Usage: Python Library

Beyond the CLI, AIDE ML can be used as a Python library, allowing you to integrate its optimization capabilities directly into your Python scripts and workflows. The core of the library is the aide.Experiment class.

The Experiment Class

This class encapsulates a single AIDE ML run. You initialize it with your task details and then call the run() method to start the agent.

Basic Example

Here is a simple example of how to use aide.Experiment in a Python script, based on the example from the README.md.

import aide
import logging

def main():
    # Set up logging to see the agent's progress
    logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(name)s - %(levelname)s - %(message)s')
    aide_logger = logging.getLogger("aide")
    aide_logger.setLevel(logging.INFO)

    print("Starting experiment...")

    # 1. Initialize the Experiment
    exp = aide.Experiment(
        data_dir="aide/example_tasks/bitcoin_price",
        goal="Build a time series forecasting model for bitcoin close price.",
        eval="RMSLE" # Root Mean Squared Log Error
    )

    # 2. Run the agent for a specified number of steps
    best_solution = exp.run(steps=5)

    # 3. Access the results
    print("\n--- Experiment Finished ---")
    print(f"Best solution validation metric: {best_solution.valid_metric:.4f}")
    print("Best solution code:")
    print(best_solution.code)
    print("-------------------------")

if __name__ == '__main__':
    main()

Breakdown

1. import aide: Imports the necessary library.
2. aide.Experiment(...): This creates and configures a new experiment.
   • data_dir: Path to the dataset directory.
   • goal: The high-level task description for the agent.
   • eval: The evaluation metric to optimize.
3. exp.run(steps=5): This is the main call that starts the agentic tree search process. The agent will run for the number of iterations specified by the steps argument.
4. Return Value: The run() method returns a Solution object, which is a dataclass containing the best solution found during the search.
   • best_solution.code: A string containing the Python code of the best solution.
   • best_solution.valid_metric: The validation metric value achieved by the best solution.

Using AIDE ML as a library is ideal for automated workflows, programmatic experiments, or integrating its code generation capabilities into a larger application.