Skip to content

On-premise users: click in-app to access the full platform documentation for your version of DataRobot.

Deployment workflows

DataRobot's MLOps monitoring is available for any models deployed in DataRobot prediction environments (including models on your own infrastructure using a Portable Prediction Server). With DataRobot MLOps, you can deploy models written in any open-source language or library and expose a production-quality REST API to support real-time or batch predictions. Custom inference models allow you to bring pre-trained models into DataRobot to make monitored predictions alongside DataRobot's models. In addition, you can configure monitoring for models running in external prediction environments with the MLOps agent. The workflows below provide high-level overviews of the most common deployment scenarios, including links to the relevant documentation for each step.

Workflow types

With the workflows provided for the common model and environment combinations below, you can learn to deploy DataRobot AutoML models and custom inference models to DataRobot prediction environments, either within DataRobot or containerized for external deployment. In addition, with the monitoring agent, you can monitor models deployed in completely external prediction environments:

Model Type Environment Type Workflow
DataRobot model DataRobot prediction environment How to deploy a DataRobot model in a DataRobot prediction environment.
DataRobot model Portable Prediction Server How to deploy a DataRobot model in a Portable Prediction Server (PPS).
Custom model DataRobot prediction environment How to deploy a custom model in a DataRobot prediction environment.
Custom model Portable Prediction Server How to deploy a custom model in a Portable Prediction Server (PPS).
Scoring Code External prediction environment How to deploy exported DataRobot model Scoring Code in an external environment with monitoring agent enabled.
External model External prediction environment How to deploy an external model in an external prediction environment with monitoring agent enabled.

Model types

The model types referenced in the deployment workflows are defined below:

Model type Description
DataRobot model A standard DataRobot model.
Custom model An external (Python, Java, or R) model assembled in the Custom Model Workshop.
Scoring Code A method for downloading select DataRobot models from the leaderboard for external deployments. Models downloaded this way are packaged as a Java Archive (JAR) file containing Java prediction calculation logic identical to the DataRobot API's calculation logic. However, Scoring Code predictions are made using a command-line interface (CLI) instead of API calls, allowing you to make low-latency predictions.
External (remote) model A model completely external to DataRobot, making predictions on local infrastructure or any other external environment. These models can be monitored by the MLOps agent, and deployment information can be reported to DataRobot MLOps.

Prediction environment types

The prediction environments referenced in the deployment workflows are defined below:

Prediction environment type Description Evaluation
DataRobot prediction environment The default DataRobot prediction environment on DataRobot infrastructure. Provides the most straightforward deployment, prediction, monitoring, and model replacement processes. However, predictions are subject to network performance limitations.
Portable Prediction Server A containerized (with all resources required to run on any infrastructure) DataRobot prediction environment for DataRobot models to make predictions on your infrastructure with MLOps monitoring. You can make API-based predictions on local infrastructure to improve performance for low-latency predictions. However, the deployment, prediction, monitoring, and model replacement processes are more complex.
Custom model Portable Prediction Server A containerized (with all resources required to run on any infrastructure) DataRobot prediction environment for custom models to make predictions on your infrastructure with MLOps monitoring. The custom model PPS bundle contains a deployed custom model, a custom environment, and the monitoring agent. You can make API-based predictions on local infrastructure to improve performance for low-latency predictions. However, the deployment, prediction, monitoring, and model replacement processes are more complex.
External prediction environment A prediction environment completely external to DataRobot and used to make predictions monitored by the monitoring agent and reported to DataRobot MLOps. External predictions or Scoring Code predictions can be made on local infrastructure to improve performance for low-latency predictions. However, the deployment, prediction, monitoring, and model replacement processes are more complex.

Updated August 31, 2023