Note

This documentation is for a prior release of Kinetica. For the latest documentation, click here.

Models + Analytics

A model is a mathematical or programmatical representation of a real-world process.

  • Point to + Add Model then click one of the options to begin the model setup/import process.
  • Type into Filter to filter down the models
  • Click show to display any archived models.
  • Click refresh to refresh the table
  • Click export to export the table's values as JSON or CSV
  • Click an existing model to display additional actions.
    • BLACKBOX models:
      • Click View Selection to open the Model Details page
      • Click Deploy to deploy the model
      • Click Clone Model to clone the model
      • Click Export Entity to export the model as a JSON object
      • Click Archive to archive the model; it will be hidden from the list of models
      • Click Description / Configuration to review summary information for the model
../img/aaw_ui_models.png

Details

The Model Details page provides a detailed look at a given model, including configuration information, feature set, training and test datasets, and any deployments. Available actions depend on the type of model being detailed.

  • BLACKBOX models:
    • Click Back to return to the Models page
    • Click Deploy to deploy the model
    • Click Clone Model to clone the model
    • Click Export Entity to export the model as a JSON object
    • Click Archive to archive the model; it will be hidden from the list of models
    • If the model has been deployed previously, click the deployment's name in the Model Deployments table to open the Deployment Details
../img/aaw_ui_model_details.png

Model Creation

New Blackbox Model

After opting to create a new Blackbox model, two paths are available:

  • Inspect Docker Container -- inspect an existing Docker container (one that is already in a Docker repository), select an available module/function, and automatically import the model

    1. From the Models and Analytics page, click + Add Model ‣ New Blackbox
    2. Under Inspect Docker Container, optionally search for and select a Docker Registry credential
    3. Provide a Docker container URI (e.g., <repo-name>/<image-name>:<tag>)
    4. Click Inspect. AAW will pull the container's functions
    5. Optionally, click View to display details about a given function
    6. Under the desired function, click Import. AAW will automatically configure the Blackbox model
    7. Optionally, adjust the Blackbox model configuration fields as necessary
    8. Click Create

  • Create a Blackbox Model Manually -- start from a blank template and provide various parameters to manually create and import a Blackbox model

    Note

    Models can be created manually from existing containers as well; this option provides more manual control over the model import process

    1. From the Models and Analytics page, click + Add Model ‣ New Blackbox.
    2. Under Create a Blackbox Model Manually, click Create.
    3. Provide a Model Name.
    4. Optionally, provide a Model Description.
    5. Provide a Blackbox Docker Container repository URI or opt to create a Blackbox model container.
      • If a Blackbox model has already been created using the Kinetica Blackbox SDK and published to a Docker repository, provide the URI in the following format: <repo-name>/<image-name>:<tag>, e.g., kinetica/kinetica-blackbox-quickstart:latest
      • If opting to create a Blackbox model container automatically using AAW, click + New Container.
        1. Provide the Docker Repository Name.
        2. Optionally, provide the Docker Repository Description.
        3. Optionally, upload a Requirements File.
        4. Upload the Module File (must be a Python file).
        5. Provide the Module Function name from the Module File.
        6. Select a Docker credential.
        7. Click Create.
    6. Provide a Module and Function from the model
    7. For Input Columns:
      1. Click Add Input Column to create input columns.
      2. Provide a Column name and Type.
    8. For Output Columns:
      1. Click Add Output Column to create output columns.
      2. Provide a Column name and Type.
    9. Click Create.

Model Deployment

After creating a model, a deployment can be created from the Models + Analytics page or the Model Details page.

  1. Select a model from the Models + Analytics page or open the Model Details page.

  2. Click Deploy in the right-hand menu.

  3. Provide a Name for the deployment.

  4. Optionally, provide a Description for the deployment.

  5. Optionally, select a Featureset.

  6. Select a compute target:

    • If selecting CPU, set the number of replicas of the model to deploy.

    • If selected GPU, set the number of GPUs used per replica and the number of replicas of the model to deploy. The Resources graphic will update accordingly.

      ../img/aaw_ui_gpu_compute_target.png

  7. Select if the deployment should be available for audits (Persist Audit):

    • True, the deployment results will be available in any audits
    • False, the deployment results will not be available in any audits

    Tip

    Deployments with Persist Audit set to False should achieve better performance

  8. Select a Mode. See Deployments for more information on the different types of modes.

    • If selecting On-Demand, skip to the next step.
    • If selecting Continuous or Batch, select a Source Table from the drop-down menu and provide an Output Table name.

  9. If Continuous mode is selected, select if the deployment should leverage Multihead I/O for better inference output performance:

    • True, the deployment will leverage Multihead I/O
    • False, the deployment will not leverage Multihead I/O

  10. Optionally, provide key and value pairs for any Environment Variables available to the model

  11. Click Deploy. The model must then be started via the Deployments interface.