PyData Paris 2024
| title | author | length |
|---|---|---|
| Building web-based engineering applications with JupyterLab components | Trung Le, Martin Renou | 27m |
Shows a concrete pipeline: assemble JupyterLab components into a CAD/GIS app, keep the scene as a small JSON shared model, and add real-time collaboration via… | ||
| Building Large Scale ETL Pipelines with Dask | Patrick Hoefler | 29m |
New Dask DataFrame implementation claims concrete wins: switching to pyarrow-backed strings cuts string-heavy memory use by over 50% and speeds some string ops… | ||
| The expanding Apache Arrow universe | Joris Van den Bossche | 32m |
Overview of concrete wins and tools: Arrow provides ADBC drivers (BigQuery among them) with a fetch_arrow_table API that avoids rowwise overhead; a nanoarrow C… | ||
| Polars Plugins: how you (yes, you!) can extend Polars Dataframes | Marco Edward Gorelli | 31m |
Claim: writing a Polars plugin is practical and yields real speedups. Workflow uses a cookiecutter template, Polars plus maturin, Rust for the UDF… | ||
| DIY Personalization: How, when why to offer self-made AI | Katharine Jarmul | 48m |
A demo builds a local RAG in a Jupyter notebook: initial LLM-as-retrieval fails, switching to BM25 + sentence encoder plus reranking and a local LLaMA-style… | ||
| Open-source AI: why it matters and how to get started | Merve Noyan | 35m |
Claims that open-source LLMs already match proprietary models and that almost 1 million models live on Hugging Face Hub. Concrete tooling and numbers: model… | ||
| Lightning talks - session 1 | 47m | |
Eight rapid demos and claims across tools papers and prototypes include concrete takeaways not fluff: Codex suggests buggy GitHub like code when training data… | ||
| Unpack business metrics to explain their evolution | Max Halford | 35m |
Provides a concrete method that decomposes KPI changes into inner and mix effects and quantifies contributions instead of handwavy statements: e.g. 2022… | ||
| Is your marketing effective? Let Bayes decide! | Emanuele Fabbiani | 30m |
A Bayesian media-mix model for WeRoad quantifies delays, saturation curves, and a positive baseline, using five channel investments to predict weekly new… | ||
| Building with Mistral | Sophia Yang | 40m |
Mistral lays out exact models, licenses and demos: pixol 12B is open-weight Apache 2.0, multimodal with a 128k context window and image-to-JSON demos, one… | ||
| An Introduction to 'metalearners', a Flexible MetaLearner Library in Python | Klein, Martí Escofet | 32m |
A Python library called metalearners implements R‑, X‑ and DR‑style meta‑learners that output CATE estimates via a fit/predict API, reuses fitted base models… | ||
| Dreadful Frailties in Propensity Score Matching and How to Fix Them | Alexandre Abraham | 27m |
Demonstrates that using standardized mean difference alone yields huge variability across ~100 propensity/matching pipelines, shows synthetic tests where… | ||
| Adaptive Prediction Intervals | Andro Sabashvili | 33m |
Compares specific adaptive interval methods: split conformal with quantile regressors (Conformal Quantile Regression), weighted residual schemes (LCP)… | ||
| An update on the latest scikit-learn features | Lemaitre, Senger | 31m |
Changelog-level details: scikit-learn 1.4–1.5 adds an metadata routing API that lets top-level tools propagate sample_weight and groups into nested… | ||
| Handling predictive uncertainty in Machine Learning | Olivier Grisel | 44m |
Shows concrete recipes: use scikit-learn probabilities, score models with log loss and Brier score (strictly proper scoring rules) and tune hyperparameters to… | ||
| Lightning talks - session 2 | Fire Duck | 51m |
A new library called FireDuck claims a pandas-compatible compiler with lazy JIT that is multi‑threaded rather than a thin wrapper, and it performs query… | ||
| Would you rely on ChatGPT to dial 911? | Nicolas Guenon des Mesnards | 29m |
