PyData NYC 2022
2022-11-09..2022-11-11
New York, United States
[PyData NYC 2018, PyData NYC 2022, PyData NYC 2023, PyData NYC 2024]
| title | author | length |
|---|---|---|
| How we upstreamed our internal goals to JupyterLab 4 | Diego Torres Quintanilla | 43m |
Two Sigma describes concrete upstream contributions that land in JupyterLab 4: benchmarks show opening large notebooks falls from about four seconds to around… | ||
| Deep learning for time series forecasting and classification in practice | Isaac Godfried | 01h 02m |
Walks through hands‑on examples using Flow Forecast, PyTorch Geometric Temporal, SHAP and ELI5 to forecast retail, healthcare, climate and sales series and to… | ||
| Understanding the News around the World with Web Scraping and NLP at Scale | Jon Wiggins | 35m |
Pipeline scrapes HTML, saves pages to S3, runs Python NLP (spaCy NER plus custom models), links mentions to Wikidata using kensho-processed Wikipedia CSVs, and… | ||
| Supercharge your Python code with Blocks | Jeff Hale | 36m |
Prefect blocks store reusable configuration as Python objects that are create via GUI or Python, expose code snippets for consumption, and persist on a server… | ||
| Practical MLOps- Do we need all the things | Brian Bush | 01h 13m |
Claim: a small subset of MLOps delivers real production value rather than every platform piece. Example: a reproducible pipeline built with Kedro catalogs and… | ||
| Hands-On Computer Vision with PyTorch | Robert Alvarez | 01h 27m |
Walkthrough of PyTorch from tensors and autodiff to CNNs with runnable Colab code and GPU runtime; builds a linear regression and shows it needs ~2,000 steps… | ||
| Introduction to Causal Inference | Roni Kobrosly | 01h 12m |
Covers practical steps: uses Jupyter notebooks to run g-computation on the LaLonde job‑training data and estimates a treatment effect near $17,000; uses the… | ||
| Troubleshooting your Data Workflows with Noteable+Dagster-live debugging of failed jobs | Jamie DeMaria | 01h 25m |
A productionized Noteable notebook hooks into Dagster to become schedulable assets, with the workshop showing how to transition an exploratory notebook into a… | ||
| Distributed Python with Ray-Hands on with the Ray 2.0 APIs for scaling Python Workloads | (unnamed presenter) | 01h 30m |
Hands-on demos use Ray 2.0 Core APIs and Ray AIR to scale Python from laptop to cluster: remote functions as tasks, remote objects as futures, and stateful… | ||
| Serving Pytorch Models in Production | Dagshayani Kamalaharan, Nidhin Pattaniyil | 01h 00m |
A BERT-based product-type classifier is already trained and then distilled and quantized to cut latency and size: F1 jumps from 0.75 to 0.91 versus the old… | ||
| Easily deploy+maintain an open source data science platform on the cloud of your choice | s Guild | 01h 15m |
Claim: Nebari installs on AWS GCP Azure DigitalOcean or on‑prem in about 30–45 minutes and the live demo finishes successfully. The deployment uses… | ||
| Building a Semantic Search Engine | Ravi Nidhin Pattaniyil | 01h 28m |
Shows a working pipeline: index 25,000 images with captions, ingest at ~3,000 docs/sec into BM25 baseline ElasticSearch with ~4 ms retrieval, compute sentence… | ||
| Using Numba Effectively Today | Cheuk Ting Ho | 37m |
Numba is a JIT compiler for NumPy that compiles functions on first use so the initial call is slower but subsequent calls run much faster; use @jit or the… | ||
| Applications in ML Drug Discovery pipelines | Estefania Barreto-Ojeda | 42m |
Claims that Data Version Control (DVC) lets drug‑discovery ML pipelines treat raw data, derived datasets, models and metrics as versioned artifacts, and… | ||
| CATs: Content-Addressable Transformers | Joshua E. Jodesty | 40m |
Claims an open-source framework that uses IPFS CIDs to content-address inputs processes outputs and IaC, orchestrates Spark workloads via Kubernetes (demo on… | ||
| A Graph-based Machine Learning early warning system to detect Ransomware | Iain Campbell | 36m |
Claims a practical early warning system that flags ransomware activity by converting Windows logs into graphs and running graph ML it processes att-29 200k… | ||
