Skip to content

Core Concepts

VerifIA’s verification pipeline is built around five key components. Each concept page dives into configuration and recommendations—together they form the end‑to‑end workflow for domain‑aware model verification.

Concept Description
Data How to provide and preprocess your original seed dataset, handle metadata, exclude out‑of‑domain or mispredicted examples, and balance seed diversity vs. runtime.
Model Guidelines for preparing a model that’s already optimized, validated, and tested—wrapping it in the appropriate VerifIA API and understanding the role of domain‑aware checks.
Domain Defining your application’s a priori knowledge—variables, feasibility constraints , and behavioral rules—in a compute‑friendly format.
Searcher Configuring the population‑based, evolutionary search algorithms that probe each seed×rule subspace to surface behavioral inconsistencies.
Run Orchestrating individual verification runs, sampling inputs, collecting per‑run metrics, and aggregating results into comprehensive reports.

Workflow Overview

  1. Prepare & attach Data
    Provide VerifIA with raw tabular data (files, DataFrame, or Dataset), validate metadata, and exclude seeds that are already mispredicted, or out of domain.

  2. Wrap & attach Model
    Use a built‑in Model wrapper or a pre-filled model‑card to load your optimized, tested model (scikit‑learn, CatBoost, XGBoost, TensorFlow, etc.) from a local disk or a model registry.

  3. Define your Domain
    Encode variables, input constraints, and directional rules in YAML (or Python dict) so VerifIA knows how to derive and evaluate new test inputs.

  4. Select a Searcher
    Choose and configure an evolutionary algorithm (e.g. GA, PSO, FFA) that will explore each constrained subspace for rule violations.

  5. Run Verification
    Launch a Run with your budget (population size, iterations, optional seed sampling). Inspect per‑run metrics (consistency rate, violation counts, infeasible points) and aggregate across runs for final reporting.

Next Steps:
- Dive into each concept page above to configure your pipeline.
- Try a quick experiment with a tutorial or a use case.
- Follow the Domain Creation Guide to iterate on your domain creation.