PyData Miami 2019
2019-01-09..2019-01-11
Miami, United States
who it's for
data scientists, machine learning engineers, natural language processing engineers, data engineers, research scientists, bioinformaticians, visualization engineers
topics
Bokeh application deployment on Linux; Dash interactive app development; nested cross validation with scikit-learn; BOHB hyperparameter optimization (hpbandster); Yellowbrick visual diagnostics for ML; Apache Arrow in-memory integration; Rapids GPU accelerated data science; productionizing TensorFlow models; H2O model production with Apache Spark; Soss probabilistic programming (Julia); unsupervised question-answer generation pipelines; single-cell bioinformatics with N-dimensional arrays
| title | author | length |
|---|---|---|
| Exposing Dark Data in the enterprise with custom NLP | Justin J. Nguyen | 34m |
Claims an open-source framework that generates reliable question‑answer pairs with minimal labeled data using unsupervised and semi-supervised learning, plus… | ||
| Responsible AI Practices: Fairness in ML | Alex Hanna | 40m |
Clear claims and concrete tools are described so watch if you want methods not platitudes. Shows a measured claim that pre trained embeddings encode… | ||
| Does This Algorithm Have a Soul? | Betina Costa | 31m |
Examples and numbers decide: Pro Publica COMPAS finding shows risk scores wrongly label Black defendants as higher risk about twice as often, facial… | ||
| Build and deploy a Python bokeh application on a Linux server | Russell Burdt | 28m |
Builds a real Python bokeh application and deploys it on a Linux IaaS server using Flask, systemd and nginx so it actually runs and restarts on reboot. Shows… | ||
| You Should Probably Be Doing Nested Cross-Validation | Sergey Feldman | 24m |
Claim: nested cross-validation gives much less biased, lower-variance estimates of final model performance on small datasets than a single… | ||
| Soss - Lightweight Probabilistic Programming with No Performance Ceiling | Chad Scherrer | 24m |
Concrete claim that Soss encodes models as manipulable Julia code and removes the usual Python/C++ split, demonstrated by a working NUTS sampler, a… | ||
| Words in Space | Rebecca Bilbro | 30m |
Visual diagnostics show specific, actionable differences: using Yellowbrick and scikit‑learn with SciPy metrics (Euclidean, Manhattan, Minkowski, Jaccard… | ||
| Ursa Labs and Apache Arrow in 2019 | Wes McKinney | 51m |
Ursa Labs forms as a nonprofit to fund Apache Arrow work with five full‑time engineers and a 2019 roadmap. Arrow defines a shared in‑memory tabular format used… | ||
| Intro to Dash | Chelsea Douglas | 33m |
Demo shows a working Dash app end to end: layout with Dash HTML and core components, callbacks with one output and multiple inputs, and interactive plotly.js… | ||
| The necessity of pipelines in Natural Language Processing | Zachariah Miller | 27m |
Argues that reproducible pipelines are the only practical way to manage dozens of preprocessing choices, demonstrating code that wraps tokenizers, cleaners… | ||
| Efficient and Flexible Hyperparameter Optimization | Dan Ryan | 26m |
BOHB claims to combine Hyperband’s multi‑resolution budgeted evaluations with a Bayesian surrogate (tree‑parzen style) to get fast and robust hyperparameter… | ||
| Visual Diagnostics for More Effective Machine Learning | Benjamin Bengfort | 35m |
Demonstrates a concrete library and patterns: the Yellowbrick Visualizer that plugs into the scikit‑learn API, producing feature, classification, clustering… | ||
| Your data fits in RAM: How to avoid cluster computing | Aaron Richter | 35m |
Claim: most “big data” workloads run on one machine if enough RAM is rented in the cloud. Evidence: AWS offers up to 768 GB instances and rentable GPUs for… | ||
| Platform for Complete Machine Learning Lifecycle | Jules S. Damji | 37m |
