John Sheffield

Major news at the intersection of climate risk and social justice from my colleagues ⁦@climate_risQ⁩ politico.com/news/2021/07/0…

The Functional API in Keras is super concise -- it lets you define complex models in about half as much code as the subclassing approach. It also saves you a lot of debugging time, because it checks all input compatibility assumptions at construction time, long before you run the model. Any model you can build, will run. This is similar to using statically typed code -- but for layers.

the absolute best intro to spatial book that I recommend to everyone.

We spent a year developing cde-small-v1, the best BERT-sized text embedding model in the world. today, we're releasing the model on HuggingFace, along with the paper on ArXiv. I think our release marks a paradigm shift for text retrieval. let me tell you why👇

Conversely, some folks believe that RAG is pretty much the point of embeddings, but there's so much more. Embeddings let you unlock the potential of unstructured data -- for exploration, for curation, for explainable models, and data understanding in general.

It took me SEVEN years to figure out that you can save a ggplot as a vectorized .svg file ggsave("file.svg"). The svg file can be opened on powerpoint and it will be completely vectorized, even text are still recognized as text boxes! No more recreating plots on illustrator!!!!!

The Barcelona clay render seemed popular, so here's the full res version for your perusing pleasure. Have fun, but it's a rather large file, though (uncompressed tif). drive.google.com/file/d/1EPOGXZ… #Barcelona #Spain #arcgispro #blender

If you could get a clustering algorithm and library specifically designed for fast clustering of embedding vectors (CLIP, sentence-transformers, Cohere-embed, etc.), what features would you most want it to have?

Wanted to share the good news that "Build an LLM from Scratch" is now in its final stages. Currently putting on the final touches on classification-finetuning chapter, and then moving on to the last chapter: instruction finetuning! The new estimated publication date is now Summer 2024. (In the meantime, there's currently a discount on the Manning website: manning.com/books/build-a-…)

And then I’ll make my annual exhortation: a relatively small number of geeks thinking they are playing a 45 year game will find they were actually playing a 15 or 20 year game due to issues mostly beyond their control. Buy term life and own occupation disability insurance.

Machine Learning can be an overwhelming field when you first start out. But with years of study, experience, and hard work you'll realize that you know even less than you had thought.

Hi #gischat — I’m hiring for a senior spatial analytics role in Sustainable Finance at Intercontinental Exchange. We play with data on climate, housing, remote sensing, social/demog, lots more. If you love GDAL and fun problems, please apply/say hi in DMs! egdd.fa.us6.oraclecloud.com/hcmUI/Candidat…

Ever wanted to do spatial clustering of orign-destination (OD) data? Leeds ITS + Turing PhD student Hussein Mahfouz - @h_mahfouz - has created this early-stage visualisation of spatial clustering of these zone-zone flows 🏗️ Case study of #Leeds. Looks beautiful AND useful 🎉

High-resolution satellite images can be insanely expensive to buy. So here's a list of free datasets you can access. These datasets can be used to build foundation models, super-resolution models, or for segmentation.

There is a new blog post at #geocompx: "An overview of the rsi R package for retrieving satellite imagery and calculating spectral indices" by Mateusz Rydzik. Read it at geocompx.org/post/2024/rsi-… #rstats #remotesensing #stac

Best resources for learning about RAGs? (Assume no prior knowledge about it). cc @HamelHusain @abacaj

Say hi to ‘Geocomputation with Python’! Our new blog post introduces the book about working with spatial data in Python. geocompx.org/post/2023/geoc… #geopython #python #geocompx #gischat

Just wrapped up the final lecture of my Spatial Data Management course this semester, diving into lightning-fast ⚡️analysis of massive geospatial 🌎 datasets with the power of DuckDB 🦆 🎥 Video: youtu.be/VSXTyKWmDVU 📒 Notebook: geog-414.gishub.org/book/duckdb/11… 🎶 Playlist: bit.ly/geog-414 #geospatial, #duckdb, #dataviz #geemap #leafmap

"Simplifying Transformer Blocks" ranks easily among my favorite research papers that I've read this year. Here, the authors look into how the standard transformer block, essential to LLMs, can be simplified without compromising convergence properties and downstream task performance. Based on signal propagation theory and empirical evidence, they find that many parts can be removed to simplify GPT-like decoder architectures as well as encoder-style BERT models: skip connections normalization layers (LayerNorm) projection and value parameters sequential attention and MLP sub-blocks (in favor of a parallel layout) The authors also did a great job referencing tons of related work motivating their experiments. I definitely recommend reading this paper just for the references alone: arxiv.org/abs/2311.01906

New blog post highlighting a suite of Jupyter Notebooks to enable programmatic access to cloud-hosted USGS 3D Elevation Program (3DEP) #lidar data. Notebooks demonstrate how to use @pointcloudpipe to access & process data hosted by Amazon Open Data. opentopography.org/blog/new-colle… 1/2

New collection of Jupyter Notebooks enables programmatic access to cloud-hosted USGS 3D Elevation Program (3DEP) lidar data Blog: opentopography.org/blog/new-colle… GitHub: github.com/OpenTopography… #lidar #dataviz #3dep