PyData London 2022
2022-06-17..2022-06-19
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 |
|---|---|---|
| SQLAlchemy and You - Making SQL the Best Thing Since Sliced Bread | Anders Bogsnes | 01h 28m |
Claims SQLAlchemy is the gold‑standard for Python SQL work and shows concrete, runnable examples: uses Core and ORM side‑by‑side, creates an engine + metadata… | ||
| Feature Engineering Made Simple | Kajanan Sangaralingam, Anindya Datta | 01h 28m |
Claims a reproducible, procedural feature pipeline built as Anovos that runs on Spark and demos on a Kaggle credit‑risk set (with six simulated time periods)… | ||
| Probabilistic Python: An Introduction to Bayesian Modeling with PyMC | Chris Fonnesbeck | 01h 27m |
Overview of what is actually done and what comes out: introduces PyMC v4 and the Aesara/JAX backends, fits models with the No-U-Turn Sampler (NUTS), and shows… | ||
| Picking What to Watch Next: Build a Recommendation System | Cheuk Ting Ho | 33m |
Builds an item‑based collaborative filtering recommender by turning the dataset into an m×n rating matrix, filtering low‑activity users and unpopular items… | ||
| Data Science at Scale with Dask | Richard Pelgrim | 01h 24m |
Introductory walkthrough of Dask shows concrete wins: scaling pandas and NumPy to disk‑heavy workloads (12 CSVs ~15 GB, an example dataset of 665 million… | ||
| Document/Sentence Similarity Solution | Ade Idowu | 01h 28m |
A working set of three demos compares actual methods: a book‑title nearest‑neighbor, a toy search engine, and an STS benchmark evaluation. Text converts to… | ||
| Parallelism the Old Way: Using MPI in Python with mpi4py | Nick Radcliffe | 01h 31m |
Demonstrates using MPI4Py for real Python parallelism: shows blocking send/recv, how small messages get copied but larger ones hit a ~32KB buffering cutoff and… | ||
| Data Validation for Data Science | Natan Mish | 01h 24m |
Tutorial shows why data validation prevents trivial failures and how to do it: use Great Expectations to build an expectation suite and run checkpoints that… | ||
| Sktime—Python Toolbox for Time Series: How to Implement Your Own Estimator | Franz Kiraly | 01h 26m |
sktime claims scikit‑learn compatibility and shows how to implement estimators via extension templates, using the provided notebooks and GitHub repo to build a… | ||
| Introducing More of the Standard Library | Simon Ward-Jones | 01h 18m |
Covers concrete uses: file handling with pathlib paths and .read_text/.write_text and saving/loading game state to a data folder, date math like “days until… | ||
| Train Object Detection with Small Datasets | Vincenzo Crescimanna | 01h 15m |
Shows how to get usable detectors with a few hundred labeled images per class by combining three concrete tactics: remove near‑duplicates with hashing + latent… | ||
| Possible Futures For Jupyter | Sylvain Corlay | 20m |
Roadmap items are concrete: a visual debugger in JupyterLab (DAP-based) already works with three kernels; collaborative editing uses Yjs CRDTs with undo/redo… | ||
| Fuzzy Matching at Scale | Thusal | 30m |
Claims that TF-IDF n-grams plus approximate nearest neighbors replace edit-distance methods for massive fuzzy matching, outperforming Levenshtein at scale… | ||
| Off with their I/Os!—Or How to Contain Madness by Isolating Your Code | Sarah Diot-Girard | 36m |
Claim: isolate all I/O at the outermost layer so the computation lives as pure functions in a functional core, using pandas.read_csv, sklearn StandardScaler… | ||
| Python-centric Feature Stores | Jim Dowling | 42m |
Claims a working Python-first feature store pipeline using Hopsworks with point-in-time joins, online/offline dual stores (RonDB + Apache Hudi), Kafka… | ||
| Beyond Pandas: The Great Python Dataframe Showdown | Juan Luis Cano Rodríguez | 40m |
Compares pandas limits to concrete alternatives: Arrow shows memory-mapped CSV→Parquet conversion that drops a 1.9 GB CSV to 1.9 GB to <0.5 GB on disk; Vex… | ||
| How Pyodide and a New Opensource Community Are Improving Children’s Social Work | Tambe Tabitha A. | 38m |
