PyData Los Angeles 2019
who it's for
machine learning engineers, data scientists, data engineers, python developers, research scientists, mlops engineers, data analysts, visualization engineers
topics
PyTorch computer vision workflows; H2O AutoML model stacks; deep Q-learning for games; Kedro and MLflow pipelines; HoloViz interactive dashboards in JupyterLab; web scraping with BeautifulSoup and Yelp API; datatable for large in-memory frames; serverless apps with Chalice and Terraform; RAPIDS GPU data science; hierarchical indices for multi-dimensional data; named entity recognition with NERDS; embedding layers for high-cardinality categoricals
| title | author | length |
|---|---|---|
| Keynote – Milana Lewis | Milana Lewis | 44m |
Built a distribution and payout platform for musicians that launches in 2015 and becomes a partner to Spotify Apple Google Amazon, aggregates revenue from 25… | ||
| Keynote – Kyle Polich | Kyle Polich | 01h 00m |
Live demo builds a serverless shoutouts board with Chalice exposing endpoints, uses chalice deploy to publish, provisions infrastructure with Terraform and… | ||
| Keynote – Carol Willing | Carol Willing | 37m |
PyData/NumFOCUS context plus concrete takeaways: a global weather agency runs 255 petabytes of data adding 250 terabytes a day and 400 billion fields, enabling… | ||
| Keynote – Sameer Singh | Sameer Singh | 48m |
Neural nets with millions to billions of parameters achieve near‑human accuracy yet make brittle, counterintuitive errors: a 94% classifier still learns a snow… | ||
| Computer Vision with PyTorch | Daniel J. Brooks | 01h 07m |
Covers concrete workflows using PyTorch and torchvision to run pretrained models like Mask R-CNN on COCO images, producing bounding boxes with confidence… | ||
| Introduction to H20 AutoML with Python | Franklin Velasquez | 01h 18m |
H2O AutoML runs in Python and Flow, trains ensembles (GLM GBM XGBoost DRF XRT deep nets) on a Freddie Mac loan set, handles imputation and encoding, and… | ||
| Analyzing Genetic Networks Using Neural Networks | Manu Flores | 01h 19m |
Claims that machine learning can recover gene regulatory wiring and shows a runnable pipeline that actually predicts the PurR regulon (31 known targets) from… | ||
| Reinforcement Learning- Pac-Man | Manu Gopinathan, Malte Loller-Andersen | 01h 31m |
Code and results: a working Pac-Man agent is implemented with a Q-learning on tiny map that reaches playable behavior after ~250 episodes and a Deep Q-network… | ||
| Kedro + MLflow - Reproducible and Versioned data pipelines at scale | Tom Goldenberg | 49m |
A concrete demo shows a Kedro project bootstraps a pipeline with a declarative data catalog, node-based pipelines and an interactive visualization, then adds… | ||
| Experimental Machine Learning with Holoviz and PyTorch in Jupeyterlab | Hayley Song | 01h 13m |
Live demo of an experimental workflow for multiclass road segmentation from SpaceNet tiles using HoloViz dynamic dashboards in JupyterLab and PyTorch FCN… | ||
| Web Scraping with BeautifulSoup & Yelp's API | Juan S Vasquez | 01h 07m |
Scrapes YellowPages HTML with BeautifulSoup + requests, targets h2.class="n", div.phone phone-primary, div.street-address and div.locality, appends… | ||
| Intro to Data Analysis with Python Data-table | Ana Castro Salazar, Pasha Stetsenko | 01h 16m |
Claims datatable is a pandas alternative built for big data with a C core with copy‑on‑write columns and virtual columns materialize lazily so many operations… | ||
| Git-ting along with others | Paul Anzel | 01h 22m |
A hands-on walkthrough of collaborative Git work shows actual commands and failures: creating branches and forks, pushing a new branch, running git fetch then… | ||
| Turn Python Scripts into Beautiful ML Tools | Adrien Treuille | 01h 09m |
