PyData Seattle 2023
2023-04-26..2023-04-28
Seattle, United States
[PyData Seattle 2017, PyData Seattle 2023, PyData Seattle 2025]
who it's for
data scientists, machine learning engineers, data engineers, MLOps engineers, python developers, open source project maintainers
topics
Ray distributed tasks and actors; Spark, Dask and Fugue portability; Delta Lake ACID, time travel and Z-ordering; MLflow experiment reproducibility; FLAML automated machine learning and tuning; prompt engineering with GPT and LangChain; embeddings, FAISS and ANN search; SymPy symbolic physics and exact collision timing; XGBoost model training and serverless deployment; Quarto and publishing Jupyter notebooks; Apache Sedona geospatial data analysis; Jupyter AI and generative AI in notebooks
| title | author | length |
|---|---|---|
| Automated Machine Learning & Tuning with FLAML | Jiang et al. | 01h 22m |
Library demo shows actual wins not promises: R^2 ~0.7→0.8 on a regression, a few-percentage-point classification lift (66.7→67.3) and AUCs reported up to ~0.81… | ||
| Build a production ML system with only Python on free serverless services | Jim Dowling | 01h 23m |
An end-to-end Python system predicts PM2.5 for Seattle by scraping live sensors and weather forecasts, writing features to Hopsworks, training an XGBoost model… | ||
| Building a Search Engine | Nidhin Pattaniyil | 01h 30m |
Demo builds a working search stack: index 25k Unsplash captions into BM25 on Lucene (Elasticsearch) at about 5000+ docs/sec, then compare against embeddings… | ||
| Building Reliable, Open Lakehouses with Delta Lake | Jim Hibbard | 01h 28m |
Shows a reproducible Docker demo environment and runs code that writes parquet files into a Delta table, updates the Delta transaction log and demonstrates… | ||
| Porting Existing Python and Pandas Code to Spark, Dask, and Ray | Kho & Holten | 01h 27m |
Porting Python/Pandas code uses Fugue to run the same functions on Spark, Dask, and Ray by calling a single transform interface, requiring an explicit schema… | ||
| Going beyond ChatGPT: introduction to prompt engineering & LLMs | Ties de Kok | 01h 00m |
Covers a concrete pipeline: uses OpenAI API (GPT-3.5/4) and LangChain in a Jupyter notebook on GitHub/Binder to classify app feedback by building a prompt… | ||
| Introduction to Working with U.S. Census Data in Python | Darren Vengroff | 01h 27m |
Shows that the Census Bureau exposes 1,300 data sets via APIs, demonstrates using open‑source Python tooling to discover and download ACS variables like… | ||
| Introduction to Ray for distributed and machine learning applications in Python | Jules S. Damji | 57m |
Hands-on coding shows Ray Core 2.0 turning Python functions into distributed tasks, actors, object store primitives that run on laptop or cloud and integrate… | ||
| Keynote: Does Open Data Science Lead to the End of the World? | Peter Wang | 42m |
Argues that open data science drove rapid adoption by lowering barriers, cites partnerships with Snowflake Microsoft Azure and Oracle, and names Pi script (a… | ||
| Keynote | Travis E. Oliphant | 44m |
History of building and distribution dominates: creation of NumPy and SciPy, an early workshop that became PyData in 2012, and a packaging focus that produced… | ||
| Building a ML Model Observability pipeline | Prabhakar & Saha | 01h 21m |
Claim: a production-ready observability pipeline is implemented using an open source Python stack and it actually scales. It profiles inference logs with y… | ||
| Robust and End-to-End Cloud Native Machine Learning & Data Processing Platform | Eduardo Apolinario | 01h 24m |
Claims a production-ready open source orchestration: Flyte runs containerized, versioned tasks on Kubernetes and converts data between runtimes (PySpark… | ||
| Hands-on intro of ipyvizzu-story | Peter Vidos | 01h 31m |
Hands-on demo shows an open‑source chart morpher called ipyvizzu-story that runs in Jupyter Notebook and via a Streamlit integration, with a C++ core compiled… | ||
| Introduction to Ray for distributed and machine learning applications in Python | Jules S. Damji | 01h 31m |
Introductory, hands-on tour of Ray Core 2.0 walks code not concepts: it converts Python functions into remote functions and actors, demonstrates the shared… | ||
| Solving complex ML problems in a few lines of code with AutoGluon | Erickson & Shirkov | 01h 25m |
AutoGluon claims it trains multimodal image‑text‑tabular data with a few lines of code and actually builds a multi-layer stack ensemble of transformer image… | ||
| Panel: Building a Stronger Open Source Python Data Community | Gomez et al. | 01h 30m |
Panel lists concrete tools and actions: nbdev and quarto for literate workflows, black and rough for code quality, and Hamilton for pipeline opinions. It names… | ||
| a workbench for creating scikit-learn like parametric objects and libraries | Betchel & Kiraly | 01h 04m |
skbase is a working meta‑toolkit that makes it easy to build scikit‑learn‑style libraries: it provides BaseObject/BaseEstimator classes, a tag and config… | ||
| A Perfect, Infinite-Precision, Game Physics in Python | Carl Kadie | 36m |
