PyData New York City 2019
2019-11-04..2019-11-06
New York City, United States
| title | author | length |
|---|---|---|
| A Few Good Public Servants: How Great Analysis Inspires Action | Kelly Jin | 40m |
Stories report concrete outcomes: a Building Intelligence System for firefighters, a mayor's dashboard that executives use daily, deployment of an analytic… | ||
| Data science at The New York Times | Chris Wiggins, Anne Bauer | 38m |
The New York Times applies machine learning end to end: Python and TensorFlow for training, Go for real‑time ranking, BigQuery and Google Cloud Storage for… | ||
| Stars, Planets, and Python | Sara Seager | 45m |
MIT-led TESS launches April 2018 and images wide sky strips, stores ~200 GB onboard and downlinks sectors that observe 20,000 targets per sector at two‑minute… | ||
| Introduction to pandas | Marc Garcia, Jeff Reback, Tom Augspurger | 01h 21m |
This tutorial is a hands on introduction to pandas. pandas is a fast, powerful, flexible and easy to use data analysis and manipulation tool, built on top of… | ||
| Advanced Software Testing for Data Scientists | Raoul-Gabriel Urma | 01h 30m |
Uses a Kickstarter dataset and a scikit-learn pipeline of transformers (category extraction, currency normalization to USD) and shows concrete bugs found by… | ||
| Role playing Annotation workshop | Sumowska, Desikan, Lawrence Warner, Konstantinovskiy | 01h 31m |
Role-play workshop has teams of three plus a Game Master and runs about an hour and 20 minutes, focused on a concrete task: classify homepage into business… | ||
| Neural Networks for Natural Language Processing | Matti Lyra | 01h 32m |
Covers concrete experiments: German dataset with 300‑dim fast text embeddings fed into feed‑forward, LSTM and transformer models implemented in PyTorch… | ||
| Introduction to Language Modeling | Aditi Khullar, Eugene Tang | 01h 24m |
Two-part tutorial that starts with n‑gram language model math then walks through a Jupyter hands‑on notebook using the Brown corpus and NLTK to build unigram/… | ||
| Introduction to NLP | Mariel Frank | 01h 14m |
Hands-on notebook work shows actual code and outcomes: a GitHub repo and Binder Jupyter notebooks run regex, nltk tokenization, Porter stemming and WordNet… | ||
| Bayesian Inference for Fun and Profit | Mitzi Morris | 01h 20m |
Worked examples use Major League baseball (first 45 at‑bats) and Women's World Cup match data (48 matches) to compare full Bayesian posteriors against… | ||
| New Trends in Estimation and Inference | Cameron Davidson-Pilon | 01h 26m |
Jupyter notebooks fit Weibull and piecewise-hazard models with scipy.optimize.minimize and automatic gradients from Autograd to avoid hand derivatives, compute… | ||
| A/B Testing with SciPy | Hillary Green-Lerman, Michoel Snow | 01h 18m |
Concrete recipe for running and evaluating A/B tests with SciPy: use scipy.stats.chi2_contingency or proportion z‑tests for categorical metrics, run the… | ||
| An Introduction to Probability and Statistics | Will Kurt | 01h 30m |
Covers a hands‑on path from basic counting to Bayesian parameter estimation and generalized linear models, using Python 3 notebooks and concrete code: a “magic… | ||
| From Raw Recruit Scripts to Perfect Python | Stanley van der Merwe, Petr Wolf | 01h 33m |
Converts a messy Jupyter notebook into modular, testable Python: replaces ad‑hoc CSV parsing with intake catalogs, adds pytest unit tests (including doctest… | ||
| Visualizing the 2019 Measles Outbreak in NYC (with Python) | Carlos Afonso | 01h 24m |
Creates concrete visual evidence: pulls NYC Health Department tables and reproduces counts (1,250 US cases, 649 in NYC), cleans CSVs, and builds a… | ||
| A Primer on Gaussian Processes for Regression Analysis | Chris Fonnesbeck | 01h 30m |
Hands-on notebooks show how to specify, fit and validate Gaussian process regressions in Python using PyMC3 with gradient MCMC (NUTS/HMC) and occasional MAP… | ||
| High-Performance Data Science at Scale with RAPIDS, Dask, and GPUs | Keith Kraus | 01h 22m |
A concrete claim: RAPIDS + Dask + GPUs move entire ETL, ML and graph pipelines onto GPU memory to avoid copies and give huge real results. Tools and bridges… | ||