Claim: a new open-source app framework called streamlet turns normal Python scripts into interactive ML tools. Concrete demos: pip install streamlet, a… | ||
| What's Data Science Reporting | Amy Tzu-Yu Chen | 31m |
Argues that data science reports must show the model’s business impact not just ML metrics, and prescribes one central metric (example: money per passerby)… | ||
| Building a Data Driven Organization | Franklin Sarkett | 42m |
Claims a 150× client growth from a MacBook and a CSV to deployed pipelines, using Python plus concrete tools: Black for deterministic formatting, Airflow for… | ||
| Write the Docs! | Hareem Naveed | 33m |
Claims that Sphinx + reStructuredText turns plain files into versioned HTML/PDF docs with five init commands and about eight build commands, and that nbSphinx… | ||
| Making Data Relevant to Business. It's Harder Than You think! | Ravin Kumar | 46m |
Prototype uses scikit-learn random forest and simple decision trees, achieves 85% accuracy in experiments but deploys as a batch double-check reducing misses… | ||
| A Python Package for Grappling with Missing Data | Joseph Kearney, Shahid Barkat | 47m |
Auto impute is a pip‑installable Python package with a scikit‑learn style API that runs end‑to‑end missing‑data workflows: explore missingness, run… | ||
| Data and ETL with Notebooks in Papermill | Matthew Seal | 39m |
Explains a concrete system: papermill parameterizes notebooks and runs them headlessly, injecting JSON params, monitoring kernels, and writing immutable… | ||
| How to Get Started with Server-less on Google, Amazon & Microsoft | Michelle Brenner | 39m |
Compares concrete setups and what actually worked: a CRUD template implemented on API Gateway + Lambda + DynamoDB, on HTTP trigger + Cosmos DB, and on Cloud… | ||
| A Supremely Light Introduction to Quantum Computing | Rodolfo Bonnin | 42m |
Plain introduction that states concrete claims and demos: explains qubits and Bloch sphere, cites real machines (IBM Q Experience, Rigetti, Google, AWS Braket)… | ||
| Bokeh Maps - Interactive Map for Your Next Web Application | Kevin Chrzanowski | 36m |
An interactive choropleth is built that visualizes 311 call volumes by neighborhood using pandas and GeoPandas with Dask for large files and Bokeh for… | ||
| Traditional & Novel Feature Selection Approaches | Ben Fowler | 30m |
Claims that combining fast gain-based pruning with final SHAP-backed backward selection and a split validation subset improves model generalization and gives… | ||
| Topological Techniques for Unsupervised Learning | Leland McInnes | 40m |
Claims that a combinatorial translation of metric spaces yields a practical dimension reducer and clustering pipeline: implements UMAP by mapping to a fuzzy… | ||
| Lightning Talks | John Mount | 24m |
Describes an early but working library that models queries as composable schema-to-schema operations rather than immediate dataframe calls, claiming parity… | ||
| A Guide to Modern Hyperparameters Turning Algorithms | Richard Liaw | 38m |
Covers concrete methods and results: compares grid/random/Bayesian search, shows HyperBand/ASHA early-stops bad trials and reports running 20+ samples with 4–8… | ||
| Gradient Boosting for Data with Both Numerical and Text Features | Vasily Ershov | 37m |
CatBoost claims automatic, high-quality handling of categorical and text features: target-leakage-free label encoding via time-ordered leave‑one‑out estimates… | ||
| A/B Testing in Python | Raul Maldonado | 39m |
A runnable Jupyter notebook generates fake ad data, implements manual proportion t/z calculations and a statsmodels proportion_ztest, and shows a nonmanual… | ||
| Dynamics Programming for Machine Learning- Hidden Markov Models | Avik Das | 39m |
Builds a Hidden Markov Model from first principles, implements a tiny Python Viterbi in a few lines and shows it recovers the most probable state path while… | ||
