Image created by the author with Midjourney
MetaGPT is a multi-agent framework for assigning roles to multiple agents leading to the formation of collaborative entities that can work together to execute complex instructions. MetaGPT bills itself as a “software as a multi-agent system company,” which gives you an idea of the intended use of these collaborative entities. MetaGPT can be used as a standalone application from the command line and as a library within your own Python scripts, allowing for the flexibility and control one would want in such a framework.
The project started in April 2023, leveraging ChatGPT, and at the time of writing has almost 40,000 stars on GitHub. Its GitHub repository describes itself as follows:
MetaGPT takes a one line requirement as input and output of user stories/competitive analysis/requirements/data structures/APIs/documents etc.
Internally, MetaGPT includes product managers/architects/project managers/engineers. It provides the entire process of a software company along with carefully orchestrated SOPs.
MetaGPT software company multi-agent scheme (gradual implementation) (from MetaGPT GitHub)
MetaGPT can be used for code generation, prototyping, project planning, and more. It has been recognized as a outstanding open source achievementand is continually a trending GitHub repository.
That's MetaGPT. Now let's talk Data interpreter, ai/” rel=”noopener” target=”_blank”>deep wisdomThe latest improvement to MetaGPT and an achievement in its own right.
Complete video to introduce the MetaGPT data interpreter
Showing how to address electric load forecasting challenges through dynamic planning, tool utilization, improved reasoning, and experience-based verification.
Repository: https://t.co/xWGS0UF9oW
Cases: https://t.co/GhNH54Ahhi… pic.twitter.com/Xc5aam1TXz—MetaGPT (@MetaGPT_) March 19, 2024
Data Interpreter is another member agent of the MetaGPT framework, an agent dedicated to evaluating and solving data-related tasks. From the newspaper:
In this study, we present Data Interpreter, a solution designed to solve with code that emphasizes three fundamental techniques to increase problem solving in data science: 1) dynamic planning with hierarchical graphical structures for real-time data adaptability; 2) dynamic integration of tools to improve code mastery during execution, enriching the necessary experience; 3) identification of logical inconsistencies in feedback and improving efficiency by recording experiences. (…) Compared to open source baselines, it demonstrated superior performance, showing significant improvements on machine learning tasks, increasing from 0.86 to 0.95. Additionally, it showed a 26% increase on the MATH dataset and a notable 112% improvement on the open tasks.
These findings are certainly impressive. And there is no need to take them literally, since they have published these results. Deep Wisdom has also made available a ai/main/en/DataInterpreter/” rel=”noopener” target=”_blank”>multitude of examples to show how their Data Interpreter agent can be used alongside the existing MetaGPT framework.
ai/main/en/DataInterpreter/detail.html?id=NVIDIAStockTrendAnalysis” rel=”noopener” target=”_blank”>This example here shows how it can be used for NVIDIA stock trend analysis. To see what the MetaGPT data interpreter message looks like, I'll duplicate it below:
Get NVIDIA Corporation (NVDA) stock price data from Yahoo Finance, focusing on historical closing prices for the last 5 years. Summary statistics (mean, median, standard deviation, etc.) to understand the central tendency and dispersion of closing prices. Analyze data for notable trends, patterns, or anomalies over time, potentially using moving averages or percentage changes. Create a chart to visualize all the data analysis. Reserve 20% of the data set for validation. Train a predictive model on the training set. Report model validation accuracy and visualize prediction result. near
You can check out the example notebook (linked above) to follow the MetaGPT process and see the results. Spoiler alert: Deep Wisdom doesn't share them because they aren't awesome 🙂
Read the complete document for all the information you can ask for. You can learn more about installation and usage on the project page. GitHub repository. I can attest from experience that MetaGPT is worth a try, and with the addition of the Data Interpreter agent, this is even more true than before.
Matthew May (@mattmayo13) has a master's degree in computer science and a postgraduate diploma in data mining. As Editor-in-Chief of KDnuggets, Matthew aims to make complex data science concepts accessible. His professional interests include natural language processing, machine learning algorithms, and exploring emerging ai. He is driven by the mission to democratize knowledge in the data science community. Matthew has been coding since he was 6 years old.
