PyData Global 2022
2022-12-01..2022-12-03
-- online --
[PyData Global 2020, PyData Global 2022, PyData Global 2023, PyData Global 2024]
as the organisers put it
PyData Global 2022 is where users, contributors, and newcomers can share experiences to learn from one another and grow together.
who it's for
data scientists, python developers, machine learning engineers, data engineers, mlops engineers, data visualization engineers
topics
Dask for single-node and cluster scaling; time series forecasting for energy; probabilistic demand forecasting; transformer model engineering; building localized language models; model traceability and reproducibility; testing pandas and property testing; reproducible publications with Quarto and Python; Ray AIR with Hugging Face Trainer; TPU-accelerated model training; expressive fast dataframes with polars
| title | author | length |
|---|---|---|
| Geometry for Social Good | Colleen M. Farrelly | 13m |
Uses persistent homology to detect urban growth and climate “tipping points” and applies this to epidemic spread, with a concrete example of positioning… | ||
| Build a web app in Python | David Chapuis | 09m |
Creates a tiny web app using Dash, loads data into a pandas DataFrame built from three dictionaries, and draws a grouped bar chart comparing fruit amounts in… | ||
| Everybody Is an Influencer–Which Influencer Are You | Dina Bavli | 08m |
Classifies influence with graph metrics and a concrete Python tool: uses NetworkX to build follower/mention graphs, measures participation as out‑degree and… | ||
| Life, Death, and Shopping | Dina Bavli | 12m |
Claims a practical pipeline that turns past event logs into labels for near‑future purchase prediction using sliding run dates and aggregated temporal… | ||
| HuggingFace + Ray AIR Integration: A Python developer’s guide to | Antoni Baum | 28m |
Shows a working integration between Hugging Face Transformers and Ray AI Runtime: the Ray AIR Hugging Face Trainer runs existing Trainer code distributedly… | ||
| Modern Analytics in the Cloud-A case for fraud detection | Marwa Ahmed | 26m |
Claim: an event‑driven, serverless pipeline on AWS detects e‑commerce fraud with real‑time detection using EventBridge, Lambda, S3, DynamoDB and either… | ||
| IMF Data Discovery and Collection | Irina Klein | 23m |
Claims that the IMF JSON RESTful web service is usable but inconsistent across series and too complex for non-developers, and shows a working Python workflow… | ||
| Critical CV-NLP Data Errors and How to Fix Them with Galileo | Nikita Demir | 29m |
Claim: 80/20 performance split — most gains come from fixing data, not model tinkering. Demonstrates concrete error types: mislabeled examples, class overlap… | ||
| Discover Inspirational Insights in Motivational Sports Speeches Using | Tonya Sims | 24m |
A pipeline transcribes pre‑recorded sports speeches with the Deepgram Python SDK, extracts summaries, topics, and sentiment, and visualizes word clouds and… | ||
| Navigating Career Adjustments in Times of Uncertainty | Jose Mesa | 28m |
Pandemic-driven layoffs push people into new roles and the concrete ways to move matter: use open source tools like OpenFOAM and Python, build API wrappers to… | ||
| Use pandas in tidy style | Srikanth | 09m |
A new Python package wraps pandas to give a tidyverse-like API: a tidyframe class that is a thin wrapper over pandas, an accessor to call tidy methods on… | ||
| How to format strings for logging in Python | Lutz Ostkamp | 09m |
Claim: f-strings are fastest when messages are emitted but percent-style logging defers formatting when a message is suppressed. Timing examples show ~890 µs… | ||
| Dask Powering lower end PC | Kefentse Mothusi | 04m |
A brief report claims Dask makes it possible to process datasets larger than memory on modest hardware by replacing eager loads with lazy evaluation and… | ||
| Bessel's Correction- Effects of (n-1) as the denominator in Standard Deviation | Saradindu Sengupta | 07m |
Claims that using n−1 (Bessel correction) yields an unbiased estimator for sample variance and standard deviation especially when samples are small… | ||
| Robyn - An asynchronous Python web framework with a Rust runtime | Shivay Lamba | 10m |
Robyn claims to be an asynchronous Python web framework that compiles Python handlers into a Rust runtime and runs them on a native thread pool, yielding a… | ||
| Tour of UnionML: An Open Source framework for Building Machine Learn Microservices | Shivay Lamba | 08m |
Claims an open source MLOps framework that reduces boilerplate and unifies tooling into one app and shows it actually builds and serves a working microservice… | ||
| Introducing Tilburg Science Hub | Banerjee, Sudhaharan | 07m |
Tilburg Science Hub claims to be a one‑stop open science platform with three concrete pillars: building blocks, tutorials, and examples. It provides… | ||
