Python API Reference

This page provides a detailed reference for using AIDE ML as a Python library. The main entry point is the aide.Experiment class.

aide.Experiment

The Experiment class is used to configure and run a complete agentic search process programmatically.

class aide.Experiment:
    def __init__(self, data_dir: str, goal: str, eval: str | None = None):
        # ...

    def run(self, steps: int) -> Solution:
        # ...

__init__(self, data_dir, goal, eval)

Initializes a new experiment run.

Parameters:

• data_dir (str): The path to the directory containing the dataset files. This can be a relative or absolute path. The agent will create a sandboxed workspace based on this data.
• goal (str): A high-level, natural language description of the task's objective.
• eval (str | None, optional): A more specific description of the evaluation metric the agent should aim to optimize. If None, the agent will infer a suitable metric from the goal.

Example:

import aide

exp = aide.Experiment(
    data_dir="./aide/example_tasks/house_prices",
    goal="Predict the sales price for each house",
    eval="Use RMSE between log-prices"
)

run(self, steps: int) -> Solution

Starts the agentic tree search process for the configured experiment.

Parameters:

• steps (int): The total number of iterative steps (improvements or debugs) the agent should perform.

Returns:

• Solution: A dataclass object containing the best solution found during the run.

Example:

# Continuing from the previous example
best_solution = exp.run(steps=10)

print(f"Final Metric: {best_solution.valid_metric}")
print(f"Final Code:\n{best_solution.code}")

aide.Solution

A simple dataclass that holds the final result returned by the Experiment.run() method.

from dataclasses import dataclass

@dataclass
class Solution:
    code: str
    valid_metric: float

Attributes:

• code (str): The Python code of the best-performing script.
• valid_metric (float): The validation score achieved by the best script.