Prediction services

Prediction Service allows you to send real-time data to a model and receive predictions immediately. You can create an endpoint of the model that is exposed as an API and upload the test dataset. This API that has the model will make predictions on the uploaded data.

Creating and Testing a Prediction Service

Use this procedure to create a prediction service or an endpoint for a model. This service can only be created for models generated after running the data pipeline in a project or for models that are manually added. After generating the model as an API, you can test it on the uploaded dataset to make predictions.

Steps to Create a Prediction Service

1. Select Artifacts & Models from the left navigation menu. The Artifacts tab is displayed, showing a list of all artifacts in the tenant.

2. Click the Models tab to see the list of models created in this tenant.

3. Click +Add in the Prediction Service column for the model whose API you want to create. The Prediction Service tab is displayed.

4. In the Details section, specify the following:

   • Name: The name of the prediction service.

   • Description: A brief description of the prediction service.

   • Environment: The environment in which you want to test the prediction service.

   • Pre-process & Post-process: If needed, add pre-processing and post-processing steps using the integrated code editor.

5. Specify the Configuration Options:

   • Timeout: The duration (in minutes) after which the incoming request should time out.

   • Concurrency: The number of parallel requests you can send (ranges from 5 to 100).

6. Enable the Save History option to view a detailed record of activities in the prediction service. Disable this option to stop tracking logs.

7. Click Save to create the endpoint for the model.

This generates a unique endpoint for the model along with a CURL command.

Testing the Prediction Service

1. Select the file format. Possible values:

   • JSON

   • CSV/XLSX/XLS File

   • Canvas Datasets

2. Click Browse to upload the file in CSV or JSON format based on the selected file format.

3. Click Test to check the prediction results.

1. Click History to view a comprehensive list of all executions of the prediction service. For more details, see Viewing the Prediction Service History.

After testing the prediction service on an uploaded dataset, you will see two options:

• Download as CSV: Download the output as a CSV file.

• Add to Canvas: Add the dataset to the canvas for further analysis.

Note: You can review logs from the past 30 days by clicking the Logs option. Logs provide detailed information, including:

• Types of queries executed

• Number of successful queries

• Number of failed queries

Logs are accessible only if the Save History toggle is enabled. Additionally, you can export logs using:

• Export: Download logs as a .txt file via the export option in the side panel.

• Open in New Tab: View logs in a separate tab for better visibility and analysis.

Viewing the Prediction Service History

Use this procedure to view the list of times the prediction service was executed in the past 30 days.

Steps to View Prediction Service History

1. Select Artifacts & Models from the left navigation menu. The Artifacts tab is displayed, showing a list of all artifacts in the tenant.

2. Click the Models tab to see the list of models created in this tenant.

3. Search for the model whose prediction service history you want to view.

4. Click the link under the Prediction Service column. This navigates to the Prediction Service page for the selected model.

5. Click History at the top of the Test Prediction Service section. This displays a record of all times the prediction service was executed in the last 30 days.

1. Review the Prediction Service History information:

• Start Time: The timestamp indicating when the prediction service execution began.

• End Time: The timestamp marking when the prediction service execution was completed.

• User: The user who initiated the prediction service run.

• Status: The current status of the prediction service run (e.g., success, failure).

• Info: Details of the request and response generated after the prediction service execution.

• Tracking ID: A unique tracking ID generated for every run.

• Request ID: A unique request ID generated for each run. To view this in the table, enable the Request ID column from the table settings.

Filtering Prediction Service Records

Use the Filter option to refine records based on their status:

• Success: Displays only successful executions.

• Failure: Displays only failed executions.

• Both: Displays all executions.