via the magic of laion_clap embeddings and umap, my live coding thingy has a sample browser at last!

I made this annotated scatter plot of 1 million FineWeb-Edu documents for @sashamtl.bsky.social's new TED talk.

Also really love how organic the plot looks with "inferno" (left) and "viridis" (right).

Map of the internet: 1.3M nodes (BGP)

The video of my talk at SciPy on DataMapPlot is up at last. If you make t-SNE or UMAP plots the talk provides some guidance on how to make plots most effective, and introduces a library to help make that easier. www.youtube.com/watch?v=-iBh...

Despite the gutting of the National Center for Educational Statistics, the dept of Ed *did* manage to release 2024 college major counts in the usual format, so I can run it through the same code I do every year. First off, the change since peak of the largest fields -- another year of drops.

I'm very much a learner, but you're maybe asking if aspects of matrix factorisation approaches to dimensionality reduction apply here. But LocalMAP is a KNN approach, with a matrix factorisation initialisation. h/t @lelandmcinnes.bsky.social for his attempts to describe these youtu.be/9iol3Lk6kyU

📢 Save the date! Join us for the next @ellis.eu x UniReps Speaker Series! 📅 27th August – 16:00 CEST 📍https://ethz.zoom.us/j/66426188160 🎙️ Speakers: Keynote by @lelandmcinnes.bsky.social & Flash Talk by Yu (Demi) Qin 🔔 Stay updated by joining our Google group: groups.google.com/u/2/g/ellis-...

🚀 We've just open-sourced Embedding Atlas – a tool for exploring large embedding spaces through rich, interactive visualizations 📊.

Meteoroid stream identification with HDBSCAN unsupervised clustering algorithm. Eloy Peña-Asensio et. al. https://arxiv.org/abs/2507.01501

Ever wanted to pan through the latent🌌 space of TikTok videos? Made using the amazing toponymy and datamapplot from @lelandmcinnes.bsky.social and data from mine and @jurgenpfeffer.bsky.social 's first complete TikTok slice. link below

🎤 Speaker Spotlight: Leland McInnes Join Leland at #SciPy2025 for his talk "DataMapPlot: Rich Tools for UMAP Visualizations." 📊 Discover powerful new ways to explore high-dimensional data! 🔗 scipy2025.scipy.org

Explore Wikipedia through a data map. Pages are grouped by semantic similarity, for topic clusters. Hover to see details, zoom to explore more fine-grained topics, click to go to a page. Search by page name to find interesting starting points for exploration. lmcinnes.github.io/datamapplot_...

I'll be giving a talk about DataMapPlot for visualizing data maps at Scipy this year. I would love to meet potential users and chat about where to go next. cfp.scipy.org/scipy2025/ta...

I also updated the ArXiv data map example to make use of new features in datamapplot. lmcinnes.github.io/datamapplot_... You can tweak parameters and build your own version: gist.github.com/lmcinnes/e11...

OMG I am so glad someone finally did this. Thank you 🙏 @lelandmcinnes.bsky.social This will now consume hours and hours of my time. lmcinnes.github.io/datamapplot_...

I also updated the ArXiv data map example to make use of new features in datamapplot. lmcinnes.github.io/datamapplot_... You can tweak parameters and build your own version: gist.github.com/lmcinnes/e11...

Great idea. Did no one think of this before?

Explore Wikipedia through a data map. Pages are grouped by semantic similarity, for topic clusters. Hover to see details, zoom to explore more fine-grained topics, click to go to a page. Search by page name to find interesting starting points for exploration. lmcinnes.github.io/datamapplot_...

🔥 Meet our Keynote Speakers for #SciPy2025! Dr Malvika Sharan, co-Director of Open Life Science (OLS) and a senior researcher at The Alan Turing Institute will be sharing with us her expertise at our favorite conference. You can't miss her ➡️ hubs.la/Q03sdlsb0

🔥 Meet our Keynote Speakers for #SciPy2025! Hon. Dr. Kathryn D. Huff 🇺🇸, nuclear engineer, policy leader, and former Assistant Secretary for the Office of Nuclear Energy will be joining us in Tacoma! 🙌 Don't miss her talk, grab your ticket now: hubs.la/Q03sdlsb0

Nature Reviews Methods Primers: Uniform manifold approximation and projection (UMAP) www.nature.com/articles/s43... 🧬🖥️🧪 read free: rdcu.be/d0YZT

Our latest paper is out: peerj.com/articles/cs-.... We added functionality to the #HDBSCAN clustering algorithm to also detect branches hdbscan.readthedocs.io/en/latest/ho... #eda #datavis #clustering

Original paper: www.nature.com/articles/s41...

No microscope? No problem! A new spatial transcriptomics method developed at the Broad lets you map gene expression with zero imaging needed. Perfect for big tissue samples and small labs! 🔬💡 #spatialtranscriptomics #bioinformatics #transcriptomics #genomics #biology #harvard #broadinstitute

If you want to annotate quickly ... embeddings/UMAP/live interaction can be your best friends! Just embed everything in two dimensions and inspect the clusters, annotating as you go along.

