August 13, 2024

Release highlights

This release brings many user experience upgrades to ValidMind, including customization for your platform UI dashboard, support for more input types in comparison tests within the developer framework, and model dependency management.

Customize your dashboard

We enhanced the ValidMind Platform UI Dashboard with new customization options:

• Rearrange and resize widgets to suit your preferences.
• Add charts from Reports to your Dashboard, for a more personalized and streamlined experience.

Customizing your dashboard

Generalize support for comparison tests

To make comparison tests easier to analyze, we’ve added support to specify an input grid when a test has many datasets or models.

You’ll need to pass a list of lists for datasets or models, since the original input is a list of inputs.

For example, the following test produces a combined GINI table:

• XGBoost on training and test dataset
• RandomForest on training and test dataset

vm.tests.run_test(
    "validmind.model_validation.statsmodels.GINITable",

    input_grid={
        "datasets": [
            [vm_train_ds, vm_test_ds],
        ],
        "model": [vm_model_xgb, vm_model_rf],
    },
)

This test produces a combined PSI report for XGBoost and RandomForest on training and test datasets:

vm.tests.run_test(
    "validmind.model_validation.sklearn.PopulationStabilityIndex",

    input_grid={
        "datasets": [
            [vm_train_ds, vm_test_ds],
        ],
        "model": [vm_model_xgb, vm_model_rf],
    },
)

Enhancements

Reports page improvements

We added new summary widgets to the Reports tab, including:

• Total Models
• Total Findings
• Open Findings
• Past Due Findings
• Closed Findings

These widgets summarize key information from your reports, making the data available at a glance:

New Findings Report summary widgets

Manage model dependencies

You can now configure model dependencies in your Model Inventory.

Manage both upstream and downstream model interdependencies:

Model dependencies diagram

Model dependency management

Bug fixes

Exclude categorical & binary features from outlier tests

We’ve excluded categorical and binary features from the outlier tests IQROutliersTable and IQROutliersBarPlot.

This fix ensures that outlier detection is applied correctly and appropriately, improving computational efficiency and leading to more relevant, accurate, and meaningful insights from the data.

Documentation

Support for GCP Private Service Connect

We’ve introduced support for Google Cloud Private Service Connect for enhanced network security and privacy. This is in addition to the existing support for AWS PrivateLink.

• Google Cloud Private Service Connect allows private connections between ValidMind and your company network. It ensures secure communication over the Google network without exposing traffic to the public internet.
• By using private endpoints within your VPC, you can reduce your network’s attack surface and maintain traffic privacy.