| Hacking the Data Science Challenge | Michoel Snow, Hillary Green-Lerman | 01h 28m |
Claims a repeatable, time-boxed recipe for finishing a data‑science take‑home by automating the dull bits and delivering a story, not just plots. Shows a Colab… | ||
| The Inspection Paradox is Everywhere | Allen Downey | 32m |
Explains the inspection paradox with concrete data and code: a live Colab notebook and one-question survey, class-size data from Purdue, MBTA train intervals… | ||
| Data-centric exploration using intake, dask, hvplot, and more | Julia Signell | 40m |
Shows an end-to-end, reproducible workflow using intake catalogs, xarray and dask to read parquet and Landsat tiles, datashader/HVPlot to rasterize and compute… | ||
| Colorism in High Fashion (featuring: K-Means Clustering) | Malaika Handa | 33m |
Transforms twenty years of Vogue covers into ~250 per-cover datasets, fits a separate k-means clustering model per image (about 250 models), extracts average… | ||
| Production Code in Data Science Consulting | Akos Furton | 39m |
BCG Gamma converts POCs into production systems: built five interacting tools (including Pathfinder for plane allocation and Sentry for day‑of reoptimization)… | ||
| Improve the efficiency of your Big Data.. | Francesc Alted, Christian Steiner | 43m |
Claims that operating on compressed arrays yields real speed and memory wins: a 22 GB Central‑Europe precipitation dataset on an 8‑vCPU VM shows about 2×… | ||
| Building Software and Communities With Peer Review | Noam Ross | 43m |
Claims concrete outcomes: four years of rOpenSci peer review drives adoption of standards, finds real bugs, and scaffolds maintainers. Process combines ~250… | ||
| Painting a Picture of Public Data | Kamal Abdelrahman | 35m |
Uses public emergency-department data to show concrete changes: average wait times rose from 46.5 to 58.1 minutes (≈25%) and urban ~62.4 vs rural ~40 minutes… | ||
| The Echo-Chamber of Your Social Media Feed | Tamar Yastrab | 40m |
Claim: recommendation pipelines on Google YouTube and Facebook concentrate attention and can radicalize rhetoric by surfacing what clicks not what’s… | ||
| Reproducibility in ML Systems: A Netflix Original | Ferras Hamad | 29m |
Claims that reproducibility is a first-class ML-infrastructure goal and shows an actual system that enforces it: per-run code packaging to S3 with versioned… | ||
| The Secret Life of Python | Steve Dower | 33m |
Python 3.8 adds audit hooks that surface module loads sockets file access and compile events so security tooling sees what Python actually does not just that… | ||
| Building a maintainable plotting library-PyData NYC | Colin Carroll, Hannah Aizenman, Thomas Caswell | 38m |
Concrete how-to for making reusable plotting code: use matplotlib rcParams style files for lab-wide defaults, wrap plotting into functions that accept an ax=… | ||
| tf-explain: Interpretability for Tensorflow 2.0 | Raphaël Meudec | 30m |
Open-source tf-explain implements interpretability tools for TensorFlow 2.0, ships seven methods including occlusion sensitivity and Grad-CAM, and saves PNG… | ||
| Clean Machine Learning Code: Practical Software Engineering... | Moussa Taifi | 44m |
Claims that applying Clean Code and Clean Architecture to ML cuts technical debt and keeps change costs low backed by concrete examples: an insurance-premium… | ||
| Every ML Model Deserves To Be A Full Micro-service | Romain Cledat | 35m |
Claim: every model ships as a real microservice with tracing, versioning and promotion. Implementation uses meta flow to store artifacts in S3, a lightweight… | ||
| Deep Dive into scikit-learn's HistGradientBoosting Classifier.. | Thomas J Fan | 32m |
scikit-learn implements histogram GBDT that claims comparable accuracy and training time to XGBoost CatBoost and LightGBM on large problems by using histogram… | ||
| Sloth & ENVy | James Powell | 19m |
Explains how the process entry from the system V ABI threads a raw envp of null‑terminated strings through libc’s start_main into Python’s posix layer, how… | ||
| Free Your Esoteric Data Using Apache Arrow and Python | Maciej Wojton | 33m |