Claims that production systems must mix deterministic rules with probabilistic LLM outputs and shows concrete tooling: rule‑first prefilter + ML classifier… | ||
| Evaluating the evaluator- RAG eval libraries under the loop | El Mawass, Knorps | 30m |
An experiment runs 200 RAG queries with embeddings stored in chrom ADB and compares evaluator libraries like RAGAS and TruLens across four metrics (answer… | ||
| Leveraging LLMs to build supervised datasets suitable for smaller models | Bel-Letoile, Carton | 31m |
Claims that LLMs can produce training data fast enough to let simpler models compete, then shows what actually changed: a salary extractor (logistic regression… | ||
| Unveiling new maps of biology with Squidpy | Inácio Medeiros | 29m |
Spatial transcriptomics gets practical: Squidpy is a Python library that loads Visium-style count matrices plus images, computes spatial statistics (Moran’s… | ||
| Geoscience at Massive Scale | Hendrik Makait | 30m |
A concrete account of scaling geoscience: a 250 TB catalog is sampled to a 12 TB subset for ~$2 of cloud compute, the cluster holds ~3.5 TB in memory while the… | ||
| xsimd: from xtensor to firefox | Serge « sans » Paille | 33m |
xsimd provides a portable C++ SIMD abstraction that compiles to real ISA intrinsics (Intel, ARM, RISC‑V, WebAssembly) with both static and dynamic dispatch… | ||
| Foundational Models for Time Series Forecasting: are we there yet | Baggi, Orlandi | 30m |
Claims center on Amazon Chronos and Salesforece's Moirai as transformer-based, zero-shot forecasters trained on massive pooled corpora (billions to tens of… | ||
| sktime: python toolbox for time series- next-generation AI | Heidrich, Kiraly | 30m |
sktime extends its scikit‑learn–style API to load and run deep learning and foundation models (e.g. morai, tiny times mix, Times of M) and composes them with… | ||
| Counting down for CRA: updates and expectations | Cheuk Ting Ho | 29m |
EU vote happens later this year with formal adoption possibly taking years; law carves out an open source steward category and adds supplier-style obligations… | ||
| Visualization of the sky in Notebooks: the ipyaladin widget extension | Matthieu Baumann, Manon Marchand | 30m |
ipyaladin embeds Aladin Lite into Jupyter and actually exchanges complex astronomy objects between JS and Python so users set targets, get astropy coordinates… | ||
| Onyxia: A User-Centric Interface for Data Scientists in the Cloud Age | Comte | 30m |
Claims a concrete replacement for JupyterHub by running on Kubernetes and Helm and providing per-user Kubernetes namespace with S3 bucket and Vault directory… | ||
| Towards a deeper understanding of retrieval and vector databases | Noe Achache | 33m |
Practical takeaways: a deduplication pipeline runs on 15 million images using DinoV2 image embeddings into an HNSW-backed vector DB, exposing that prefiltering… | ||
| Chainsai:- facilitating sampling of multimodal probability distributions | Simeon Carstens | 31m |
Chainsail is an open source Python web service that implements Replica Exchange (parallel tempering) to sample truly multimodal posteriors, interoperates with… | ||
| MLOps at Renault: A Generic Pipeline for Scalable Deployment | Carton, Tiran-Cappello | 32m |
Claims a Python-first Kubeflow + GCP Vertex AI pipeline that deployed 12 supply‑chain models in months, now used by 16 projects with 5 in production and… | ||
| Python 3.12's new monitoring and debugging API | Johannes Bechberger | 30m |
Python 3.12 exposes PEP 669 low‑impact monitoring that registers callbacks in the mon namespace, enables events per function and per thread, and patches… | ||
| High Performance Data Visualization for the Web | Tim Paine | 28m |
Builds a high performance web frontend using Perspective and demonstrates realtime filtering sorting pivoting and aggregation on large and streaming datasets… | ||
| Exploring Quarto Dashboard for impactful and visual communication | Christophe Dervieux | 30m |