| Customizable probabilistic record linkage with Name Match | Melissa McNeill | 45m |
Probabilistic record linkage tool Name Match trains a supervised model using records that include first name last name and DOB plus some records with a ground… | ||
| Holistic MLOps for better science | Isabel Zimmerman | 34m |
Claims that MLOps practices help small teams and then shows an implementation using the open source vetiver framework to version share deploy and monitor… | ||
| 20 ideas to build social capital in the Data Science ecosystem | Lawrence Wilson Gray | 35m |
Twenty concrete moves are listed not platitudes: submit a pull request, volunteer at conferences, start a local Meetup. Specifics include contributing to… | ||
| Improving Your Data Modeling Work Through Open-Source Software | Blancas, Michael | 01h 23m |
Shows concrete tools and outcomes: parameterized notebooks run as a 45‑experiment grid and produce per‑run pickles and reports; a CLI called debuglater… | ||
| Install Python. Quarto Render All the Things | Daniel Chen | 43m |
Shows how to get Python installed and use Quarto (the uploader description: Quarto built on Pandoc) to run live code, convert notebooks and qmd files, and… | ||
| Contextual Multi-Arm Bandit and its applications to digital experiments | Peterson, Qin | 45m |
Claims a production-ready contextual MAB that builds a model from historical randomized data then assigns users to contextual cohorting and runs a vanilla MAB… | ||
| Shiny for Python: Interactive apps and dashboards made easy-ish | Joe Cheng | 44m |
Shiny for Python is announced as an Alpha release that brings Shiny’s decade‑old reactive graph/decorators model to Python, trading Streamlit’s simplicity for… | ||
| I hate writing tests, that's why I use Hypothesis | Cheuk Ting Ho | 42m |
Claims that property-based testing with Hypothesis replaces example thinking, shows concrete tools: given and strategies drive generation, ghostwriter… | ||
| Implementing a Workflow Engine in Python | Sanjay Siddhanti | 44m |
Engineered a Python workflow engine that actually runs production automations for hospitals it serializes inputs with Pydantic, exposes task enqueueing via… | ||
| Simulations in Python: Discrete Event Simulation with SimPy | Lara Kattan | 43m |
Builds an M/M/1 queue in SimPy + simpy_helpers inside a Jupyter notebook, using numpy draws for exponential interarrival and service times (mean service 20)… | ||
| Discover Inspirational Insights in Motivational Sports Speeches Using Speech-to-Text | Tonya Sims | 35m |
Builds a Python pipeline that uploads audio to Python + Deepgram speech-to-text, runs summarization, topic detection, sentiment analysis, profanity filtering… | ||
| Predicting Weather-Caused Rare Events: A Utility Outage Prediction Use Case | Song, Srinivasan | 42m |
Claims a deployed ML pipeline that predicts weather-caused utility outages hourly for 24–48 hour horizons and lets operations mobilize crews 12 hours ahead to… | ||
| ipyvizzu-story- a open-source tool to build create+share animated data stories w/ python in jupyter | visu (team member) | 01h 26m |
Shows that ipyvizzu-story wires Vizzu’s C++/WebAssembly engine to a Python API and a Jupyter UI, using an animate(data, config, style) call to morph charts… | ||
| Bagging to BERT- A tour of applied NLP | Benjamin Batorsky | 01h 26m |
Compares concrete methods on the IMDb 50k sentiment dataset and shows what actually moves the needle: a scikit-learn Count/TF‑IDF + Logistic baseline hits… | ||
| Can we optimize communities | Melissa Weber Mendonça | 53m |
A two‑year Chan Zuckerberg–backed pilot supports NumPy SciPy Matplotlib and pandas with paid contributor‑experience leads and after one year shows concrete… | ||
| Herding Entities- Information Search and Synthesis in the Context of Transaction Data | Martin Jiminsky | 39m |
Ntropy claims a production entity search that enriches noisy transaction text by combining Elasticsearch indexes, milvos vector nearest‑neighbor lookups, and a… | ||
| Scaling Python - Bank Edition | Andrew Fulton, Adam Lewis | 40m |
Claims a working replacement of a legacy single‑threaded C++ engine by a distributed Python pipeline that actually handles 24.5 TB (140,000 parquet files, 6M… | ||
