Usage: Command-Line Interface (CLI)

The primary way to interact with AIDE ML is through its command-line interface, aide. The CLI allows you to define a task, specify the data, and configure the agent's behavior.

Basic Usage

The fundamental command structure requires three main arguments: data_dir, goal, and eval.

aide data_dir="path/to/your/data" \
     goal="Your machine learning objective in plain English" \
     eval="How to measure success (e.g., your evaluation metric)"

Argument Breakdown:

• data_dir: The file path to the directory containing your dataset files (e.g., train.csv, test.csv). AIDE ML will create a sandboxed workspace and copy this data into it.
• goal: A natural language string describing the overall objective. For example, "Build a model to predict customer churn."
• eval: A natural language string explaining the evaluation metric the agent should optimize. For example, "Maximize the F1 score on a validation set." or "Minimize the Root-Mean-Squared-Error (RMSE) between the logarithm of the predicted value and the logarithm of the observed sales price."

Advanced Options & Configuration

You can override any setting from the default configuration file directly from the command line using dot notation. This is powerful for quick experiments.

Common Flags

Here are some of the most common flags you might use:

Flag	Purpose	Default	Example
agent.steps	The number of improvement iterations the agent will run.	20	agent.steps=50
agent.code.model	The LLM used by the agent to write code.	gpt-4-turbo	agent.code.model="claude-3-5-sonnet-20240620"
agent.search.num_drafts	The number of initial solution drafts to generate before starting the improvement loop.	5	agent.search.num_drafts=3
exp_name	A custom name for your experiment run. If not provided, a random one is generated.	null	exp_name="churn-prediction-v1"
exec.timeout	The execution timeout in seconds for each script run by the interpreter.	3600	exec.timeout=600

Example with Advanced Flags

This command runs an optimization for 50 steps using Anthropic's Claude 3.5 Sonnet model and gives the experiment a custom name:

aide agent.steps=50 \
     agent.code.model="claude-3-5-sonnet-20240620" \
     exp_name="house-prices-claude-50-steps" \
     data_dir="example_tasks/house_prices" \
     goal="Predict the sales price for each house" \
     eval="RMSE between log-prices"

For a complete list of configurable options, please refer to the Configuration page.