Examine the Azure Machine Learning schema

• 4 minutes

The azure_ml schema of the azure_ai extension allows your database to interact with the capabilities of custom-built machine-learning models. By utilizing the azure_ml schema, you can seamlessly integrate your PostgreSQL database with Azure Machine Learning services. This integration enables you to deploy and serve machine learning models directly from your database, making real-time inferencing efficient and scalable.

Real-time inference with the azure_ml schema

Azure Machine Learning is a cloud-based platform streamlining end-to-end machine learning workflows. Within Azure Machine Learning, models are developed using popular frameworks like PyTorch and TensorFlow. Once trained, these models are deployed as endpoints, stable URLs where predictions can be requested.

With the online endpoints providing real-time inference, Azure Machine Learning integrated with the azure_ai extension equips you to make accurate predictions directly from your database. The inference function within this schema is designed to facilitate making predictions or generating outputs using a trained model from Azure Machine Learning. When you deploy a model, the inference function allows you to invoke the model and obtain predictions on new data.

• azure_ml.inference(jsonb,integer,boolean,text)

The inference() function expects the following input parameters:

Parameter	Type	Default	Description
input_data	jsonb		A JSON object containing the input_data object required for interacting with Azure Machine Learning models.
deployment_name	text	NULL::text	(Optional) Name of the model deployment to target at the specified Azure Machine Learning endpoint.
timeout_ms	integer	NULL::integer	The timeout...
throw_on_error	boolean	true	desc...
max_attempts	integer	1	Number of times to retry the call to Azure OpenAI service in the event of a failure.
retry_delay_ms	integer	1000	Amount of time, in milliseconds, to wait before attempting to retry calling the Azure OpenAI service endpoint.

Azure Machine Learning inferencing endpoints expect a JavaScript Object Notation (JSON) object as input. However, the structure of this object is dependent upon the underlying model. For example, a regression model trained to predict daily rental prices for short-term housing rentals in the Seattle, Washington area, given specific inputs like the neighborhood, ZIP Code, number of bedrooms, number of bathrooms, and more, has the following shape:

{
  "input_data": {
    "columns": [
      "host_is_superhost",
      "host_has_profile_pic",
      "host_identity_verified",
      "neighbourhood_group_cleansed",
      "zipcode",
      "property_type",
      "room_type",
      "accommodates",
      "bathrooms",
      "bedrooms",
      "beds"
    ],
    "index": [0],
    "data": [["False", "False", "False", "Central Area", "98122", "House", "Entire home/apt", 4, 1.5, 3, 3]]
  }
}

The expected input object structure can be retrieved by examining the Swagger definition associated with your deployed endpoint. This definition specifies the ServiceInput and ServiceOutput structures, which you can use to determine your inputs and outputs.

Configure a connection to Azure Machine Learning

Before using the azure_ml.inference() function to perform real-time inference, you must configure the extension with your Azure Machine Learning scoring endpoint and key. The value for azure_ml.scoring_endpoint is the REST endpoint for your deployed model. The value for azure_ml.endpoint_key can be either the primary or the secondary key for this endpoint.

SELECT azure_ai.set_setting('azure_ml.scoring_endpoint', '{endpoint}');
SELECT azure_ai.set_setting('azure_ml.endpoint_key', '{api-key}');