PyData Los Angeles 2018
who it's for
data scientists, python developers, machine learning engineers, data engineers, data visualization developers, research scientists, analytics managers
topics
tensor decompositions for neural nets; Ray and Modin for pandas scaling; JupyterLab evolution; RAPIDS GPU dataframes; CatBoost gradient boosting; variational autoencoders for anomaly detection; interrupted time series analysis; mixed effects random forests; Pyodide/Iodide in the browser; Bokeh dashboard development; text pipelines with spaCy and gensim; Docker for reproducible data science
| title | author | length |
|---|---|---|
| Building a Player-Focused Data Team | Andrea Trevino | 35m |
Describes concrete systems: player-level models for matchmaking and personalization, a champion recommendation engine, positional-play detection and… | ||
| Tensorly: A Flexible Python Framework for Machine Learning | Anima Anandkumar | 45m |
Tensorly is a Python front end that wires to NumPy SciPy or PyTorch TensorFlow and MXNet backends and provides Jupyter notebooks that define tensorized neural… | ||
| Building an Open Platform for Sustaining Data Science Innovation | Andy Terrel | 58m |
NumFOCUS claims an operational platform: 26 sponsored and 24 affiliated projects producing roughly 50 packages that combine into a reusable stack. Example… | ||
| What is a Data Desk? | Ben Welsh | 51m |
Shows concrete claims and repeatable tools: how Python + pandas in Jupyter notebooks and requests + BeautifulSoup scrape county sites, how GeoPandas spatial… | ||
| The New Science of Cause and Effect | Judea Pearl | 01h 06m |
Argues that data science needs an actual causal engine: a graph‑based Structural Causal Model plus surgical interventions and do‑calculus that let one answer… | ||
| Learning to Scale Data Science, Machine Learning, and Pandas with Ray and Modin | Devin Petersohn | 29m |
A unified runtime called Ray actually runs Ray remote tasks and actors with a bottom-up scheduler, per-node object store, Redis-backed global control store… | ||
| When Rotten Tomatoes Isn’t Enough: Analyzing Twitter Movie Reviews | Amanda Moran | 01h 22m |
Pipeline uses Apache Cassandra (DataStax Enterprise Analytics) as the store, Apache Spark to tokenize and remove stop words, Python + Jupyter notebooks to glue… | ||
| Introduction to Data Manipulation and Visualizations in R | David Sung | 01h 25m |
Hands-on RStudio session that reads a Kaggle CSV with 25 variables installs and attaches packages, and demonstrates actual code: reading CSV, setting working… | ||
| Network Science, Game of Thrones and US Airports | Mridul Seth | 01h 25m |
Hands‑on tutorial that loads the repo https://github.com/MridulS/pydata-networkx and uses NetworkX, Pandas, Matplotlib to build graphs from the Game of Thrones… | ||
| Tips, Tricks and Topics in Text Analysis | Bhargav Srinivasa Desikan | 01h 27m |
Hands‑on pipeline that loads ~300 Australian news articles and a 20‑newsgroups subset, uses spaCy for lemmatization, POS and NER, builds bigrams, creates a… | ||
| Docker for Data Science: Reproducibility and Deployment | Hareem Naveed | 01h 02m |
Live demos show concrete wins and gotchas: reproducible environments replace juggling six virtualenvs, an docker run hello-world and docker build -t dev… | ||
| Applying Statistical Modeling & Machine Learning to Perform Time-Series Forecasting | Tamara Louie | 01h 26m |
Walks through a reproducible notebook that ingests a 23 MB CSV (1.2M rows), aggregates to daily counts (408 listings on the last date), and shows that raw… | ||
| Testing with Pytest for Data Science | Ravin Kumar | 01h 28m |
Step‑by‑step use of pytest shows actual code, not platitudes: reproduce the Python2 vs Python3 split_bill failure and see tests catch it; run tests locally and… | ||
| CatBoost - The New Generation of Gradient Boosting | Vasily Ershov | 01h 27m |
CatBoost claims to outperforms existing implementations on public benchmarks and ships an open source library with CPU and GPU engines, native handling of… | ||
| An Introduction to Julia | Jane Herriman | 01h 38m |
Quick, concrete signals: Julia claims “looks like Python, feels like Lisp, runs like C” and shows it — benchmark: 10 million elements → handwritten Julia ~11… | ||
| Making Computation Easier with Cool Numpy Tricks | Kirit Thadaka | 01h 31m |
Demo shows concrete speedups and working code: creating a 100,000-element array takes ~1.04 ms vs ~5.2 ms in plain Python, summing takes ~462 µs vs ~19 ms, and… | ||
| Time, Interrupted: Measuring Intervention Effects with Interrupted Time-Series Analysis | Ben Cohen | 45m |
Explains how interrupted time‑series turns the counterfactual into a prediction problem, fitting Bayesian time‑series models (uses Prophet (Stan backend)) to… | ||
| Serverless for Data Scientists | Mike Lee Williams | 46m |
Claim: serverless removes OS management so data science runs as short-lived functions. Demonstrations use AWS Lambda with deployment via Zappa and parallel… | ||
| Data Mining JIRA Tickets to Gain Insights into Organizational Behavior | Wendy Grus | 33m |
Claims that built a Flask app that mined JIRA for Glyph links using the Python jira library and simple visuals. It extracts issue fields, counts issue types… | ||
| 1D Convolutional Neural Networks for Time Series Modeling | Nathan Janos, Jeff Roach | 36m |
Claims that learned 1D convolutional filters plus a small context network beat the existing TBATS baseline by about MASE 0.85–0.86 vs TBATS 0.90 on… | ||
