Skip to content

Diabetes Classification Example

This example demonstrates optimizing a Scikit-Learn model.

Quick Start Guide

You can verify your installation immediately by running our pre-configured Scikit-Learn example without any setup.

  1. Open Control Panel: In the UI, stay on the Control tab.
  2. Set Root Directory: Set the Root Directory field to example/diabetes_sklearn.
  3. Configure eval_cmd:
    • In the Global Variables list, find eval_cmd.
    • Ensure it is set to: python evaluator.py (ensure your environment's python is used).
  4. Save & Start: Click 💾 Save Config, then click ▶ Start Agent.
    • The agent will immediately start its first cycle, analyzing the diabetes data and generating model hypotheses.

Location

example/diabetes_sklearn

Reference Experiment Structure

The example/diabetes_sklearn/ directory provides a reference implementation for organizing machine learning experiments. It is designed to be robust and prevent "cheating" by the LLM.

File Organization

  • Evaluation Script (evaluator.py): The ground truth for assessment. Includes built-in anti-leakage checks and enforces time limits.
  • Seed Experiment Directory: Contains the initial strategy.py and auxiliary files.
  • Experiment History (history.json): Automatically stores metrics and optimal results.

Evolutionary Progress Logic (B, L, S)

The default workflow uses a Branch (B), Level (L), Step (S) logic: * Branch (B): A new conceptual direction. * Level (L): Progression in depth (advanced on significant improvement). * Step (S): Explorations within the same level (retries/mutations).

Structure

  • strategy.py: Defines the Model and Hyperparameter Space.
  • evaluator.py: Runs Grid Search and reports MSE.