| Think Inside the Box(es) Excel-Hosted Dashboards With Python Graphics | Ted Conway | 11m |
Shows a reproducible pattern: Python code runs from Excel cells, a macro exposes Excel width and Excel height variables, Python renders plots and writes static… | ||
| Utilizing Word Embeddings and Gradient Boosting | Aadit Kapoor | 10m |
Combines log word embeddings with gradient boosting on 23 sensor columns to consolidate unstructured logs and tabular streams from dental milling machines into… | ||
| Deep Into the Tweet | Dina Bavli | 19m |
Detailed how-to for harvesting and parsing Twitter data with tweepy and Python, extracting JSON fields (full_text, ids, timestamps, user metadata, hashtags… | ||
| Bayesian Optimization- Fundamentals, Implementation, and Practice | Nguyen Hong Quan | 28m |
Explains concrete claims: Bayesian optimization finds global optima with far fewer expensive evaluations by combining a Gaussian process surrogate with… | ||
| I hate writing tests, that's why I use Hypothesis | Cheuk Ting Ho | 27m |
Argues that Hypothesis replaces hand-picked examples with property-based testing that generates inputs and finds edge cases, using pytest integration via the… | ||
| Keynote – Pia Mancini | Pia Mancini | 30m |
Open Collective builds a platform plus fiscal hosts so open source projects can receive and hold money without becoming companies, with 3,343 projects… | ||
| Classification Through Regression:Unlock the True Potential of Your Labels | Topaz Gilad | 28m |
Argues that recasting some classification tasks as regression or ordinal problems measurably lowers risk and overconfidence: implements bilinear cross entropy… | ||
| Reproducible Publications with Python and Quarto | Tom Mock | 30m |
Quarto claims a reproducible publishing pipeline that actually executes Python via IPython/Jupyter kernels, keeps a live kernel preview to avoid repeated… | ||
| Developing Battery Materials with Python | Gabriel Birnbaum | 25m |
Shows how density functional theory produces quantitative energies used to predict battery metrics, gives the explicit equilibrium‑voltage formula requiring… | ||
| Speed up Python data processing with vectorization | Itamar Turner Trauring | 29m |
Vectorization means using array-backed APIs to bypass per-item Python overhead and get order-of-magnitude speedups. Using NumPy and Pandas, a Python for loop… | ||
| A dive into time series for the energy sector | Rosana de Oliveira Gomes | 29m |
Covers concrete time‑series workflows for energy: imputation strategies tied to resolution and holidays, decomposition into trend/seasonality to spot… | ||
| Media Mix Modeling:How to Measure the Effectiveness of Advertising | Hajime Takeda | 31m |
Explains a Bayesian-style marketing mix model implemented in Python and demonstrates lightweight_mmm on four years of weekly data with seven media channels… | ||
| Scalable Feature Engineering with Hamilton | Elijah ben Izzy, Stefan Krawczyk | 29m |
Hamilton is an open source Python framework that turns each feature into a function so a DAG of dependencies is inferred and the driver handles execution… | ||
| Probabilistic demand forecasting at scale | Hagop Dippel | 29m |
Shows how probabilistic forecasts replace blind point estimates by outputting quantiles or learned distribution parameters, trained with pinball loss or… | ||
| What-if- Causal reasoning meets Bayesian Inference | Benjamin Vincent | 36m |
A Python package called CausalPy package implements quasi‑experimental designs and can fit either scikit‑learn estimators or PyMC Bayesian models; the demo… | ||
| A Practical Approach To Unlock Value From Data and Analytics | Maria Feria | 24m |
Claims concrete numbers and fixes: only 5% of data is actually used and 97% of data engineers report burnout, so the focus is on measurable change not hype… | ||
| Mixing art with Python- an introduction to Style Transfer | Isac Moura Gomes | 25m |
Concrete pipeline: load content and style images, use VGG19 to extract layer activations, compute Gram matrices for style, set weights and run optimization… | ||
| Is it possible to have entities within entities within entities? | Victoria Slocum | 24m |
spaCy’s open-source SpanCategorizer claims it can extract overlapping and arbitrary spans that NER cannot, replacing a brittle NER+text-classification pipeline… | ||
| Implementation and analysis of deep learning models for speech classification | Yashasvi Misra | 22m |
Data set is assembled from recordings and TV audio into 25 hours, 5,000 code-switched samples and 2,000 English plus 2,000 Hindi 10s clips. Pipeline extracts… | ||
| ML Model Traceability and Reproducibility by Design | Basak Eskili, Maria Vechtomova | 29m |
Claims a reproducible deployment pipeline that ties code, runs, artifacts and serving together using GitHub Actions to pass git SHA, MLflow tags with… | ||
