CanvasXpress AI and Copilot

CanvasXpress AI: Powered by LLM

CanvasXpress AI is a feature that allows users to interact with their data using large language models (LLMs). This functionality enables you to explore data and create visualizations or code with a single click by simply describing what you want in plain English. You can use provided datasets or upload your own files via drag-and-drop.

Below are examples of how CanvasXpress AI works with different data types and visualizations.

Chatting with Data in CanvasXpress

Here are some examples of how to interact with CanvasXpress AI using large language models (LLMs). You can either use the provided datasets or drag and drop your own tab-delimited file into any of the CanvasXpress visualizations (currently shown as data tables for clarity). Simply describe in plain English what you want the visualization to look like, including the column names you want to use.

Examples - Update on a table

• Create a Boxplot of len groupped by supp and dose showing the original data jittered.
• Create a Violin plot of len by dose in a vertical orientation. Rotate the sample labels 90 degrees.
• Create a Bar graph of len color by supp. Orient vertically and show legend on top in two columns.

Examples - Update on a plot

• Create a Scatter plot of cty and hwy color by class.
• Create a Regression plot of cyl and displ.
• Create a Loess fit of cty and hwy color by class.

Examples - Append on a plot

• Create a Scatter plot of mpg and wt.
• Create a Filled contour of mpg and wt.

Copilot for CanvasXpress

**Copilot for CanvasXpress** is an integrated feature designed to help users create effective prompts for configuring parameters accurately. This functionality was instrumental in generating the few-shot examples required to train our LLM model.

When you use Copilot, there is no need to ask the LLM to generate visualizations. Instead, CanvasXpress will generate the configuration parameters needed for the visualizations. If the prompt cannot be validated, CanvasXpress will forward it to the LLM for processing.

Setting up the correct configuration isn't complicated. It's simply about translating the parameters into plain English—there's no need to craft it into poetry.

How it works

Copilot can be enabled or disabled using a toggle widget in the chat interface. As you start typing, you'll see a set of initial suggestions for how to begin your new sentence. You can simply press the tab key to accept a suggestion. If there are multiple options available, pressing tab will select the first one in the list.

Initial Suggestions

• Create a
• Orient the graph
• Set the title for the
• Increase the font size of the
• Decrease the font size of the
• Change the color of the
• Set the color of the
• Color by
• Shape by
• Size by
• Group by
• Show original data
• Jitter the data
• Transform the x-axis to
• Transform the y-axis to
• Change the scale of the
• Hide the legend
• Show the legend
• Put the legend on the
• Rotate the
• Use the
• Exclude data where
• Include data with
• Filter data where

Based on the initial suggestion, Copilot will propose ways to complete the sentence. Several of these suggestions exhibit a well-defined structured format.

Example of a Structured Suggestion

• Create a **`{graph}`** of **`{column}`** **`{concatenator}`**

When you begin typing your suggestion, Copilot will automatically complete the sentence with the available graph types in CanvasXpress, such as bar graph, line graph, scatter plot, and more. Once you select the graph type, Copilot will add the term 'of' and suggest the columns from your dataset. After you finish typing the first column, Copilot will propose a comma, a period, or 'and' to help you concatenate the entries.