Skip to main content
The run command is the primary entry point for starting an ML experiment session. It orchestrates the entire autonomous experimentation process from data profiling to final report generation.

​
Command Syntax

autopilot run [OPTIONS]

​
Required Arguments

Three arguments are required for every run:
​
--data
path
required
Path to your dataset file. Supports CSV and Parquet formats.Short form: -dExample:
autopilot run --data train.csv --target price --task regression
​
--target
string
required
Name of the target column for prediction. Must exist in your dataset.Short form: -tExample:
autopilot run --data data.csv --target SalePrice --task regression
​
--task
enum
required
Type of machine learning task. Must be either classification or regression.Options:
  • classification - For predicting categorical outcomes
  • regression - For predicting continuous values
Example:
autopilot run --data customers.csv --target churned --task classification

​
Optional Arguments

​
--constraints
path
Path to a constraints file in Markdown format. Use this to guide the autopilot’s experimentation strategy.Short form: -cDefault: NoneExample:
autopilot run -d data.csv -t price --task regression -c constraints.md
See Constraints for details on writing constraint files.
​
--max-iterations
integer
Maximum number of experiment iterations to run. The autopilot will stop after this many iterations, even if the time budget hasn’t been exhausted.Short form: -nDefault: 20Range: 1-100Example:
autopilot run -d data.csv -t price --task regression --max-iterations 50
​
--time-budget
integer
Time budget in seconds for the entire experiment session. The autopilot will stop when this time limit is reached.Default: 3600 (1 hour)Range: 60-86400 (1 minute to 24 hours)Example:
# Run for 2 hours
autopilot run -d data.csv -t price --task regression --time-budget 7200
​
--output-dir
path
Directory where results will be saved. If not specified, a timestamped directory will be created automatically in ./outputs/.Short form: -oDefault: Auto-generated (e.g., ./outputs/experiment_20240315_143022/)Example:
autopilot run -d data.csv -t price --task regression -o ./my_experiment
​
--verbose
boolean
Show detailed Gemini reasoning and thought processes during experimentation. Useful for understanding why the autopilot makes certain decisions.Short form: -vDefault: falseExample:
autopilot run -d data.csv -t price --task regression --verbose
​
--resume
path
Resume from a previously saved state file. Use this to continue an interrupted experiment.Default: NoneExample:
autopilot run --resume ./outputs/experiment_20240315_143022/state.json
When resuming, the state file contains all configuration from the original run, so you don’t need to specify --data, --target, etc. again.

​
Workflow

When you run the command, the autopilot executes the following workflow:
  1. Configuration - Validates arguments and loads API credentials
  2. Data Profiling - Analyzes dataset characteristics, distributions, and quality
  3. Experiment Design - Gemini creates an experiment plan based on data profile and constraints
  4. Experiment Execution - Runs experiments, tracks metrics, and logs results
  5. Analysis & Iteration - Analyzes results, generates hypotheses, and plans next experiments
  6. Reporting - Generates final report with best models and insights

​
Examples

​
Minimal Example

Simplest possible command with just required arguments: 
autopilot run --data train.csv --target SalePrice --task regression

​
Production Example

Full-featured command for a production run: 
autopilot run \
  --data housing_data.csv \
  --target price \
  --task regression \
  --constraints production_constraints.md \
  --max-iterations 50 \
  --time-budget 14400 \
  --output-dir ./production_run \
  --verbose

​
Classification Example

Customer churn prediction: 
autopilot run \
  -d customer_data.parquet \
  -t churned \
  --task classification \
  -c churn_constraints.md \
  -n 30 \
  -v

​
Resume Example

Continuing an interrupted experiment: 
# Original run (interrupted with Ctrl+C)
autopilot run -d data.csv -t price --task regression -n 100

# Resume from saved state
autopilot run --resume ./outputs/experiment_20240315_143022/state.json

​
Exit Codes

  • 0 - Successful completion
  • 1 - Error or user interruption (Ctrl+C)

​
Output Structure

The output directory contains: 
outputs/experiment_20240315_143022/
├── state.json              # Resumable state file
├── experiments/            # Individual experiment results
│   ├── exp_001.json
│   ├── exp_002.json
│   └── ...
├── models/                 # Saved model artifacts
│   ├── best_model.pkl
│   └── ...
├── plots/                  # Visualization outputs
│   ├── feature_importance.png
│   └── ...
└── report.md              # Final experiment report

​
See Also

Build docs developers (and LLMs) love

Get started for freeTalk to us