What is Kangas and what does this notebook use it for?

Kangas is an open-source, mixed-media dataframe-like tool from Comet. This notebook uses it to create a DataGrid that displays review data alongside 2D projections of OpenAI embeddings, allowing you to visualize and explore embedding structure interactively.

What input data do I need to use this notebook?

You need a CSV file containing text data (such as reviews) with a 1536-dimension embedding column and a label field (such as Score). The notebook will fetch example data from GitHub if you're not running inside OpenAI's own repository.

How can I interact with the resulting visualization?

The rendered DataGrid shows each row alongside its 2D embedding projection, with point colors representing the label field (e.g., review Score). You can group rows by the label field to inspect clusters and explore whether embeddings separate meaningfully by that label.

Is this tool suitable for production use?

No, this is a lightweight visualization and exploration tool rather than a production embedding pipeline component. It's designed for data scientists and developers who want to quickly inspect embedding structure interactively without building a custom plotting pipeline.

Prompt Chain

Visualize Data Embeddings with Kangas

Name: Visualizing Embeddings In Kangas
Availability: OnlineOnly
Author: OpenAI Cookbook

A notebook that builds a Kangas DataGrid with 2D embedding projections, colored by score and groupable in the browser.

Copy chain

Works with github openai

OpenAI Cookbook

Own this? Claim it

Spark score

out of 100

Updated 16 days ago

Version 1.0.0

Models

gpt 4o

Add to Favorites

Why it matters

Visualize high-dimensional data embeddings in a 2D projection using the Kangas DataGrid. This asset helps in understanding data relationships and patterns by rendering embeddings with color-coded scores.

Outcomes

What it gets done

Load data and embeddings into a Kangas DataGrid.

Convert numerical embeddings to a dedicated Embedding datatype.

Render 2D projections of embeddings, color-coded by score.

Group data points by score for analysis.

Install

Add it to your toolbox

Run in your project directory:

curl -fsSL https://spark.entire.vc/get/oai-visualizingembeddingsinkangas | bash

Steps

Steps in the chain

Setup

Constructing a Kangas DataGrid

Render 2D Projections

Overview

Visualizing Embeddings In Kangas

A notebook that loads review data and embeddings into a Kangas DataGrid, converts them into a proper Embedding column, and renders a 2D projection colored and groupable by score. Use it when you want to inspect embeddings and their projections in a dataframe-native tool colored by score - it's built around Kangas' own rendering, not a general plotting library.

What it does

This notebook builds a Kangas DataGrid containing review data alongside 2D projections of their embeddings. Kangas is an open-source, mixed-media, dataframe-like tool for data scientists built by Comet, a company focused on reducing friction in moving models to production.

After installing Kangas, the notebook reads a CSV of reviews (each with a 1536-float embedding) directly from GitHub - so the notebook works even outside OpenAI's own repo - into a DataGrid, inspects its fields and sample rows, then converts the raw embedding numbers into a proper Embedding-typed column and saves the DataGrid. Rendering the DataGrid directly in the notebook shows each row with its embedding projection visible by scrolling right; the projection point's color represents the review's Score, and grouping by Score clusters rows by group. A hosted example of the resulting DataGrid is linked for reference.

When to use - and when NOT to

Use it when you want an interactive, dataframe-native way to inspect embeddings alongside their source data and a score field - reviewing projection clusters by score, without exporting to a separate visualization tool. It's built around Kangas' own DataGrid rendering rather than a general-purpose plotting library, so it fits best when Kangas (or Comet's broader tooling) is already part of your workflow.

Inputs and outputs

Input is a CSV of text data with an associated 1536-dimension embedding per row and a score field. Output is a saved Kangas DataGrid with an Embedding-typed column and a rendered 2D projection, viewable and groupable directly in the notebook or via a hosted Kangas link.

Integrations

It uses the Kangas Python library (pip install kangas) for DataGrid construction and 2D projection rendering, reading source data from a CSV hosted on GitHub.

Who it's for

Data scientists who want to explore embeddings and their projections inside a dataframe-native tool, colored and grouped by an existing score field, rather than building a separate plotting pipeline.

Source README

Visualizing the embeddings in Kangas

In this Jupyter Notebook, we construct a Kangas DataGrid containing the data and projections of the embeddings into 2 dimensions.

What is Kangas?

Kangas as an open source, mixed-media, dataframe-like tool for data scientists. It was developed by Comet, a company designed to help reduce the friction of moving models into production.

1. Setup

To get started, we pip install kangas, and import it.

2. Constructing a Kangas DataGrid

We create a Kangas Datagrid with the original data and the embeddings. The data is composed of a rows of reviews, and the embeddings are composed of 1536 floating-point values. In this example, we get the data directly from github, in case you aren't running this notebook inside OpenAI's repo.

We use Kangas to read the CSV file into a DataGrid for further processing.

We can review the fields of the CSV file:

And get a glimpse of the first and last rows:

Now, we create a new DataGrid, converting the numbers into an Embedding:

The new DataGrid now has an Embedding column with proper datatype.

We simply save the datagrid, and we're done.

3. Render 2D Projections

To render the data directly in the notebook, simply show it. Note that each row contains an embedding projection.

Scroll to far right to see embeddings projection per row.

The color of the point in projection space represents the Score.

Group by "Score" to see rows of each group.

An example of this datagrid is hosted here: https://kangas.comet.com/?datagrid=/data/openai_embeddings.datagrid

FAQ

Common questions

Discussion