PyData London 2017
2017-05-05..2017-05-07
London, United Kingdom
[PyData London 2016, PyData London 2017, PyData London 2018, PyData London 2019, PyData London 2022, PyData London 2024, PyData London 2025, PyData London 2026]
| title | author | length |
|---|---|---|
| Word Embedding For Fun and Profit | Lev Konstaninovskiy | 01h 03m |
Trains gensim word2vec on a ~170k‑word movie‑plots corpus from the movie-plots-by-genre repo, compares baselines: majority label 40%, tf‑idf bag‑of‑words 46%… | ||
| Make your research interactive with Jupyter Dashboards | Pavlo Andriychenko | 01h 16m |
Shows how to turn notebooks into live web apps with Jupyter Dashboards, using Bokeh + ipywidgets for interactivity and a kernel-gateway to route execution… | ||
| Introduction to Convolutional Neural Networks using TensorFlow and Keras | Oliver Zeigermann | 01h 33m |
Uses TensorFlow and Keras to classify German speed‑limit signs from a tiny dataset (~379 images, six labels) by resizing everything to 64x64 images… | ||
| Twisted up in a Distributed Tornado | Andrew Stretton | 01h 10m |
Begins by showing concrete exercises not platitudes a t SNE image pipeline pulling images then parallelizing with process and thread pools and scaling to… | ||
| Jupyter Notebooks for geospatial data analysis | Julia Wagemann, Stephan Siemen | 01h 27m |
Jupyter Notebooks show executable workflows that fetch ECMWF reanalysis via OGC WMS/WCS and owslib, run WCPS server-side processing (e.g. Kelvin→°C, monthly… | ||
| Building Web-based Analysis & Simulation Platforms | James Powell | 01h 49m |
Live code shows a working pipeline: a Numpy Monte Carlo integrator (used to approximate π, ~10 million samples ~10s) is wrapped as a Python script, exposed… | ||
| Test-Driven Data Analysis | Nick Radcliffe | 01h 11m |
Claims a reproducible workflow: capture inputs and outputs, run reference tests that can auto‑regenerate expected results, and infer constraints on Pandas… | ||
| Bayesian optimisation with scikit-learn | Thomas Huijskens | 39m |
Demonstrates Bayesian optimisation implemented with Gaussian processes as priors and an expected improvement acquisition function, using the scikit-learn GP… | ||
| Scale out from the very beginning | Jens Nie | 38m |
Implements a nightly distributed filesystem scanner over 1.5 PB storage that crawls ~60 million files and ~500k directories, produces CSV indexes pushed to… | ||
| High-Performance Distributed Tensorflow | Chris Fregly | 36m |
Demos show concrete optimizations: 8-bit quantization and weight rounding shrink a 32k float model to ~13k and activation calibration is needed… | ||
| Using Random Forests in Python | Nathan Epstein | 27m |
Shows concrete results: out‑of‑the‑box scikit-learn regressor fits a noisy xsin(x) curve with no tuning, a random forest classifier* perfectly separates… | ||
| Journeys through JuPyteR | Alex Glaser | 29m |
Rewrites Tariq Rashid’s Python neural‑net code into R and Julia, trains on MNIST and reaches 97% accuracy with the reference single‑hidden‑layer network… | ||
| To explain or to predict? | Nick Sorros | 38m |
Argues that predictive models win at prediction while regression gives causal significance, and shows concrete tradeoffs: an SVM classifier (trained on split… | ||
| Yellowbrick: Steering Machine Learning with Visual Transformers | Rebecca Bilbro | 39m |
Yellowbrick is a pure‑Python library that extends Scikit‑Learn with a core Visualizer object so Visualizers fit and transform inside pipelines and produce… | ||
| Analyzing the ElectroCardioGram (ECG) | Emlyn Clay | 46m |
Home-recorded three-lead ECG (V2) loads from disk with h5py, or from Physionet via python-wfdb, then a Butterworth bandpass and smoothing reduce noise, wavelet… | ||
| A beginner's guide to data analysis in cosmology using Jupyter Notebook | Caroline Clark | 32m |
Downloads three CMB maps plus one foreground map and a power spectrum from the Planck release, uses healpy read_map and FFTPack to transform maps into… | ||
| Picasso's terminal; data science and AI in the visual arts | Gene Kogan | 56m |
Shows concrete results not promises: deep dream amplifies doglike hallucinations and spawns fractal puppy slugs, fast tensorflow style transfer runs real time… | ||
| Diversity and Data: Cases in the Music Industry | Delger Enkhbayar, Mike Sumner | 35m |