Built a browser-run Python error checker using Pyodide in the browser and WebAssembly so sensitive case data never leaves the workstation, with front-end… | ||
| Making Fake Data Generators for Open Source Healthcare Data Science Projects | Matthew Cooper, Jennifer Hall | 39m |
Shows a working pipeline that produces fake tabular and image data for NHS projects and a separate synthetic route using synthvae trained on MIMIC‑III… | ||
| Unlocking the Power of Gradient-Boosted Trees (using LightGBM) | Pedro Tabacof | 37m |
LightGBM is shown to handle missing values natively and to do categorical encoding without one‑hot preprocessing, producing a baseline on the Kaggle… | ||
| Can you Read This? How I Improved Text Readability Online for Visually Impaired | Asya Frumkin | 37m |
Uses deep learning to detect illegible text over images and compares OCR engines tesseract and EasyOCR against a pixelwise text detector CRAFT model. Extracts… | ||
| Large Language Models for Real-World Applications - A Gentle Intro | Jay Alammar | 36m |
Jay Alammar Presents:
Large Language Models for Real-World Applications - A Gentle Intro
Machine language understanding and generation has been undergoing… | ||
| Data Pipelining for Real-time ML Models | Gabor Bakos | 39m |
Optiver builds a custom Python pipeline that ingests about 100 GB per minute from 50 exchanges and produces live ML outputs that match backtests, trading at… | ||
| Lightning Talks (01) | 01h 16m | |
Day‑long lightning set gives concrete outcomes and tools: an event called Big Data Barbecue and EuroPython/EuroSciPy dates; a claim that data marketplaces let… | ||
| Key Challenges In the PyData Ecosystem and How we Can All Make a Difference | Tania Allard | 54m |
PyData claims 25–40 million users and traces deep technical debt: core libraries still rely on Fortran 77, the NumPy-based stack remains largely… | ||
| Solving Real-World Business Problems with Bayesian Modeling | Thomas Wiecki | 42m |
Announces PyMC 4 with JAX and GPU sampling and shows concrete Bayesian builds: a CAC estimate that yields 25% probability a channel is profitable… | ||
| Using Graph Neural Networks to Embrace the Dependency Within Your Data | Usman Zafar | 39m |
Claim: graph neural networks beat naive tabular models when relationships matter by encoding structure into adjacency and Laplacian matrices and learning via… | ||
| Beyond Medical Image Segmentation. The Road Towards Clinical Insights | Tomasz Bartczak, Adam Klimont | 36m |
A multi-stage pipeline processes CT scans for abdominal aortic aneurysm planning using sub-organ multilabel segmentation and a skeletonization-based loss… | ||
| A Hitchhiker Guide to MLOps | Davide Frazzetto | 40m |
Claims that monitoring the ML lifecycle is practical end to end using only open source, and demonstrates it with MLflow for run/metric/artifact tracking and… | ||
| Notebooker: Production and Scheduling for your Jupyter Notebooks | Jon Bannister | 34m |
Notebooker converts Jupyter Notebooks into reusable templates that are parametrised reports, executed by Papermill execution and stored in a Mongo-backed… | ||
| Clusterf*ck: A Practical Guide to Bayesian Hierarchical Modeling in PyMC3 | Hanna van der Vlis | 36m |
Predicts Kenyan maize yield with PyMC3 using a log transform (added 0.1) on 110,789 rows, comparing pooled, unpooled (30 counties × 2 thetas = 60 posteriors)… | ||
| Hypothesis Testing Stop Criterion with "Precision Is The Goal" | Eyal Kazin | 43m |
Claims that Bayesian methods do not magically fix sequential stopping and instead recommends stopping on interval precision is the goal rather than p-value… | ||
| Extreme Multilabel Classification in the Biomedical NLP Domain | Nick Sorros | 34m |
A transformers‑based NLP model is built to tag biomedical grants with 29K MeSH tags, trained on a corpus of about 14 million publications, and it uses… | ||
| Clean Architecture: How to Structure Your ML Projects to Reduce Technical Debt | Laszlo Sragner | 37m |