| Building an ML Application Platform from the Ground Up | Sean Sheng | 24m |
Claims that BentoML lets teams separate runtimes for IO and inference so preprocessing runs on memory‑heavy nodes and model inference runs on GPU or CPU nodes… | ||
| Leading machine learning research journeys with happy (into-production) endings | Topaz Gilad | 28m |
Shows concrete playbooks and outcomes: a team ships live‑streaming AI into production using test-driven machine learning (API-first blocks, unit tests… | ||
| Testing Pandas- Shoots, leaves, and garbage! | Matt Harrison | 30m |
Shows concrete, runnable steps to make Pandas pipelines testable: a chained monthly_summary function on Alta ski data that shrinks ~14,000 rows to ~2,000 after… | ||
| Daft - The Distributed Python Dataframe for Complex Data | Jay Chia, Sammy Sidhu | 23m |
Daft claims a Python-native distributed dataframe that handles images video and other non-tabular types, using Apache Arrow Parquet and Iceberg for… | ||
| Data pipelines != workflows: orchestrating data with Dagster | Sandy Ryza | 22m |
Claims that Dagster replaces workflow-focused Airflow by treating pipelines as graphs of persistent data assets rather than task DAGs and shows concrete… | ||
| Better Python Coding with Prefect Blocks | Jeff Hale | 28m |
Claims that Prefect blocks are sharable Python objects that store reusable configuration on a server and can be created via UI or Python code or with pip… | ||
| Machine Learning in the Warehouse with Python | Allan Campopiano | 27m |
Demonstrates that Snowpark for Python lets notebooks prototype, train, save and run models without exporting data: code registers a local function and a… | ||
| Improving production workflows for scikit-learn models with skops | Merve Noyan | 26m |
skops provides a reproducible repo layout, a zipped skops format that avoids pickle by serializing numpy/scipy/scikit‑learn objects, and a programmatic model… | ||
| Creating behavioral profiles of your customer | Dominika Basaj, Barbara Rychalska | 27m |
Cleora is an open source graph-embedding tool that converts event streams into compact vectors using simple iterative matrix multiplication on a Markov… | ||
| Practical MLOps for better models | Isabel Zimmerman | 27m |
Claim: vetiver actually versions, deploys, and monitors models end to end with concrete tooling. It uses vetiver model objects pinned into pins boards, stores… | ||
| Extending Awkward Array into the broader PyData Ecosystem | Doug Davis | 29m |
Provides native support for Awkward Array inside PyData by shipping dask-awkward and awkward-pandas. dask-awkward exposes a partitioned Awkward Array that… | ||
| Things I learned running neural networks on microcontrollers | Saradindu Sengupta | 21m |
Shows concrete decisions: deploy on edge for very low latency, privacy, poor networks, or high throughput; use federated training if data must never leave… | ||
| Metadata Systems for End-to-End Data & Machine Learning | Alejandro Saucedo | 29m |
Claims that metadata must describe entire ML systems not just model artifacts and shows concrete work: adoption of the open inference protocol and an open… | ||
| Industrial Strength DALLE-E:Scaling Complex Large Text & Image Models | Alejandro Saucedo | 28m |
Demo shows a reproducible pipeline that productionises a DALL-E image model hooked to GPT2, with runnable code, Docker files and a Kubernetes manifest. It uses… | ||
| How to Properly Test ML Models & Data | Shir Chorev | 30m |
Claims continuous validation across training to deployment and a working open source stack that automates checks for data, splits and model behavior. Concrete… | ||
| Applying contrastive learning to fight anisotropy in language models | Aleksander Molak | 26m |
Claim: applying SimCSE (contrastive objective) with dropout augmentation turns cone-like BERT/Roberta embeddings into a more spherical, isotropic distribution… | ||
| Maps, Maps, Maps! | Geir Arne Hjelle | 30m |
Live demos show creating interactive web maps with folium (leaflet wrapper), centering and saving maps as HTML and adding markers; building a geopandas… | ||
| Building Large-scale Localized Language Models | Miguel Martínez, Meriem Bendris | 31m |
A project trains 1.7 billion parameter models for Spanish and French from OSCAR, cleans and deduplicates 282GB to 117GB, tokenizes in ~45 minutes, and converts… | ||
| Responsible AI-What, Why, How and Future! | Sonal Kukreja | 26m |
Starts from the concrete: recounts the Microsoft Tay failure as an explicit cautionary incident, cites tools actually used like IBM Watson Studio and AI… | ||
| Using embedded SQL engines for plotting massive datasets on a laptop | Eduardo Blancas | 27m |
Claims that a laptop can plot larger‑than‑memory data by pushing aggregations and percentiles into SQLite or DuckDB, computing boxplot percentiles and… | ||