Shows concrete pipelines and numbers: reading CSV with Apache Arrow drops measured latency from 23 ms → 8 ms and JSON from 144 ms → 15 ms, converting strings… | ||
| Implementing Lightweight Random Indexing for Polylingual Text Classification -PyData NYC | Ian Whalen | 23m |
Implements lightweight random indexing to merge multilingual TF‑IDF corpora into one training matrix by assigning each token a sparse random index vector with… | ||
| Type-Driven Automated Learning with Lale | Martin Hirzel | 40m |
Lale is an open-source Python library that claims to make semi-automated data science practical by letting users search over pipelines in a few lines of code… | ||
| Discover your latent food graph with this 1... | Alex Egg, Emily A Ray, Parin Choganwala | 39m |
Claims an automated latent food knowledge graph built from behavioral and content data and shows it works. Uses Representation Learning with query2vec… | ||
| Dealing With Imbalanced Classes in Machine Learning | Aditya Lahiri | 31m |
Skewed datasets are not uncommon. And they are tough to handle. Usual classification models and techniques often fail miserably when presented with such a… | ||
| Julia for Pythonistas | Kelly Shen | 32m |
Claims concrete speedups and working projects: Python code translates with little change and Julia runs numeric routines far faster (example sum: Julia ~10 ms… | ||
| Zarr vs. HDF5 | Joe Jevnik | 52m |
Direct comparison of actual behavior and extensibility: Zarr says 2015, pure Python, codecs via numcodecs (Blosc, JSON, msgpack), and multi‑threaded reads with… | ||
| Geo Experiments and Causal Impact in Incrementality Testing | Jessica Tyler | 35m |
Uses geo experiments on DMAs and a Python implementation of CausalImpact to estimate incremental conversions and incremental cost per conversion. Builds… | ||
| Simplified Data Quality Monitoring of Dynamic Longitudinal.. | Jacqueline Gutman | 34m |
A functional library of modular checks monitors dynamic longitudinal oncology data for over 2 million patients across ~20 cancer-specific datasets that are… | ||
| Using Graph Nets (GNs) to predict molecular properties | Chaya D Stern, Yuanqing Wang | 23m |
Graph Nets are applied to molecular graphs and claim to approximate per-atom, per-bond and per-molecule quantum properties with errors within QM uncertainty… | ||
| Effective Python and R collaboration | Daniel Rodriguez | 35m |
Claims concrete interoperability: reticulate runs a Python interpreter inside R and translates dataframes and lists so an R Markdown file can run pandas and… | ||
| Conda-press, or Reinventing the Wheel | Anthony Scopatz | 39m |
Conda-press claims to create wheels from conda packages by walking conda-forge artifacts, producing skinny or fat wheels, and even merging many wheels into… | ||
| To comment or not | Veronica Hanus | 41m |
Argues that comments are practical signals not philosophy citing a messy but working climbing wall scraper built with BeautifulSoup and Stack Overflow snippets… | ||
| Working with Maps: Extracting Features for Traffic Crash... | Jenny Turner-Trauring | 37m |
Feature engineering pipeline extracts city road networks with OpenStreetMap via osmnx, normalizes projections with pyproj and geopandas, indexes geometry with… | ||
| Bringing mental health data to doctors | Bill Lynch | 40m |
A product team turns questionnaires, Apple Health steps and heart rate, and in-app behavior into a single risk score from questionnaires + Apple Health that… | ||
| Quantifying uncertainty in machine learning models | Samuel Rochette | 33m |
Random forest code uses fully expanded trees and keeps leaf training targets to build conditional distributions and 90% prediction intervals contain 91 of 100… | ||
| Is Spark still relevant? Multi-node CPU and single-node GPU workloads.. | Eric Dill | 34m |
Compares Spark Dask and RAPIDS across exploration modeling prediction and dashboarding and concludes Spark remains relevant because enterprises prioritize… | ||
| Scalable Machine Learning with Dask | Tom Augspurger | 38m |
Dask scales NumPy, pandas and scikit-learn onto a cluster so code stays familiar while data and compute move out of core; examples include a cluster of 100… | ||