Builds a perfect infinite-precision engine in Python using SymPy and runs a live demo and article, producing exact collision times and symbolic velocity… | ||
| Building Machine Learning Microservices & MLOps using Union ML | Shivay Lamba | 36m |
UnionML claims to turn notebooks into microservices with decorator functions (reader, trainer, predictor, evaluator) so data pipelines avoid boilerplate. It is… | ||
| Enterprise-grade Full Stack ML Platform | Goyal & Ramakrishnan | 36m |
Describes an enterprise ML platform built at Dell using Metaflow and Outerbounds that claims to let data scientists run end-to-end workflows without deep SRE… | ||
| Keynote: Distributed Computing 4 Kids -- with Spark | Holden Karau | 44m |
A concrete claim: kids can learn distributed systems concepts first and syntax later by mapping work to garden gnomes and using Spark as a teaching vehicle… | ||
| Keynote: Scientific Computing and the Gateway to Open Source | Katrina Riehl | 01h 03m |
History and impact of scientific open source gets concrete claims: NumFOCUS fiscally sponsors 120 affiliated projects, routes millions in funding while… | ||
| Let’s program to fight the impacts of climate change! | Ying-Jung Chen | 36m |
Claims concrete numbers: training a GPT‑3–scale model uses ~1,287 MW and produced ~552 tons CO2, and ChatGPT emits ~23.04 kg CO2 per day; shows tools and demos… | ||
| Scaling Altair visualizations with VegaFusion | Jon Mease | 28m |
Claims that VegaFusion automatically moves Vega/Altair data transformations from the browser into the Python kernel so charts scale Altair charts to millions… | ||
| How Data Science Complements the Six Sigma Methodology in Manufacturing | Eloisa Elias Tran | 21m |
Claims that pairing DMAIC with CRISP‑DM maps concrete steps: 35 stratified samples vs millions, use DOE and regression to find causal features, and validate… | ||
| Data Mapping for Data Exploration | Leland McInnes | 38m |
Demonstrates converting embeddings into an interactive 2D map using UMAP, fast nearest-neighbor search against vector databases, and hierarchical clustering… | ||
| How to build stunning Data Science Web applications in Python with Taipy | Gosselin & Jacta | 39m |
A new low‑code Python package claims to build complete data‑science web apps using Taipy Core for scenario and pipeline management and Taipy GUI for… | ||
| It's not just code: managing an open source project | Tracy Teal | 28m |
Claims that most projects run on one or two people do most work and that success depends on explicit choices about goals, stage, governance, licensing, and… | ||
| Publishing Jupyter Notebooks with Quarto | J.J. Allaire | 41m |
Quarto turns Jupyter notebooks into production‑quality websites, books, blogs, presentations, PDFs and Word docs by rendering notebook cells into markdown… | ||
| Building a Marketing Analytics Web App with Python | Leo Anthias | 36m |
Claims a full replacement for a proprietary marketing analytics platform by wiring Shopify data into a Python stack and shipping a dashboard plus downloadable… | ||
| Shiny: Data-centric web applications in Python | Joe Cheng | 44m |
Shiny for Python is a ground-up reimplementation (not an R wrapper) that uses Transparent Reactive Programming to build data apps with Seaborn, pandas/polars… | ||
| Diversity Panel: Allyship is a journey, not a destination | Eloisa Elias Tran | 59m |
A consulting engagement asks for an AI that identifies LGBTQI+ people from protest videos so a government can count and act on them; the request prompts an… | ||
| Experimentation and the gold standard of data champions | Timothy Chan, PhD | 39m |
Claims that companies can run 10,000 experiments simultaneously and get reproducible causal answers, with tooling built around feature flagging and a… | ||
| Because SQL is everywhere but you don't want to use it | Cloud & Forsynth | 32m |
Claim: use Python expressions to build SQL that actually runs on the engine, not fragile string templates. Demonstration uses Ibis to connect to DuckDB and… | ||
| Jupyter AI — Bringing Generative AI to Jupyter | David Qiu | 39m |
Jupyter AI ships as a JupyterLab/IPython extension that actually works: IPython magics and chat panel let users call multiple providers (AI21, Anthropic… | ||
| Managing a search engine for over 600 million openly licensed media records | Madison Swain-Bowden | 45m |
Openverse is a search engine that aggregates — not hosts — openly licensed images and audio, claiming over 600 million records and integrated into WordPress… | ||
| MLOps Deployment Patterns with Delta Lake and MLflow | Mary Grace Moesta | 32m |
Three concrete deployment patterns appear: deploy model, deploy code, hybrid, and the clear claim is that the choice depends on which artifacts move and who… | ||
| Monitoring in the era of Generative AI, LLVMs, and embeddings | Bernease Herman | 32m |
Argues that monitoring LLMs, vision models, and embeddings scales only when raw data is profiled locally and summarized as compact statistics rather than… | ||
| Plant a Touch-Me-Not: Train Models Without Anyone Touching Your Data with Flower | Krishi Sharma | 32m |