Argues that a small, concrete engineering stack reduces ML technical debt: refactor pipelines into a class-based product, run the same code in laptop test and… | ||
| Accelerating Machine Learning at Scale with | Alejandro Saucedo | 36m |
Takes a GPT-2 model from HuggingFace, converts and quantizes it with Optimum into ONNX conversion artifacts, runs inference locally on MLServer and deploys to… | ||
| What is X up to? - NER and Relationship Extraction for Information Extraction | Ahmet Melek | 32m |
A concrete pipeline extracts named entities and classifies pairwise relations into a graph, using transformer-based NER and relation classifiers (tokenizers to… | ||
| Rethinking Data Visulisation with PyScript | Valerio Maggio | 44m |
Claim: run Python in the browser as a usable data‑science platform. Implementation: Pyodide via WASM underlies PyScript, with a pyenv YAML loader, micropip and… | ||
| AUC is Worthless: Lessons to Transition From Academic to Business Data Science | Dillon R Gardner | 40m |
Asserts AUC is worthless for business decisioning and builds a concrete loan example: fill a confusion-matrix value table, derive expected utility per… | ||
| Building Successful Data Science Projects | Ian Ozsvald | 40m |
Clear specifics decide whether to watch: is there a written specification and de‑risking plan, and does the $1 million client claim trace to measurable… | ||
| Whatever I Can do to Entertain You in 30 Minutes | James Powell | 26m |
Laptop fails to boot so notes are gone and a short live coding session runs on a terrible keyboard; claim: generators for simulation control are the right… | ||
| Keynote: Lighthearted Thoughts on Conversing About Data | Susan Mulcahy | 34m |
Program runs five postgraduate students on a six-week data analytics sprint with industry projects (gas distribution, price optimisation, law firm… | ||
| Feature Engineering for Time Series Forecasting | Kishan Manani | 43m |
Claims that machine learning (not classical ARIMA) wins on large cross-series problems: LightGBM wins M5 and yet gains are modest (top 50 beat exponential… | ||
| Machine Learning 2.0 With Hugging Face | Julien Simon | 49m |
A concrete workflow shows up: DistilBERT fine-tuned on 100k Amazon shoe reviews (20k per star), trained for a quick epoch on GPU, pushed to the Hugging Face… | ||
| Lessons Learned About Data & AI at Enterprises and SMEs | Alexander CS Hendorf | 39m |
Claim: enterprises and SMEs behave differently enough that cannot buy AI must build it becomes operational advice — commodity APIs ok, domain models require… | ||
| Understanding your Bank Statement in 100ms | Chady Dimachkie, Robin Kahlow, Jonathan Kernes, Ilia Zintchenko | 36m |
Claim: Ntropy parses financial transactions in under 100ms with near human accuracy. Pipeline details: attention‑based NER with multilingual training… | ||
| Models Schm-odels: Why you Should Care about Data-Centric AI | Marysia Winkels | 41m |
Claim: iterating on data beats fiddling with models for practical gains. Concrete actions include cleaning noisy labels, rebalancing train and validation… | ||
| Testing, Testing: On Experimental Drift and Data Driven Product Design | Yizhar Toren | 38m |
Claims that A/B tests break when the environment shifts: a measured 5% lift from a February experiment on pet beds flips by summer because the user mix moves… | ||
| Questions and Practices to Make Algorithmic Decision Making More Fair | Adrin Jalali | 40m |
Gives a measurement framework borrowed from social sciences and a concrete checklist of questions to ask about what is actually being measured, how proxies… | ||
| Signature Methods for Time Series Data | Sam Morley | 48m |
Signatures turn irregular, oscillatory paths into fixed‑size feature vectors and the pipeline is concrete: compute signatures with e‑cig stream_to_sig and… | ||
| Lightning Talks (02) & Closing Notes | chris lafra | 01h 02m |
Quick scan: a live demo claims Feinman instrumentation monkey‑patches functions and draws execution in the browser, showing thousands of modules loading and… | ||
| nothing matches | ||