Domino makes it easy to deploy models for batch scoring to make inferences (predictions) in bulk over a large batch of records. Learn how to use Domino jobs to perform batch prediction with the web UI and API.
Batch scoring jobs are a good choice for "fire-and-forget" scoring jobs that don’t require a strict SLA (Service Level Agreements) for results. If you need a scoring endpoint with better response times, especially for customer-facing applications, use the Model API REST endpoints.
-
Set up the executable inference file.
-
Create a prediction script that is the entry point for the batch scoring program.
-
Use the prediction script as the executable when you configure a new job run. See details for how to start a Domino job here.
-
-
Configure execution and compute resources.
-
Select a custom Domino environment if you require additional packages not supported by the default environment.
-
Select the size of the storage volume in the job.
-
Select the appropriate compute resources needed. For simplification, Domino offers a list of available Hardware Tiers that represents a pool of nodes your Domino Administrator configures for such executions.
-
If your scoring code works with large-scale compute clusters like Spark, Ray, or Dask, attach one of those compute clusters to the job to enable their use. Domino provides on-demand Spark, Ray, Dask, and OpenMPI clusters made available to the executable in the job.
-
Run the job created in the previous section. Optionally, run the job on a schedule to process periodically generated data that requires batch scoring.
Batch job durations depend on the volume of processed data and the available compute resources. Monitor the job’s progress on the jobs dashboard.
Configure the notifications settings to receive the results of an execution. By default, you are notified of all failed runs via email. On successful execution, you receive an email with the last few lines of stdout and up to ten results files from your execution. Domino detects which files were added or changed during your execution and captures those as the execution’s results. You can also set up custom emails as part of the notification.
The results of an execution are the set of files that your scoring code generates or modifies when it runs. View the results of a given execution through the Runs tab of your Project or on the Results tab of your Job. See Execution results to learn how to retrieve results, customize the results dashboard, or automate the delivery of results via an Airflow pipeline.
To troubleshoot your scoring job, Domino provides access to log files related to the execution. If you need help, contact Domino Support to review and resolve execution issues.
-
For errors that occur before execution, i.e., during setup, download the Setup Output files.
-
Download the User Output log file for errors in the scoring job to examine and address any changes to your scoring code.
-
For runtime errors related to the deployment itself, download the Deployment Logs file to analyze the failure and share it with Domino Support if needed.
For model deployments that require interactivity, see Deploy models at REST.