Kolmogorov.ai Predicate

Kolmogorov.ai Predicate (Predicate) is an application for monitoring and quantitative validation. It enables you to create monitoring projects using a vast library of metrics and generate validation reports based on multiple projects.

Configure projects for monitoring static and business metrics. Use Predicate’s extensive range of prebuilt metrics or add custom formulas for calculation of indicators.

Predicate key features

• Library of validation and monitoring projects.
• Library of metrics with visualizations in Plotly and support for adding custom metrics in Python.
• Preset package of standard metrics for model quality validation, including data drift and concept drift, as well as domain-specific metrics for credit and operational risks and customer analytics.
• Running projects according to a pre-defined procedure.
• Library of templates for standard monitoring and validation projects.
• GUI validation report builder.
• Ability to add custom preprocessing code to the execution of monitoring and validation projects (e.g., employing an ad hoc model or factor calculation).
• Traffic lights for individual metrics and evaluation of the general project status.

1. Install the components required by Predicate.

   Here is a list of required components which are not included in the default Predicate bundle:

   • Kubernetes-compatible cluster with a separate namespace. In the namespace, create a service account with CRUD permissions for all basic object types. The minimum resource configuration is 8 vCPUs, 16 GB RAM, and disk size of 100GB. Also, install the following in the cluster:
     • NFS StorageClass: RWM storage class in Kubernetes backed by an NFS server.
     • Ingress controller: Controller providing external access to Kubernetes resources.
   • Keycloak: Authentication and authorization service.
   • S3 storage (Minio or Yandex Object Storage): Storage for temporary files and metric outputs. Create a separate bucket in the storage. Assign roles for getting, updating, and deleting objects to the service account (storage.editor or storage.admin).

2. Test Predicate.

• Quick generation of reports on modeling results to share with key project participants and stakeholders.
• Automation of quantitative validation tasks.
• Dashboards for routine monitoring of models and solutions in their production versions.
• Standards and templates for monitoring procedures in various scenarios.

Data Sapience
You can contact support at contact@kolmogorov.ai.

Yandex Cloud
Yandex Cloud does not provide technical support for this product. If you have any issues, please refer to the vendor’s information resources.