notebook",
  "ModelSchemaName": "example_schema",
  "ModelName": "",
  "DeltaSchemaName": "example_schema",
  "DeltaTableName": "my_table",
  "DeltaTableUpdateType": "Replace",
  "NotebookParameters": { "Param1": "Value1", "Param2": "Value2" }
}

Variable Reference​

The following variables are supported:

  • AdditionalNotebooks (Optional) - The path to other notebooks, Python files etc., referenced by the main notebook.

  • DatabricksClusterId (Optional) - The Databricks Cluster to use for this task.

  • DeltaSchemaName (Required) - The name of the Schema this transformation lives in.

  • DeltaTableBusinessKeyColumnList (Optional) - Comma-separated list of Business Key columns in the Lakehouse Table. This is required if 'Lakehouse Table Update Type' is 'Dimension' or 'Merge'. If a value is specified, a uniqueness test is performed against this (composite) key for both the result of the Enrichment and the Lakehouse Table.

  • DeltaTableComments (Optional) - Comments to add to the Lakehouse Table.

  • DeltaTableName (Required) - The name of the Table representing this transformation.

  • DeltaTablePartitionColumnList (Optional) - Comma-separated ordered list of columns forming the Partitioning strategy of the Lakehouse Table.

  • DeltaTableUpdateType (Required) - Indicates what type of update (if any) is to be performed on the Lakehouse Table.

  • ExtractControlVariableName (Optional) - For incremental loads only, the name to assign the Extract Control variable in State Config for the ExtractControl value derived from the Extract Control Query above.

  • ExtractControlVariableSeedValue (Optional) - The initial value to set for the Extract Control variable in State Config - this will have no impact beyond the original seeding of the Extract Control variable in State Config.

  • IsFederated (Optional) - Makes task available to other Insight Factories within this organisation.

  • Links (Optional) - NULL

  • MaximumNumberOfAttemptsAllowed (Optional) - The total number of times the running of this Task can be attempted.

  • MinutesToWaitBeforeNextAttempt (Optional) - If a Task run fails, the number of minutes to wait before re-attempting the Task.

  • ModelAlias (Optional) - The Alias of the ML Model e.g. Champion/Challenger. If Model Version is specified, this will be used instead of Model Alias.

  • ModelName (Required) - Name of the ML Model.

  • ModelSchemaName (Required) - The Schema the ML Model resides in.

  • ModelVersion (Optional) - The Version of the ML Model. If this is specified, it will be used instead of Model Alias.

  • NotebookParameters (Optional) - Parameters for use in the Databricks Notebook. This is JSON format e.g. { "Param1": "Value1", "Param2": "Value2" }.

  • NotebookPath (Required) - The relative path to the Databricks Notebook.

  • PartitionDepthToReplace (Optional) - The number of columns in 'Lakehouse Table Partition Column List' (counting from the first column in order) to use in a Partition Replacement. NOTE: This cannot be greater than the number of columns defined in the 'Lakehouse Table Partition Column List'. Defaults to 1 if only one column has been specified in 'Lakehouse Table Partition Column List'.

  • SkipCreateVolumeAndSchema (Optional) - If a Schema and/or Volume has already been created, you can opt to skip this check - it will lead to better performance.