| Model Upgrade Schemes- Considerations for Updating Production Models | Emmanuel Naziga, Munaf A Qazi | 40m |
Describes an implemented CI/CD + continuous monitoring and training pipeline with feature store integration, a random holdout feedback loop and delayed labels… | ||
| Human-Friendly, Production-Ready Data Science Stack with Metaflow & Kubernetes | Goyal | 36m |
Metaflow claims to already power hundreds of ML projects across domains and provides concrete features: automatic execution snapshots to S3 that record code… | ||
| Testing Big Data Applications (Spark, Dask, and Ray) | Han Wang | 39m |
Compares concrete testing techniques and a working tool: Spark Dask and Ray create brittle, cluster‑dependent tests while an abstraction — Fugue abstraction… | ||
| Prompt Engineering ⚙️ - Addressing the sensitivity of Large language models | Wan (possibly Wan) | 38m |
Mantis runs systematic comparisons of different prompts across over 6 models (GPTJ 6B, GPT‑3 175B, Bloom, Jurassic‑1, OPT, Cohere, Goose AI and others) on 5… | ||
| Large Language Models for Real-World Applications - A Gentle Intro | Hemant Jain | 51m |
Clear, concrete rundown: shows how GPT3 and Dall‑E class models are trained and deployed, demonstrates embeddings and fine‑tuning for classifiers and semantic… | ||
| Apache Beam on Dask- Portable, Scalable, Scientific Python | Alex Merose | 42m |
Claims a working Beam-on-Dask translator that visits Beam pipelines and emits Dask high‑level graphs, with the change landed upstream in Apache Beam and a Dask… | ||
| Scalable Feature Engineering with Hamilton | Elijah ben Izzy | 41m |
An open source framework called Hamilton converts each feature into a single Python function, decouples execution via a driver, and adds runtime data… | ||
| pandas at a Crossroads, the Past, Present, and Future | Jeff Reback | 37m |
pandas sits at a concrete crossroads: maintainers choose between a risky full rewrite, steady incremental improvement, or embracing an API layer (Ibis) that… | ||
| ML Latency No More- Useful Patterns to Reduce ML Prediction Latency to Sub X ms | (unnamed presenter) | 42m |
Argues that latency is a business constraint not a nicety: treat ~100 ms as the cutoff (examples use <10 ms timeouts), measure optimizing vs satisficing… | ||
| Coldstart- A library for automatic data curation + feature engineering | Piero Ferrante | 32m |
Coldstart is a Python package that automates data curation and feature engineering by running concurrent templated SQL against dialects like BigQuery via… | ||
| I'm from the government and I'm here to help | Chelle Gentemann | 39m |
NASA launches a program that changes requirements and builds cloud tooling so datasets and mission software are open with no embargo and egress‑free access… | ||
| Causal machine learning for a smart paywall at The New York Times | Rohit Supekar | 40m |
Claims a deployed prescriptive system called Dynamic Meter that uses causal meta‑learners trained on randomized control trials to pick per‑user meter limits… | ||
| A Guide to Data Science as a Creative Discipline | Ilinca Barsan | 32m |
Argues that data science is inherently creative and shows concrete work not just theory: uses DALL·E 2 and GPT-3 as context, trains small neural nets on… | ||
| Ibis- Expressive analytics in Python at any scale | Charles Cloud, Gil Forsyth | 44m |
Ibis is a pure Python library that builds type-checked expressions and compiles them to SQL or native backends (duckdb, sqlite, postgres, spark, clickhouse… | ||
| Level up your viz skills- from Matplotlib to HoloViz | Sophia Yang | 30m |
Starts from Matplotlib syntax and shows concrete HoloViz tools: hvplot over Bokeh, Panel for dashboards, and Datashader for big data. Demonstrations include 11… | ||
| Deploying Dask | Matthew Rocklin | 41m |
Deploying Dask is messy but solvable: an Optuna + XGBoost demo requires a MySQL replacement, spins up 20 machines on AWS, and surfaces a profile where 97%… | ||
| JAX for Bayes | Colin Carroll | 41m |
Implements a regression-for-classification on a small golf dataset across five model libraries and runs inference with four inference libraries (JAX-backed… | ||