| Pandas vs Koalas: The Ultimate Showdown! | Amanda Moran | 34m |
Claim: Koalas implements the pandas API on top of PySpark so pandas code mostly runs on Spark with minimal edits. Demo loads a full dataset of 26 million rows… | ||
| Time series for scikit-learn people | Ethan Rosenthal | 34m |
Frames forecasting as supervised ML and ships a scikit-learn compatible skits library with autoregressive, differencing and horizon transformers so you can… | ||
| Semantic modeling of data science code | Evan Patterson | 35m |
Programming languages and libraries are proliferating in the data science community. In an effort to reduce communication barriers and enable automation and… | ||
| Build an AI-powered Pet Detector in Visual Studio Code | Katherine Kampf | 31m |
An AI pet detector gets built inside Visual Studio Code using TensorFlow, Jupyter support and Azure Machine Learning; training uses MobileNet transfer learning… | ||
| Generating realistic, differentially private data sets using GANs | Joshua Falk | 36m |
Generates synthetic individual-level data with differential privacy by adding noise to gradients during GAN training and releases a PyTorch implementation… | ||
| Same API, Different Execution | Saul Shanabrook | 26m |
Claims a small Python-first framework called meta DSL records expressions and rewrites them so the same array API runs with different executors. Demonstrations… | ||
| Propensity Score Matching: A Non-experimental Approach to Causal... | Michael Johns | 34m |
Explains a reproducible recipe: build a propensity model (usually logistic regression), score every unit, match exposed to unexposed using a 0.25 SD caliper on… | ||
| Cleaning, optimizing and windowing pandas with numba | Diego Torres Quintanilla | 39m |
Claims a concrete refactor: separate windowing into indexer, aggregator, kernel so rolling, expanding and ewm reuse kernels and share code; implements a… | ||
| Small Big Data: using NumPy and Pandas when your data... | Itamar Turner Trauring | 34m |
Concrete tactics and numbers: shows how NumPy and Pandas handle data that won’t fit in RAM using compression, chunking, indexing and cheap extra memory, with… | ||
| A Crash Course in Applied Linear Algebra | Patrick Landreman | 35m |
Shows concrete claims and runnable tools: y = Ax decomposition as a unifying model, practical demos in NumPy, and when low matrix rank yields compression and… | ||
| Managing Stakeholders: The Key to Successful Data Science for Business | Lauren Oldja | 33m |
Claims that implementation is a people problem and not just technical get operationalized: use multi-level stakeholder analysis, prioritize the KPI owner as… | ||
| Spark Backend for Ibis: Seamless Transition Between Pandas... | Li Jin, Hyonjee Joo | 37m |
A PySpark backend for Ibis converts Ibis expression trees into native PySpark DataFrame operations so the same Python Ibis expressions run on both pandas and… | ||
| Launching a new warehouse with SimPy at Rent the Runway | Meghan Heintz | 33m |
Rent the Runway uses SimPy discrete event simulator to model launching a second warehouse in Texas, encoding generators for customer orders, resources for SKU… | ||
| The physics of deep learning using tensor networks | Marianne Hoogeveen | 34m |
Argues that convolutional networks map to tensor networks and that convolutional arithmetic circuits use linear activations with product pooling, not ReLU/max… | ||
| What we learned by running a large custom Bayesian... | Jens Fredrik Skogstrom | 35m |
Kolonial.no builds a custom Bayesian demand-forecasting pipeline in PyMC3 that models customer habit segments and transition-based state space dynamics… | ||
| A How-to guide for migrating legacy data applications | Marius van Niekerk, Rohit Kapur | 38m |
A team takes ownership of a decade-old mission‑critical data pipeline and modernizes it by automating large-scale edits across tens of thousands of files… | ||
| Should I develop my own DS library? Maybe. | Piero Ferrante | 36m |
Aetna decides to build an internal library called PI Connect that wires conda setup, Jupyter, multiple back ends, and Hive/Spark defaults into a simple API; it… | ||
| nothing matches | ||