Quarto converts Jupyter notebooks or .qmd sources into a dashboard layout of cards, value boxes and tab sets, using a CLI with quarto run --execute and live… | ||
| Solara: Pure Python web apps beyond prototypes and dashboards | Breddels, Rotko | 25m |
Solara claims to let you build large production Python web apps (not just prototypes) by running IPyWidgets on a server backed by Starlette/FastAPI and a… | ||
| Bridging the worlds: pixi reimplements pip and conda in Rust | Nichita Morcotilo | 29m |
Pixi reimplements conda and pip ecosystems as a single Rust binary and claims a working, cross‑platform package manager and task runner. It integrates the new… | ||
| Unveiling Mamba 2.0: The Future of Fast Package Management | Johan Mabille, Julien Jerphanion | 30m |
Mamba 2.0 claims a full refactor into a modular C++ framework with a unified codebase so mamba and micromamba share logic, producing a single 5 MB static… | ||
| Bridging Scientific and Industrial Worlds for Advanced Signal and Image Processing | Pierre Raybaut | 31m |
DataLab is an open-source desktop platform for processing 1D signals and 2D images that runs locally on Windows/Linux and embeds into Python workflows. It… | ||
| Boosting AI Reliability- Uncertainty Quantification with MAPIE | Cordier, Laurent | 30m |
Library MAPIE implements conformal prediction methods that produce prediction intervals and sets with provable marginal coverage under exchangeability… | ||
| JupyterLite, Emscripten-forge, Xeus, and Mamba | Tuloup, Thomas | 31m |
JupyterLite runs entirely in the browser using WebAssembly kernels so deployments serve only static files and scale to millions via static files; Xeus provides… | ||
| Open Source Sustainability & Philanthropy: Building Contributor Communities | Wright, Dave | 33m |
A corporate volunteer incentive model converts logged contribution hours into charitable donations via an open source dollars for your hours program partnered… | ||
| Fast NetworkX and How Accelerated Backends Are Changing Graph Analytics | Erik Welch | 31m |
NetworkX now dispatches hundreds of algorithms to accelerated backends so the same API runs on GPUs (cugraph + cuDF), GraphBLAS implementations, or falls back… | ||
| Color-composite images from the James Webb Space Telescope | Jesse Averbukh | 26m |
Open-source Jdaviz loads JWST FITS from MAST, aligns by WCS, maps filters to RGB and tweaks per-layer stretch/opacity to produce color composites. The demo… | ||
| Jupylates- spaced repetition for teaching with Jupyter | Nicolas M. Thiéry | 32m |
A small Jupyter extension called Jupylates builds exercises as regular notebooks with hidden solution and test cells, randomized variants, and a lightweight… | ||
| Enhancing RAG-based apps by constructing and leveraging knowledge graphs with open-source LLMs | Silva | 31m |
Demo builds a Metallica dataset, stores combined text embeddings in LanceDB with gte‑tiny embeddings and shows vector RAG returns correct song answers… | ||
| Track your code's CO2 emissions with Code Carbon | Luis Blanche, Benoît Courty | 31m |
CodeCarbon is an open source Python package that measures CPU GPU and RAM energy use and multiplies it by regional carbon intensity (via CO2 Signal /… | ||
| On the structure and reproducibility of Python packages - data crunch | M. Knorps, Z. Zhang | 35m |
Data-driven analysis compares 1,180 top PyPI packages with ~2,260 biomedical projects and uses FawltyDeps to detect undeclared imports. Key findings: 99.9% of… | ||
| Processing medical images at scale on the cloud | Guillaume Desforges | 28m |
Whole Slide Images (WSIs) are tens of gigabytes each so the project shows how to avoid copying whole files, read only the tiles needed, and feed patches into… | ||
| Collaborative editing in Jupyter | David Brochart | 28m |
Shows that Jupyter now uses CRDTs (yjs in practice) to make notebooks locally editable and automatically converge, with both centralized Jupyter Collaboration… | ||
| nothing matches | ||