LIDA

LIDA is a tool designed to streamline data exploration, producing visualizations and infographics via large language models (LLMs) such as ChatGPT and GPT4.

It features a conversational approach for automatically creating visualizations from data, regardless of the grammar used. LIDA is structured into four modules: the Summarizer, which provides a brief natural language summary of the data; the Goal Explorer, which lists potential visualization objectives based on the data; the VisGenerator, which creates, improves, runs, and selects visualization code; and the Infographer, which generates accurate and stylized graphics using image generation models. LIDA supports any programming language or visualization grammar, enabling users to develop visualizations in languages like Python (using Altair, Matplotlib, Seaborn), R, and C++.

The tool also allows for operations on existing visualizations, including explanation, self-assessment, automated repair, and suggestions. Capabilities include data summarization, automated data exploration, versatile visualizations, and infographic creation.

It utilizes the language and code generation capabilities of LLMs to deliver key automated visualization features. LIDA can also perform operations on existing visualizations, such as explaining them, assessing their quality, fixing them, and recommending improvements. LIDA's design merges LLMs and image generation models (IGMs) to manage the complex process of creating visualizations.

It is an open-source tool that offers a Python API and a user interface for making interactive charts, infographics, and data-driven narratives. While LIDA may struggle with visualization grammars not well-represented in the LLM's training data and its performance can vary depending on the visualization libraries and code generation capabilities used, it is still a valuable tool for automating visualization.