Get started

This page describes how to access the MLOps app.

Start an instance

1. In the H2O.ai App Store, enter MLOps in the search bar. Select the H2O MLOps tile to open a page that contains basic details about the app and a list of existing instances of the app.

   

2. To start an instance of the app, click the Run button.

3. To open the app in a new browser tab, click the Visit button.

Pause or end an instance

To pause or end a running instance, click My Instances.

• To pause an instance, locate the instance you want to pause and click the Pause button.

• To end an instance, locate the instance you want to end and click the drop-down button next to the Pause button, then click Terminate.

Note

Instances can also be paused or ended on the app details page.

View instance logs

To view a log of processes related to an instance of MLOps, click My Instances, then click Instance log.

Workflow

The following is a typical workflow for using H2O MLOps with Driverless AI:

1. Create a new project in MLOps. The new project appears on the Driverless AI Projects page.

2. In Driverless AI, open your project and create a new experiment. You can also link any existing datasets and experiments to the project. For more information, see the page on the Projects workspace in the Driverless AI documentation.

3. After the experiment is completed, link that experiment to your project. For more information, see the page on linking experiments in the Driverless AI documentation.

4. In MLOps, register the experiment as a model, then deploy the model to Dev or Production.

Next steps

To learn more about using MLOps, refer to the following pages:

• Projects

  • Understanding projects in MLOps

  • Creating projects

  • Viewing projects

• Experiments

  • Understanding experiments in MLOps

  • Adding experiments

• Model registry

• Model deployment

  • Understanding deployments in MLOps

  • Deploying a model

  • Configuring node affinity and toleration

  • Shapley values support

• Key terms

• MLOps Python client:

  • Installing the Python client

  • Using the Python client

  • Python client examples:

    • Driverless AI example

    • H2O-3 example

    • MLflow examples:

      • MLflow PyTorch example

      • MLflow scikit-learn example

      • MLflow custom Python model example

    • A/B Test deployment

    • Champion/Challenger deployment

    • Model monitoring:

      • List monitored deployments

      • Get monitored deployment

      • Get health metrics

      • Get drift metrics

    • Deploying a model with Kubernetes resources and replicas

    • Getting metadata for experiments

    • Getting deployment statuses for experiments

    • Creating secure endpoints

    • Sharing projects

    • Pump data

    • Authentication methods in Keycloak

Feedback

• Send feedback on the MLOps app to cloud-feedback@h2o.ai