Kolmogorov.ai Predicate

Kolmogorov.ai Predicate — an application for performing monitoring or quantitative validation, which enables the organization of a monitoring project based on a vast library of metrics and allows the creation of a validation report from multiple projects.

Configure monitoring projects from statistical and business metrics. The Predicate Metrics Catalog contains a wide array of built-in metrics, as well as allows for the addition of custom calculation algorithms for indicators.

Predicate provides the following functionality:

• A catalog of validation and monitoring projects.
• A catalog of metrics with Plotly visualizations, addition of custom Python metrics.
• A pre-configured package of standard model quality assessment metrics (including data / concept drift), as well as metrics specific to subject areas like credit risk, operational risk, and customer analytics.
• Scheduled project execution according to a set regimen.
• A catalog of template projects for standard monitoring and validation.
• A GUI for constructing validation reports.
• The ability to add custom preprocessing code to the execution of monitoring/validation projects (for example, ad-hoc model application, factor calculation, etc.).

1. Prerequisites1. Prerequisites

Components that are not included in the default delivery of the Predicate application but are required for its operation:

• A Kubernetes-compatible cluster with the ability to create a separate namespace. In this namespace — the ability to create a service account with CRUD rights for all basic types of objects. Minimum resource quota — 8CPU, 16GB RAM, 100GB disk space. Additionally, the following components are required in the cluster:
  • NFS StorageClass — RWM storage class in Kubernetes that uses an NFS server.
  • Ingress Controller — a controller that provides access to Kubernetes services from outside.
• Keycloak — a service for authorization and authentication using Keycloak.
• S3 storage (Minio or Yandex Object Storage) — storage where temporary files and metric results are saved. In this storage, there should be the ability to create a separate bucket. In this bucket — a service account with the role of storage.editor or storage.admin (GET/UPDATE/DELETE objects).#### 2. Installation and Configuration of Components

2. Installation and configuration of components and cluster2. Installation and configuration of components and cluster

3. Instruction for checking Predicate’s operability.3. Instruction for checking Predicate’s operability.

• Rapid generation of modeling results reports for communication with key project participants and stakeholders.
• Automation of quantitative validation tasks.
• Operational dashboards for scheduled monitoring of models and solutions in an industrial format.
• Standardization and templating of monitoring regulations for diverse scenarios.
• Setting up notifications and triggers for automatic calibration and retraining based on a flexible traffic light system.

You can reach out to technical support via email at contact@kolmogorov.ai