funsearch

My implementation of https://deepmind.google/discover/blog/funsearch-making-new-discoveries-in-mathematical-sciences-using-large-language-models/

Local Code Execution Sandbox with SQLite Database

Prerequisites

• Docker installed on your machine. Get Docker

Setup

Make sure you execute all 'docker' commands from the /sandbox directory.

1. Navigate to the sandbox directory:

cd sandbox

2. Create a .env file in the sandbox directory with the following content:

OPENAI_API_KEY=your_openai_api_key GROQ_API_KEY=your_groq_api_key # Optional, but make sure none of the seed_db examples use groq.

3. Build the Docker image:

docker build -t code-sandbox .

Usage

Run a container from the image:

docker run -it --rm -v $(pwd)/data:/data -v $(pwd)/scripts:/sandbox/scripts code-sandbox

This command will:

1. Initialize the SQLite database (if it doesn't exist)
2. Mount the local scripts directory to /sandbox/scripts in the container
3. Start a Python interactive shell

The database will be stored in the data directory, which persists between container runs.

Interacting with the Database

In the Python shell, you can interact with the SQLite database:

import sqlite3

conn = sqlite3.connect('/data/example.db')
cursor = conn.cursor()

# Query the users table
cursor.execute("SELECT FROM users")
print(cursor.fetchall())

conn.close()

Customizing Initialization

You can modify the init.sh script in the project directory to customize the startup process or add more initialization steps.

Maintenance

Rebuild the Docker image when you change the Dockerfile, requirements.txt, or init.sh:

docker build -t code-sandbox .

If you update .env or scripts after building, you don't need to rebuild. These are mounted at runtime.

Updating Scripts

The scripts directory is mounted as a volume, so any changes made to the local scripts directory will be immediately reflected in the container. You don't need to rebuild the image to update scripts.

Note for Contributors

• Do not commit your .env file to the repository if it contains sensitive information.
• Update requirements.txt if you add new Python packages.