| Developing Dashboard Applications Using Bokeh | Bryan Van de Ven | 45m |
Release-ready Bokeh 1.0 details and concrete patterns for interactive dashboards: streaming updates use ColumnDataSource.stream and .patch, deployment options… | ||
| Why You Should Do Text Analysis in Python (Even if You Don't Want to) | Bhargav Srinivasa Desikan | 49m |
Exports of group chats from WhatsApp exports get parsed year by year to count who starts conversations who uses the most emojis and to run topic models that… | ||
| (Deep) Learn You a Neural Net for Great Good! | Michael Stewart | 43m |
Live code builds a fully-connected neural net from scratch using numpy in a RISE slideshow, trains it first with explicit Python for-loops and then wraps it as… | ||
| JupyterLab: The Evolution of the Jupyter Notebook | Ian Rose, Grant Nestor | 40m |
JupyterLab positions itself as a next‑generation Jupyter Notebook UI with a modular extension system, implemented in TypeScript using NPM/Yarn, Webpack and… | ||
| Accelerating Data Science with RAPIDS | Mike Wendt | 41m |
RAPIDS keeps data on the GPU to avoid CPU copy/convert penalties, demonstrating loading ~150–200 GB of CSV (seven years of loan data) into GPU memory… | ||
| Hot Water Leak Detection Using Variational Autoencoder Model | Jay Kim | 38m |
SoCalGas needs to distinguish hot water leaks from other abnormal gas usage and applies a Variational Autoencoder to screen hourly meter profiles rather than… | ||
| Iodide and Pyodide: Bringing Data Science... | Michael Droettboom, Brendan Colloran, Hamilton Ulmer | 31m |
Iodide and Pyodide claim to run Python NumPy pandas matplotlib in the browser by compiling CPython and C libraries to WebAssembly, loading interpreters and… | ||
| Extracting Structured Data from Legal Documents | Zack Witten | 44m |
Uses a conditional random field with hand engineered features (spaCy POS tags, baby name lists, number and month detectors) on OCRed PDFs via Google Cloud OCR… | ||
| Attacking Clustered Data with a Mixed Effects Random Forests Model in Python | Sourav Dey | 37m |
Shows a Python implementation of MERF as a pip-installable MERF with a scikit-learn style estimator and an EM loop that alternates training a random forest and… | ||
| I Just Inherited 50,000 Lines of Code! What Now? — A Practical Guide | James Powell | 59m |
Claims concrete tactics for inheriting 50,000 lines: find runnable entry points, run c-loc and tests, expect many tests to fail, treat ~20% of code as… | ||
| Detecting Signed and Unsigned Documents with Deep Learning - Beyond Transfer... | Jordan Bramble | 32m |
A pipeline uses transfer learning on a pretrained YOLO object detector plus Google Cloud OCR to localize signed and unsigned signature boxes, trained on 40,000… | ||
| Measuring Model Fairness | Stephen Hoover | 31m |
Case study analyzes anonymized Civis Analytics client data (~130k rows, ~10k labeled African American, ~120k white) using a logistic regression classifier and… | ||
| StaticFrame: An Immutable Alternative to Pandas | Christopher Ariza | 36m |
StaticFrame is an explicit Python 3 wrapper around numpy that enforces immutable numpy arrays and immutable Series/Frame objects, offers a frame_geo grow-only… | ||
| Python at City Scale | Hunter Owens | 43m |
Short technical deep dive showing concrete results: Predicting Displacement using Scikit Learn and Pandas yields a model that flags Ellis Act buildings for… | ||
| Customer Lifetime Value: Models, Metrics and a Multitude of Uses | Brian Bloniarz | 36m |
Claims that probabilistic CLV models recover missing future purchases from sparse histories using real credit-card transactions and open tools: modified BG/NBD… | ||
| What's the Science in Data Science? | Skipper Seabold | 46m |
Argues that credibility comes from concrete design not buzzwords and shows which methods actually rescue observational work: instrumental variables, matching… | ||
| Building a Visualizer in Yellowbrick | Nathan Danielsen | 38m |
Yellowbrick extends the scikit-learn API to produce matplotlib plots glued with numpy and packaged as a class with init fit draw poof lifecycle. Workflow shows… | ||
| Deploy and Use a Multiframework Distributed Deep Learning Platform on... | Animesh Singh, Tommy Li | 29m |
Fabric for Deep Learning (FfDL) is a Kubernetes microservice platform that runs training in containerized learners, stores data and models in object storage… | ||
| Using Data to Get Your Next Raise | Michelle Brenner | 35m |
A practical method for tracking success statements and turning them into ROI: collect duration in seconds, start date, recurrence and optional employee count… | ||
| Using Simpson’s Paradox to Discover Interesting Patterns in... | Nazanin Alipourfard, Peter Fennell | 31m |
Package claims it detects Simpson’s paradox automatically in large behavioral datasets by fitting a global logistic/linear trend then disaggregating on… | ||
| Extending Pandas with Custom Types | Will Ayd | 28m |
Pandas v0.23 introduces an extension interface that implements integer NA with mask by storing a value array plus a boolean mask so integers keep precision and… | ||
| Train, Evaluate, Repeat: Building a Credit Card Fraud Detection System | Leela Senthil Nathan | 47m |
Describes concrete solutions for Stripe’s fraud stack: chooses merchant‑agnostic labels like chargebacks and issuer fraud records, experiments to set a 10%… | ||
| nothing matches | ||