Open source MLflow from Databricks claims an end-to-end ML platform by providing concrete APIs and formats: a tracking API that logs params, metrics and… | ||
| Machine Learning & Healthcare | Michoel Snow | 28m |
Claims that algorithms outscore radiologists hinge on narrow setups not real work: CheXNet vs radiologists uses single frontal x‑rays with no prior studies or… | ||
| Accelerating Single-cell Bioinformatics with N-dimensional Arrays | Ryan Williams | 31m |
Claim: parallelize single-cell workflows across Spark Dask Pywren and Apache Beam and store arrays as Zarr instead of HDF5. Implements Zarr format in scanpy… | ||
| How We Lie to Ourselves With Charts | Alberto Cairo | 48m |
Explains concrete failure modes: the hockey‑stick claim shows reconstructed estimates plus uncertainty, the 2016 map misleads because area ≠ voters so a… | ||
| Productionizing H2O Models with Apache Spark | Jakub Hava | 33m |
A concrete workflow shows H2O models embedded inside Apache Spark so a single serialized artifact handles training, feature transforms and serving; it uses… | ||
| Nasa Influence on Open Data Projects | Anita Kirkovska | 28m |
Overview of concrete outcomes not fluff: a description of NASA Space Apps Challenge numbers (25,000 participants, 25 solutions in 2017, 150,000+ engaged over… | ||
| Model Evaluation for Humans | Dante Gates | 27m |
Provides concrete evaluation steps: frame success as a business metric (profit or dollars), set a deployable last-week rating baseline, check for target… | ||
| Deep Learning vs. Conventional Machine Learning | Chao Han | 39m |
Compares concrete models and results: a triplet Siamese encoder with pretrained fastText embeddings LSTM attention and triplet loss on 20k QA pairs outperforms… | ||
| Smart Streams: A Real-time framework for scoring Big and Fast Data | Vedant Jain | 30m |
Claims a working real-time scoring pipeline that ingests device CSV/JSON from S3, runs ETL and windowed aggregation in Spark Structured Streaming, calls a… | ||
| Introduction to Data Analysis Methods in Particle Physics | Will Hunter | 28m |
Brief overview of the Standard Model and the 27 km LHC with the 3000‑ton ATLAS calorimeter leads to an actual data pipeline: CERN/ATLAS Monte Carlo and… | ||
| Pandas and Plots for Data Analysis | Abhijit Dasqupta | 01h 29m |
Concrete workflow: live notebooks on a Google Colab live notebook (Binder link fails), using the Anaconda Python distribution, JupyterLab, numpy then pandas… | ||
| Introduction to PyTorch | Robert Alvarez | 01h 32m |
Hands‑on PyTorch code runs on Google Colab with GPU and a linked GitHub repo; examples include a toy adding network (learning takes ~2,000 training samples), a… | ||
| Effortless APIs with Flash and AWS Lambdas | Santiago Basulto | 01h 10m |
Demonstration shows a Flask app packaged with Zappa and deployed to AWS Lambdas in minutes, running an image color extraction service using scikit-learn KMeans… | ||
| Putting Tensor Flow Models Into Production | Bugra Akyildiz | 01h 18m |
Concrete deployment steps and measured behavior appear: build a Docker image that runs TensorFlow Serving, expose gRPC and HTTP ports (500 and 501), and point… | ||
| GPU Data Science with Rapids | Joshua Patterson, Keith Kraus | 01h 08m |
RAPIDS shows concrete work: GPU-accelerated ETL and ML using cuDF & cuML, Arrow in GPU memory, Dask for multi‑GPU scaling and CUDA IPC/UCX to avoid copies. A… | ||
| Applying Rigorous Machine Learning Methods in Business Strategy | Mash Zahid | 01h 39m |
Data is profiled with pandas profiling, a high‑missing smoker column is dropped, numeric gaps get group‑averaged imputation, and categorical fields get one‑hot… | ||
| Deep Learning and Modern NLP | Zachary S. Brown | 01h 59m |
Step‑by‑step notebooks use scikit‑learn tf‑idf and PyTorch to build a perceptron, RNN/LSTM and an encoder‑decoder seq2seq, with integer‑encoded embeddings and… | ||
| nothing matches | ||