Altair SmartWorks Analytics

 

Model Metadata

A new feature of SmartWorks MLOps is the ability to add metadata to ML models when they are be registered. Collectively, we refer to these data as Model Metadata.

There are a few different kinds of Model Metadata that can added:

  • Prediction Type – the type of prediction the model will be making. Current options are:

    • CLASSIFICATION_BINARY – a classification model with one output label and two possible classes for that label

    • CLASSIFICATION_MULTICLASS – a classification model with one output label and more than two possible classes for that label

    • REGRESSION – a regression model

    • CLUSTERING – a clustering model

    • CUSTOM – any other kind of custom-type model

  • Dependent Variables – name(s) of the output(s) that your model is predicting

  • Independent Variables – name(s) of the input feature(s) that your model will process

  • Custom Tags – any custom tags that you wish to add to your model. They can be descriptions of the model, project labels, team labels, or any other key-value pair

Adding Model Data in the MLOps Node

To add Model Metadata in the MLOps Node, we start by following the exact same process as before to register a model: upload it to the library, start a new model registration, and fill out all the information such as model name, type, etc.

At the bottom of the page there is a new section called “Metadata“. Notice how already you can select your model’s Prediction Type from the dropdown.

From there, you will have the option to add the rest of the Model Metadata in one of two ways: Table or JSON.

Table

Steps:

  1. Click on the “Add New Variable“ button.

  2. From there, you can set:

    • The Name of your new variable

    • The Type of your new variable:

      • Independent Variable

      • Dependent Variable

      • Tag (for custom tags)

    • The DataType of the variable

You can add as much of these variables as you wish.

Any data you add is automatically synced with the JSON view.

 

JSON

You can add Model Metadata using a JSON. There is a specific schema to follow, for which an example is shown below. You will also see a similar example in the platform UI.

 

 

Adding Model Metadata in the MLflow UI

When you train and produce a model using SmartWork's AutoML node, your model will automatically be saved with the following Model Metadata:

  • Prediction Type

  • Independent Variables

  • Dependent Variables

If you wish to add more metadata such as Custom Tags, you can simply edit the modelMetadata tag in the MLflow UI as shown below. The only requirement is that you follow the proper JSON schema.

 

Viewing Model Metadata

Steps:

  1. Select your desired model from the Model Registry by double-clicking it.

  2. Select your desired model version by double-clicking it.

  3. This will bring you to a page with detailed information about your Model Metadata. The variables which you added before are viewable in both table and JSON format.