PPL reports a repertoire database with over 10 million recordings and receives ~919,000 recordings this year so scaling is real. A recommendation engine is… | ||
| Variational Inference and Python | Peadar Coyle | 35m |
Compares Markov Chain Monte Carlo to variational inference and demonstrates fitting a Bayesian neural network with PyMC3 and Edward, using ADVI and minibatch… | ||
| Pythonic Polling Analysis and Comments on 2016's Polling Surprises | Aileen Nielsen | 48m |
Claims that 2016 polls were catastrophically wrong give way to specifics: national aggregates mostly track the result while state-by-state estimates miss… | ||
| One workshop that data scientists don't want you to attend... | Oliver Laslett | 01h 17m |
Shows how a real system is built and what actually works: labels about 1,500 labeled articles, vectorizes titles+descriptions with pandas and a scikit-learn… | ||
| Ten Steps to Keras | Valerio Maggio | 01h 37m |
Ten Jupyter notebooks walk through concrete Keras code: ten Jupyter Notebooks show building an MLP, CNN, RNN, transfer learning and fine‑tuning, ResNets… | ||
| Easy Bayesian regularization for fitting financial time series and curves | Egor Kraev | 35m |
Claim: massively overparameterize curves and then remove degrees of freedom with convex regularizers and you get robust fits. Uses Python notebooks (code on… | ||
| How distributed representations make chatbots work (at least a bit) | Nils Hammerla | 42m |
Distributed word vectors produce the actual claim: similar-context words sit near each other so a simple classifier can route queries to content. The pragmatic… | ||
| Automated factchecking in the era of fake news | Will Moy, Mevan Babakar | 40m |
Full Fact describes a working automated fact‑checking engine that scans Hansard, online papers and TV subtitles, matches lines to a database and surfaces prior… | ||
| Julia: A Fresh Approach to Machine Learning | Mike Innes | 43m |
Julia claims real-world speed without leaving high-level code and shows concrete tooling: Flux for differentiable models, native GPU kernels and easy… | ||
| Smelly London: visualising historical smells. | Deborah Leem, Barbara McGillivray | 39m |
Smelly London brings a dataset of 5,500 MOH reports into reproducible analysis and visualization, mining OCR text with Python 3, NLTK, SQLite and… | ||
| Dimension Reduction and Extracting Topics - A Gentle Introduction | Tariq Rashid | 38m |
Explains singular value decomposition (SVD) and uses a term‑document matrix with tf‑idf weighting to extract topics from real datasets, including the Chilcot… | ||
| Bayesian Deep Learning with Edward (and a trick using Dropout) | Andrew Rowan | 39m |
Claims that Edward (TensorFlow + probabilistic random-variable classes) runs blackbox variational inference on deep nets and that MC Dropout is formally… | ||
| Recommender systems with Tensorflow | Guillaume Allain | 38m |
Shows how to express matrix factorisation and feature-augmented embeddings in Tensorflow while using Pandas and Numpy for data prep, trains on MovieLens style… | ||
| Data for good: Lessons from the frontline | Emma Deraze, Charlie Harrison | 47m |
DataKind UK describes concrete results: a WeFarm data dive processes 64,000 users / 240,000 messages, uses topic modeling and simple classifiers to detect… | ||
| An Algorithm of Style | Ed Snelson | 42m |
Thread scales eight stylists to about half a million clients by combining stylist-curated candidates with ML: a simple non-negative matrix factorization on… | ||
| Forecasting social inequality using agent-based modelling | James Allen | 39m |
Agent‑based model of UK demography uses ONS and census inputs and Python (numpy, pandas) to run 150‑year simulations on 10,000 agents across many AWS cores… | ||
| Is having dementia linked to where you live? | Frank Kelly, Ondrej Urban | 37m |
Claims a link between diesel-related particulates (PM2.5/PM10/NO2) and dementia, with a hint of increased early-onset cases in high-exposure areas; uses Python… | ||
| Efficient and portable DataFrame storage with Apache Parquet | Uwe L. Korn | 29m |
Apache Parquet stores column chunks split into ~16 KB pages, uses dictionary encoding plus run‑length and bit‑packing before codecs like snappy, and supplies… | ||
| PyData London 2017, Lightning Talks - Saturday | 36m | |
Collection of five rapid demos shows concrete claims and tools: Facebook Prophet produces probabilistic daily forecasts with changepoints and uncertainty… | ||
