MADS stands for Multi-Agents for Data Science, and is an open-source framework that uses multi-agents to automate the data science process. Its core goal is to let everyone apply machine learning with just two inputs.
Main capabilities and usage: MADS is positioned as a platform in which users can update a dataset, after which multiple agents automatically execute all necessary steps in the data science pipeline. Users only need to define the project goal, and the agents are responsible for completing the remaining work. In the end, users will receive: a trained model, the corresponding prediction results, and a report containing insights provided by each agent.
Installation and configuration requirements:
- The system needs Python >=3.11.7 installed.
- It is recommended to first create and activate a virtual environment: python -m venv .venv ..venv\Scripts\activate
- Install the library via pip: pip install pymads
- Before running the script, create a .env file and fill in the API Key: GROQ_API_KEY="your_api_key" or OPENAI_API_KEY="your_api_key".
- Although the default Docker code execution setting is false, some users encounter the issue that the agents still try to use the Docker engine. To avoid this, add AUTOGEN_USE_DOCKER="False" to the .env file.
Development roadmap and expansion directions (according to the to-do items listed in the materials):
- Implement a reinforcement learning agent to improve the current prompts.
- Create tools that agents can use. For example: provide the model-building agent with tools similar to the Nixtla library to improve forecasting for time series problems; create tools for the model advisor agent to optimize the selected model.
- Introduce a new agent that interacts with the data analyst and generates visualizations useful for the final report.
Applicable scenarios and nature: MADS is aimed at users who want to automatically complete the data science pipeline with minimal input, and is especially suitable for scenarios that require quickly obtaining models, predictions, and comprehensive reports. The project is open source, uses the MIT license, is built on the autogen framework, and welcomes community contributions (fork the repository, create a feature branch, submit improvements, open a pull request).
Limitations and notes: The materials do not provide specific information such as model performance, supported modeltypes, data size limits, or pricing, and these contents are not yet provided. To use it, you need to prepare a GROQ or OpenAI API Key yourself. If you encounter problems, you can create an issue on GitHub as described in the materials, and try to provide reproduction steps, expected behavior, actual behavior, and error information or stack trace.