| Expressive and fast dataframes in Python with polars | Juan Luis | 38m |
Polars uses a Rust core over Arrow and shows a lazy expression planner that fuses operations and parallelizes independent aggregates, letting a cloud demo scan… | ||
| Zeno Does Data Science- The Paradoxical Quest for Reproducibility | Kjell Wooding | 40m |
Describes concrete fixes: adopt one virtual environment per repo, use git for readable text artifacts only, separate a “what I want” env YAML from a locked… | ||
| Chasing the Overton Window | Allen Downey | 41m |
General Social Survey data (≈70,000 interviews) map beliefs to a 15-question conservatism scale using item response theory and Bayesian imputation; analysis… | ||
| How to build a serverless electricity price prediction service in Python with Hopsworks + Streamlit | Fabio Buso | 31m |
Builds a serverless electricity price prediction service in Python and actually wires it end to end: Pandas feature engineering feeds a Hopsworks feature store… | ||
| Git for Data- Data Versioning for Reproducible Data Science with Dolt | Zach Musgrave | 43m |
Claims that Dolt is “Git for data” get tested: a live demo shows the CLI using branch and merge SQL tables, push/pull to DoltHub, fork/clone and open DoltHub… | ||
| How to make friends and generate impact | Lauren Oldja | 38m |
NumFOCUS runs fiscal sponsorship for ~100 projects (numpy scipy pandas PyMC Jupyter Julia), provides small development grants and a developer advocate on… | ||
| Parallelism in Numerical Python Libraries | Thomas J Fan | 41m |
Shows which libraries actually parallelize (NumPy linear algebra vs elementwise ops), links NumPy to BLAS/MKL/OpenBLAS, and demonstrates threadpoolctl and… | ||
| Gentle introduction to scaling up ML service with Kubernetes + Mlflow | Kei Nemoto | 39m |
Deploys a Flask + Gunicorn model inside Docker, moves runtime orchestration to Kubernetes using k3s, and adds serverless scaling with Knative so pods scale… | ||
| Using Interconnected ML Models to Tackle Retail Challenges | Kshetrajna Raghavan | 42m |
Shopify builds a multitask hierarchical classifier (5500+ taxonomy nodes) that uses multilingual BERT for text and MobileNetV2 for images, caches embeddings to… | ||
| Closing Notes | Lauren alja | 06m |
PyData NYC reports concrete outcomes: 675 attendees, 84 speakers across four keynotes and 79 talks/tutorials, 39 diversity scholars, 32 sprint participants and… | ||
| Lightning Talks | 01h 02m | |
Five-minute snapshots run through concrete claims and tools: a Trump tweet analysis shows over 20,000 iPhone tweets and 73 tweets between 3:00–3:15 AM with an… | ||
| From Science to Open Source and Back Again | Thomas Caswell | 40m |
A practiced path between lab and open source shows concrete results: the Bluesky Project and stewardship of matplotlib produce reusable stacks rather than… | ||
| PyScript & Data Science- PyData stack on the Browser | Fabio Pliger | 42m |
PyScript claims to run CPython via Pyodide/WebAssembly and actually loads the PyData stack in the browser so code using pandas, NumPy, Seaborn, bokeh, panel… | ||
| Why do I need to know Python- I'm a pandas user | James Powell | 43m |
Argues that concrete Python knowledge pays off: using contextlib.closing fixes openpyxl file locks, pathlib + regex picks CSVs reliably, and itertools.product… | ||
| NixtlaVerse, bridging the gap between statistics and deep learning for time series | (unnamed presenter) | 33m |
NixtlaVerse delivers concrete tooling and numbers: StatsForecast (PyTorch + NumBa) runs AutoARIMA/ETS at scale — claim is 10 million series in under 43 minutes… | ||
| Fairness for Scikit-Learn Pipelines with Lale | Martin Hirzel | 41m |
Experiments use the credit G dataset (1,000 rows, ~20 features) with protected attributes age and personal status and report concrete numbers: dummy classifier… | ||
| Building highload ML powered service | Arseny Kravchenko, Thymo ter Doest | 40m |
Asserts Python backends can run high‑load ML in production and demonstrates it with a transaction‑enrichment pipeline handling millions of samples, supporting… | ||
| nothing matches | ||