Accelerate tSNE with openTSNE

Scale tSNE to millions of data points.

Instantly build and test tools for your agents!

Robust and properly tested tools are the key to building powerful agents. Otherwise, your Agents will not hold much practical value.

Postman’s AI Agent Builder provides a Tool Generation API—a massive time saver for anyone working with AI agents.

With just three clicks, you can:

• Find an API in Postman’s Network.

• Pick a programming language and agent framework.

• Generate production-ready code that integrates seamlessly into your AI agent.

Postman’s AI Agent Builder also offers:

• Quick API and LLM multiple testing

• Compare responses, costs, and performance

• A no-code visual canvas for building multi-step agent workflows

Overall, Postman’s AI Agent Builder will save you plenty of time and manual workload if you're working with agentic workflows.

Thanks to Postman for partnering on today’s issue!

---

Accelerate tSNE with openTSNE

Yesterday, we learned how to accelerate tSNE using GPUs with the tSNE-CUDA library. Here’s the visual from that post for a quick recap:

But what if you don’t have a GPU?

openTSNE is another optimized Python implementation of tSNE that lets us scale t-SNE to millions of data points.

Note: We discussed tSNE in complete detail and implemented it from scratch here →

A comparison is shown in the image below:

As depicted above, the openTSNE implementation is 20 times faster than the Sklearn implementation.

As per the benchmarks shown below:

• openTSNE can produce a low-dimensional projection of 1M data points in just ~15 minutes.

• Sklearn implementation takes two hours with just ~250k data points.

Further reading:

• While this was just about tSNE, do you know we can accelerate other ML algorithms with GPUs? Read this article to learn more: Sklearn Models are Not Deployment Friendly! Supercharge Them With Tensor Computations.

• Also, do you know how tSNE works end-to-end? We formulated and Implemented the t-SNE Algorithm From Scratch here →

Here’s the colab notebook if you want to try openTSNE: openTSNE Colab notebook.

👉 Over to you: What are some other ways to boost the tSNE algorithm?

Thanks for reading!