| Next generation of word embeddings in Gensim | Lev Konstantinovskiy | 39m |
Gensim presents concrete comparisons: fastText uses subword n‑grams so unseen forms like "knights" get plausible vectors and it outperforms on… | ||
| Outlier detection methods for detecting cheaters in mobile gaming | Andrew Patterson | 42m |
Mobile cheaters are flagged by measurable claims not topics: the pipeline uses single‑variable caps, Mahalanobis distance and minimum volume ellipsoid for… | ||
| Astrophysics to data Science: how the Milky Way is like my company. | Kathryn Harris | 35m |
Claim: an astrophysics skillset transfers to building practical data products fast inside an SME. Evidence: a brand‑new two-person data team ships recommender… | ||
| Analyzing 3D objects with power of Deep Learning and Cython | Alexandr Notchenko | 43m |
Claims a working Python module PySparseConvNet that marries C++/CUDA kernels to Python via Cython and specific .pxd/.pyx patterns, with attention to explicit… | ||
| Leveraging recommender systems to personalise search results | Soraya Hausl | 32m |
Claims that recommender outputs can re-rank search results are demonstrated as a proof of concept using Pyspark to build user/item factors and Elasticsearch to… | ||
| Interactively Analyse 100GB of Data | Raoul-Gabriel Urma, Valentin Dalibard | 45m |
Live demo that interactively analyses a 100GB Open Library dump using Apache Spark on Amazon EMR with an Apache Zeppelin notebook, running the same… | ||
| WTF am I doing? An introduction to NLP and ANN's | Jeff Abrahamson | 33m |
Project records active window titles once per minute and saves screen snapshots as 100x100 thumbnails, producing about 300,000 observations. Text uses… | ||
| PyData London 2017 Lightning Talks - Sunday | 39m | |
Fast, concrete wins: a one-step graphical Windows installer built with Conda constructor that bundles .NET and Java and actually installs; heritage-site… | ||
| PyData London 2017 - Closing Notes | 20m | |
Twitch-powered closing notes are assembled from a Google Doc (bitly/twitch-pydata) that exposes 26 slots, producing on-screen slides live while AV glitches and… | ||
| "Lights, camera, AI!" - Automated sports videography | Zack Akil | 36m |
Builds a Raspberry Pi–mounted camera plus servo that points at play, runs a compressed machine‑vision pipeline (frame differencing then column summing) and an… | ||
| Extracting Insight From A Muslim Marriage App | Laila Alabidi | 37m |
Analyzes anonymized MuzMatch data with Python, using NetworkX after neo4j crashes, community detection via Louvain modularity, and a random forest plus tf‑idf… | ||
| Static Type Analysis for Robust Data Products | Marco Bonzanini | 31m |
Shows actual adoption steps: use the typing module (Python 3.5+) to annotate functions, run mypy on CI, catch concrete errors like “incompatible return value… | ||
| Machine learning with ventilator data | Ian Ozsvald, Gusztav Belteki, Giles Weaver | 42m |
Processes 100Hz dual-channel ventilator data from neonatal ICU traces, about 160 ventilator days collected (each day ≈600–700 MB) producing roughly 43,000… | ||
| Building robust machine learning systems | Stephen Whitworth | 38m |
Systems use concrete practices: automated pipelines with Luigi and BigQuery, offline feature code reused online, model provenance saved as dictionaries and… | ||
| Ranking hotel images using deep learning | Nuno Castro | 29m |
Expedia builds a model that ranks hotel photos using transfer learning from VGG and trains a regressor in Keras with the TensorFlow backend, labeling 100,000… | ||
| SaaaS - Sampling as an Algorithm Service | Vincent D Warmerdam | 39m |
Explains that sampling recovers posteriors by building a naked for loop MCMC sampler in Python, then swaps to PyMC3 for speed and applies it to a Santa Kaggle… | ||
| Machine Learning in Financial Credit Risk Assessment | Soledad Galli | 39m |
Describes building a production credit‑risk pipeline that combines applicant fields and credit‑agency history at scale (over £2bn lent to ~240,000 borrowers)… | ||
| Building a ChatBot with Python, NLTK and scikit | Edward Bullen | 01h 31m |
Walks through runnable Python notebooks that build a Q&A chatbot using a SQL-backed core, uses NLTK tokenization and POS tagging and the Stanford CoreNLP… | ||
| nothing matches | ||