| RAPIDS- Open Source GPU Data Science | Brad Rees | 38m |
RAPIDS claims end-to-end GPU data science that keeps data on the GPU so ETL, joins, grouping and ML avoid repeated copies and run orders of magnitude faster: a… | ||
| Machine Learning Model Evaluation Metrics | Maria Khalusova | 40m |
Shows concrete comparisons and tools: logistic regression scored 96% accuracy while a dummy classifier scores 94.2% on a 10,000-sample set with class imbalance… | ||
| Open Sourcing Natural Disaster Preparedness and Relief | Nick Acosta | 29m |
IBM’s Code in Response runs Call for Code as an open-source pipeline that claims to move prototypes into deployed relief: Project OWL wins 2018 with an IoT… | ||
| Simplicity For Scale- Analyzing 15 Million DNA Samples With Python | Tim Orme | 51m |
Pipeline ingests 15 million DNA tests and replaces a year‑long Hadoop/Java batch system with event‑driven Python using Celery plus a C++ germline executable… | ||
| The Promise of Hierarchical Indices for Data Beyond 2 Dimensions | Christopher Ariza | 32m |
Claims that the pandas Panel deprecated leads to choices and compares actual tools: Panel, xarray, pandas MultiIndex and static_frame HLock. Presents concrete… | ||
| Open Source is Better Together- GPU Python Libraries Unite | Dante Gama Dessavre | 42m |
Rapids presents concrete interoperability work: it leverages Dask and UCX to reduce communication stalls, adapts Apache Arrow for columnar exchange, and… | ||
| Machine Learning Models Versioning Using Open Source Tools | Dmitry Petrov | 32m |
Shows a concrete workflow for tracking models and artifacts using open source pieces: git lfs for small model files, CI pipelines that capture reproducible… | ||
| Tackling Homelessness with Open Data | Eric Busboom | 42m |
San Diego regional data library scrapes HUD and local PDFs, builds a reproducible pipeline with a Python image extractor plus OCR, issues GitHub tickets and… | ||
| Accelerate NumPy Data Science Workloads and Deep Learning Applications | Hao Jin | 29m |
Apache MXNet exposes a numpy‑compatible interface that keeps NumPy interoperability while adding parallel kernels, deep‑learning ops and autodiff. Benchmarks… | ||
| Preparing Messy Data for Supervised | John Mount, Nina Zumel | 32m |
A Pythonic vtreat package converts messy tables into all-numeric frames, adds missing-value flags and replacement values, uses impact coding for… | ||
| Build an AI-powered Pet Detector in Visual Studio Code | Jeffrey Mew | 34m |
Builds a pet classifier inside Visual Studio Code using Jupyter notebooks, TensorFlow and MobileNet transfer learning, trains on the Oxford Pet dataset (37… | ||
| MAP All the Things | John Healy | 48m |
Embeddings are concrete: numeric vectors plus a distance, built here with UMAP and demonstrated end-to-end on 1.5M BeerAdvocate reviews collapsed to ~104 beer… | ||
| Building Named Entity Recognition Models Efficiently Using NERDS | Sujit Pal | 35m |
An open source framework called NERDS wraps gazetteers, CRF, BiLSTM±CRF and ELMo models and exposes a fit/predict/save/load API; a fork fixes NLTK deps, adds… | ||
| AI and Fashion- Product Retrieval with Multi-modally Generated Data | Ivona Tautkutė | 25m |
Claims a working pipeline that turns an image + text query into a synthetic image used for visual search, built with FashionIQ dataset, ResNet pretrained… | ||
| What You Got Is What You Got | James Powell | 35m |
Argues that numpy pandas and networkx are powerful but hard to embed in production: ndarray are contiguous unboxed machine types so they give speed and simple… | ||
| Using Embedding Layers to Manage High Cardinality Categorical Data | Fletcher Riehl | 35m |
A production system turns sparse search queries into usable features by mapping 10 million keywords into 50 dimensional vectors using pretrained GloVe where… | ||
| nothing matches | ||