Using Machine Learning (ML) with Amazon Athena (Preview)

Machine Learning (ML) with Amazon Athena (Preview) lets you use Athena to write SQL statements that run Machine Learning (ML) inference using Amazon SageMaker. This feature simplifies access to ML models for data analysis, eliminating the need to use complex programming methods to run inference.

To use ML with Athena (Preview), you define an ML with Athena (Preview) function with the USING FUNCTION clause. The function points to the SageMaker model endpoint that you want to use and specifies the variable names and data types to pass to the model. Subsequent clauses in the query reference the function to pass values to the model. The model runs inference based on the values that the query passes and then returns inference results. For more information about SageMaker and how SageMaker endpoints work, see the Amazon SageMaker Developer Guide.

Considerations and Limitations

• Available Regions – The Athena ML feature is available in preview in the US East (N. Virginia), Asia Pacific (Mumbai), Europe (Ireland), and US West (Oregon) Regions.

• AmazonAthenaPreviewFunctionality workgroup – To use this feature in preview, you must create an Athena workgroup named AmazonAthenaPreviewFunctionality and join that workgroup. For more information, see Managing Workgroups.

• SageMaker model endpoint must accept and return text/csv – For more information about data formats, see Common Data Formats for Inference in the Amazon SageMaker Developer Guide.

• SageMaker endpoint scaling – Make sure that the referenced SageMaker model endpoint is sufficiently scaled up for Athena calls to the endpoint. For more information, see Automaticaly Scale SageMaker Models in the Amazon SageMaker Developer Guide and CreateEndpointConfig in the Amazon SageMaker API Reference.

• IAM permissions – To run a query that specifies an ML with Athena (Preview) function, the IAM principal running the query must be allowed to perform the sagemaker:InvokeEndpoint action for the referenced SageMaker model endpoint. For more information, see Allowing Access for ML with Athena (Preview).

• ML with Athena (Preview) functions cannot be used in GROUP BY clauses directly

ML with Athena (Preview) Syntax

The USING FUNCTION clause specifies an ML with Athena (Preview) function or multiple functions that can be referenced by a subsequent SELECT statement in the query. You define the function name, variable names, and data types for the variables and return values.

Synopsis

The following example illustrates a USING FUNCTION clause that specifies ML with Athena (Preview) function.

USING FUNCTION ML_function_name(variable1 data_type[, variable2 data_type][,...]) RETURNS data_type TYPE SAGEMAKER_INVOKE_ENDPOINT WITH (sagemaker_endpoint= 'my_sagemaker_endpoint')[, FUNCTION...][, ...] SELECT  [...] ML_function_name(expression) [...]

Parameters

USING FUNCTION ML_function_name(variable1 data_type[, variable2 data_type][,...])

ML_function_name defines the function name, which can be used in subsequent query clauses. Each variable data_type specifies a named variable with its corresponding data type, which the SageMaker model can accept as input. Specify data_type as one of the supported Athena data types that the SageMaker model can accept as input.

RETURNS data_type TYPE

data_type specifies the SQL data type that ML_function_name returns to the query as output from the SageMaker model.

SAGEMAKER_INVOKE_ENDPOINT WITH (sagemaker_endpoint= 'my_sagemaker_endpoint')

my_sagemaker_endpoint specifies the endpoint of the SageMaker model.

SELECT [...] ML_function_name(expression) [...]

The SELECT query that passes values to function variables and the SageMaker model to return a result. ML_function_name specifies the function defined earlier in the query, followed by an expression that is evaluated to pass values. Values that are passed and returned must match the corresponding data types specified for the function in the USING FUNCTION clause.

Examples

The following example demonstrates a query using ML with Athena (Preview).

USING FUNCTION predict_customer_registration(age INTEGER) 
    RETURNS DOUBLE TYPE 
    SAGEMAKER_INVOKE_ENDPOINT WITH (sagemaker_endpoint = 'xgboost-2019-09-20-04-49-29-303') 
SELECT predict_customer_registration(age) AS probability_of_enrolling, customer_id 
     FROM "sampledb"."ml_test_dataset" 
     WHERE predict_customer_registration(age) < 0.5;