Combine the power of LLMs and R to help guide exploration of a dataset.
DISCLAIMER: This package is a proof of concept and was created for a 2-day hackathon. It's currently just a fun side project. Don't use it for anything serious.
You can install the development version of aidea from GitHub with:
remotes::install_github("cpsievert/aidea")To use this package, you'll also need credentials for the LLM that powers assist().
By default, assist() uses OpenAI, so you'll need to set an environment variable named OPENAI_API_KEY using the key from https://platform.openai.com/account/api-keys
We recommend setting that variable via usethis::use_renviron(). See {elmer}'s prerequisites if you plan on using a different model.
This package currently contains just one function, assist(), which takes a data frame as input, and provides a chat bot experience tailored for that dataset:
# Load a dataset
data(diamonds, package = "ggplot2")
# Start the aidea app assistant
aidea::assist(diamonds)You'll be welcomed with overview of what's in the data (e.g., interesting summary stats, variable types, etc) as well as some questions to ask about the data.
When you ask a question about the data, it'll offer R code to assist in answering that question.
That R code will include an option to run the code in browser:
When clicked, the code is run in a sidebar, and results displayed below the interactive code editor.
When you're unsure of how to interpret the results, press the interpret button. This will open an additional sidebar with an interpretation of the current results:
This package uses a combination of {elmer} and {shinychat} to provide the LLM assisted chatbot experience.
It does not send all of your data to the LLM, just basic summary stats (e.g., number of rows/columns) and data characteristics (e.g., variable types).
It will, however, send any results you choose to interpret to the LLM.
If you are worried about privacy, consider using a local model (i.e., assist(chat = elmer::chat_ollama())) instead of OpenAI