Sure there's more than a couple dozen… But why not? Burrows Wheeler transform was just a random DEC technical report in 1994. Today that would mean just posting to arxiv and never bother w/ peer review. Actually, even more on point that's exactly what UMAP is -- arxiv.org/abs/1802.03426

Announcing BERTopic v0.17 🥳 This is a feature-packed update that includes the amazing Model2Vec, more interactive DataMapPlot functionalities, a method for lightweight installation, and much more!

1. Struggling to integrate single-cell datasets? Finding it hard to resolve clear differentiation trajectories? Reveal the underlying structure in your data with CONCORD.

Did #colliderfest today on the Viper HPC stand! It was fun talking to people about supercomputing/HPC and AI, but also exhausting. https://colliderfest.co.uk/ Used my UMAP word cloud demo to talk about how foundational language models (e.g. like ChatGPT, Ollama, etc) work, and how […]

To learn in depth about UMAP + tSNE + PCA, please join my Instats seminar: Linear and Nonlinear Dimensionality Reduction instats.org/seminar/line... on the 26th of February 2025

A note on the relations between mixture models, maximum-likelihood and entropic optimal transport As they note, this is an interesting summary written for clarity rather than a new result. But, I've long tried to convince people that all of ML is deeply interconnected. arxiv.org/abs/2501.12005

Lovely example of how weird real-world data is. I put all 55 million points in Overture maps into a Nomic Atlas map. Lots of points spuriously appearing in the ocean. But the 1° square around 'null island' (0° lat, 0° long) is suspiciously clear -- someone has cleared out bad data just there.

Exploring the hidden states of #ModernBERT with UMAP umap-learn.readthedocs.io/en/latest/

arxiv.org/abs/2412.10924 - Tokens, the oft-overlooked appetizer: Large language models, the distributional hypothesis, and meaning. by Julia Witte Zimmerman et al. from 1Computational Story Lab at UVM Super fun figures!

really interesting SAE project with Llama (with api and data map) from Goodfire — including interesting research areas for steering vs factuality www.goodfire.ai/papers/mappi...

Learn how to compute embeddings with @motherduck.com and @marimo.io, visualize them with UMAP, and host the resulting notebook on @hf.co — all in our latest blog!

Hi NeurIPS! Explore ~4,500 NeurIPS papers in this interactive visualization: jalammar.github.io/assets/neuri... (Click on a point to see the paper on the website) Uses @cohere.com models and @lelandmcinnes.bsky.social's datamapplot/umap to help make sense of the overwhelming scale of NeurIPS.

Spent the day playing with this. I'm absolutely blown away @enjalot.bsky.social! - Chose any embedding from HF - Project with UMAP, cluster with HDBSCAN - Use Ollama to label the clusters (Works incredibly well!)

A first pass at UMAP on a sphere! The scikit-learn digit dataset, embedded with Python, visualized with JS, packaged & deployed with Observable Framework: pamacha.observablehq.cloud/spherical-um...

New blog post! Updated for 2024, my favorite example of why alphabetical ordering is bad for geographic features -- US presidential results since 1828. The left image shows regional patterns in a geographic ordering that the right (alphabetical) simply loses. benschmidt.org/post/2024-11...

Good, published, benchmarks of machine learning / data science is crucial. But so hard. Well-cited "SOTA" methods typically crash often. They tend to be very computational expensive. Both make a systematic study impossible. Finally, reviewers always ask for more methods, and more "SOTA".

I've organized and participated in many unconferences in the past, and they are always the most intense exchange of ideas and information that I've experienced. Given the energy we're seeing in the registration this one is poised to be no different! register today! hiddenstates.org

We've hit a critical mass of registrations! The caliber of attendees is exciting, we've got researchers from companies big and small, academic and indie. We've got prototypers and UXers who have worked on bleeding-edge interfaces as well as house-hold names. let's talk about the unconf experience:

Hidden States is happening next week in SF! It's a one-day unconference gathering researchers, designers, prototypers and engineers interested in pushing the boundaries of AI interfaces, going below the API and working with the hidden states. hiddenstates.org

I'm working on a project that looks at video game box art by captioning the boxes with multimodal models. Still need to edit some of the topics, but a lot of interesting patterns: atlas.nomic.ai/data/nabsidd... #AI, #computervision, #rstats, #digitalhumanities, #dh

Uniform manifold approximation and projection is a nonlinear dimension reduction method used for visualizing data and as preprocessing. This Primer introduces the UMAP algorithm, intuitions behind it, how best to apply it, and how to interpret and understand results. www.nature.com/articles/s43...

One way I've been thinking about ML models for some time is as a lens. The weights are crystalized patterns whose structure emerges from the crushing pressures of backpropagation. By shining a piece of data through this lens you see the patterns diffracted in the hidden states.

I recently wrote a primer on UMAP for Nature Reviews Primers. If you are looking for an overview of the method, a getting started primer, or best practices it is a good place to start. rdcu.be/d0YZT

🧵 A huge thanks to the creators of umap-learn for their work on UMAP! While trying to get parametric UMAP to work for my project, I ran into some challenges, so I decided to dive in and build my own PyTorch-based version from scratch. You can check it up here github.com/mr-fcharles/...