Claims use federated learning with Flower to train without moving raw data; evaluation compares PySyft, TensorFlow Federated, and Flower and chooses Flower for… | ||
| Quantifying Uncertainty in Time Series Forecasting | Max Mergenthaler, Fede Garza | 37m |
Claims that conformal prediction produces theoretically calibrated, distribution-free prediction intervals by estimating error distributions via time-series… | ||
| The Importance of Synthetic Data in Data-Centric AI | Fabiana Clemente | 25m |
Claims that synthetic tabular data can match or improve ML outcomes while preserving privacy using ydata-synthetic and CTGAN style models and shows concrete… | ||
| Untangling the complexity of demand forecasting models: building a Market Simulator | Pablo Alfaro | 33m |
A working Market Simulator is built in Python (numpy, pandas and an ensemble approach) that models expected demand as a price‑dependent curve, samples Poisson… | ||
| Combining IPython with Open Source Papermill to enhance your Jupyter Notebook | Pierre Brunelle | 32m |
Shows how papermill and origami plus a genai plugin link IPython to GPT‑class models to generate and execute code, materialize runs into orchestrators like… | ||
| How Apache Sedona is Revolutionizing Geospatial Data Analysis | Jia Yu | 46m |
Apache Sedona extends cluster engines to run distributed spatial SQL and datasets and claims real-world speedups: it extends Apache Spark and Apache Flink… | ||
| Growing the open source quantum ecosystem | Nate Stemen | 40m |
Overview of concrete outcomes and tools: Python-based front ends Cirq and Qiskit frontends power actual code examples, and work targets compilers and… | ||
| Indian Sign Language Recognition(ISLAR) | Akshay Bahadur | 27m |
Claims an AI system that converts regional ISL into text and speech in real time, developed with the Centre for Research and Development of Deaf & Mute and… | ||
| Notebooks as Serverless Functions | Koushik Krishnan | 31m |
Jupyrest claims to turn Jupyter notebooks into HTTP functions by pairing a small config.json (ID, input/output JSON Schema) with a notebook that calls… | ||
| Open Source meets Enterprise: The right way. | Naty Clementi | 28m |
Shows concrete changes: Dask QN reduces memory use ~80%, with a live demo where the old scheduler spills to disk and the new one runs about 6–7x faster on the… | ||
| Python Anytime, Anywhere with Anaconda Notebooks | Sophia Yang | 31m |
Anaconda Notebooks runs a managed Jupyter environment in the cloud with a default Python 3.9 kernel and 300+ preinstalled packages, claims a 7.1 billion… | ||
| Python in Bioinformatics | Trent Hauck | 38m |
Python glues compiled CLI tools into pipelines, running BLAST, Bowtie, Samtools inside containers, uploading results with boto3 and Docker and orchestrating… | ||
| The Python Data Ecosystem: Navigating a fragmented landscape | Umare & Tong | 46m |
Claims a concrete orchestration: a strongly typed Python DSL that registers containerized tasks and workflows, builds images for you, and runs elastic… | ||
| Hidden Gems in MLFlow that Improve Experiment Reproducibility | Krishi Sharma | 38m |
Shows three concrete MLflow tactics that actually reduce debugging time: use mlflow.autolog() to get consistent parameter and metric captures across branches… | ||
| Deep Learning Model Interpretability for Computer Vision based Models | Sumedh Datar | 35m |
Shows concrete methods and outcomes: uses shap and lime and gradient heat maps like grad cam to localize features, runs PCA/t‑SNE on embeddings for clustering… | ||
| Increase productivity and simplify data operations in Python | Tomasz Drabas | 25m |
Claims a concrete workflow: use Ibis to write one set of Python expressions, serialize via Substrait, and export models with ONNX so the same featurization… | ||
| How to incrementally scale existing workflows on Spark, Dask or Ray? | Liu & Wang | 38m |
Shows a concrete path to scale pandas workloads by layering Fugue over Spark Dask or Ray, using a transform wrapper and cost-partitioning so existing Python… | ||
| Nine Rules for Writing Python Extensions in Rust | Carl Kadie | 40m |
Nine concrete rules and examples show how ported genomics C++ extension to Rust (a plink bed file reader) replaces OpenMP with Rayon, uses pyo3 and… | ||
| Scaling data workloads using the best of both worlds: pandas and Spark | Eng & Kwon | 36m |
Pandas API on Spark lets code scale by swapping import pandas for pyspark.pandas and shows near linear speedup as nodes increase in their benchmarks, with API… | ||
| U-Net-style neural networks for feature identification in 1D time-series | Michael Byington | 36m |
U‑Net–style 1D networks identify discrete events (pipe joints) by expanding the convolutional receptive field >10,000 points so each label sees many repeats… | ||
| Particle Swarm Classification for Next-Gen Recommendation Engines | Eugene Ciurana | 30m |
Claims an optimal recommendations engine for consumer banking that yields high-response tailored results from anonymized and incomplete data and actually ships… | ||
| nothing matches | ||