| Everything you need to know about Transformer Models | Mike Rothenhäusler | 28m |
Explains the actual mechanism: numpy implementation of scaled dot product attention and how queries, keys, and values are learned and computed; shows that… | ||
| Keynote – Quincy Larson | Quincy Larson | 01h 00m |
Claims that freecodecamp.org records the equivalent of 8,000 years of learner time in 2022, that it has 200,000 certified alumni (1M followers), and that it… | ||
| Scale Data Science by Pandas API on Spark | Xinrong Meng, Takuya Ueshin | 25m |
Shows how Pandas API on Spark runs real pandas code on Apache Spark, explains the internal frame plus Catalyst optimizer rules used, and reports benchmarks… | ||
| Vaex: the perfect DataFrame Library for Python Apps | Jovan Veljanoski, Maarten Breddels | 26m |
Vaex claims real interactivity on a single node for datasets larger than RAM — memory‑mapped columns, C++ kernels and lazy evaluation let it stream a 350 GB… | ||
| 100x Faster NetworkX: Dispatching to GraphBLAS | Jim Kitchen, Erik Welch | 27m |
NetworkX adds an opt in plugin layer that dispatches decorated algorithms to compiled back ends so Python code runs orders of magnitude faster. It wires via a… | ||
| Keynote: Gabriela de Queiroz | Gabriela de Queiroz | 50m |
Community work focuses on concrete offerings and measurable impact: GitHub Codespaces runs notebooks in the browser with reproducible environments and GitHub… | ||
| How to Eliminate the I-O Bottleneck & Continuously Feed the GPU While Training in the Cloud | Lu Qiu | 24m |
Claim: a caching layer called Alessio mounts S3 Azure and HDFS and streams data so GPUs never wait. Implementation uses the POSIX-like process API to load and… | ||
| MLOps for the rest of us- A poor man's guide to putting models in production | Duarte Carmo | 26m |
Argues small teams can reliably ship models without a full ML platform, showing concrete tools and numbers: only 22% of companies have models in production… | ||
| Testing Big Data Applications (Spark, Dask, and Ray) | Han Wang | 34m |
Compares Spark Dask and Ray testability, shows a reusable pattern: separate core computing logic from orchestration and cluster setup, unit-test pandas or… | ||
| Keynote: DJ Patil | DJ Patil | 58m |
Memo dated March 14 2020 describes a quick playbook: acquire messy tests and hospital censuses, clean and join them, and run a Johns Hopkins epidemiological… | ||
| The Pythonic Common Chemical Universe | Suliman Sharif | 29m |
Claims an open knowledge graph called Global-Chem (pip-installable) that indexes common chemical lists from Enamine (20B), ZINC (2B) and PubMed (68M) and… | ||
| Steering a data science project | Morgane Mahaud | 23m |
Checklist named Portillo guides project direction with concrete choices: an autoencoder embeddings pipeline creates image vectors and sequences, a user… | ||
| Converting sentence-transformers models to a single tensorflow graph | Georgios Balikas | 21m |
Converts sentence-transformers (example all-miniLM-L6-v2) into a single TensorFlow SavedModel that accepts raw strings, folding tokenizer, padding, model and… | ||
| Lessons Learned Building Our Own Dashboard Solution | Dix et al | 27m |
An interactive public dashboard now monitors data quality for 2.2 million LEIs and detects daily outliers that monthly reports missed, built over three years… | ||
| Machine Learning Frameworks Interoperability | Christian Hundt, Miguel Martínez | 26m |
Pipeline reads CSV straight into cuDF on the GPU, uses CuPy FFT and a Numba/CUDA kernel to segment and embed heartbeats, then trains a PyTorch VAE on 100k… | ||
| Exploring Feature Redundancy and Synergy with FACET 2.0 | Jan Ittner, Mateusz Sokół | 29m |
FACET 2.0 implements SHAP vector projection to turn per-instance SHAP values into feature vectors, then uses angles and an orthogonal decomposition to quantify… | ||
| How to maximally parallelize the entire pandas API | Rehan Durrani | 24m |
UC Berkeley research yields Modin, a distributed drop-in replacement for pandas with a VLDB paper describing a novel parallelization scheme, lazy metadata… | ||
| Super Search with OpenSearch and Python | Laysa Uchoa | 28m |
OpenSearch is an Elasticsearch fork that the demo uses with the Python client to index ~220k recipe documents and run concrete queries against a live cluster… | ||
| Why we do ML model retraining wrong, and how to do better | Emeli Dral | 28m |
Machine learning models degrade over time so the concrete proposal is: monitor model quality and error distributions, run per-step data checks inside… | ||
| Supercharging your pandas workflows with Modin | Alejandro Herrera | 23m |
Claim: Modin is a drop-in replacement for pandas that you pip install and run with the same syntax. Benchmarks show concrete gains: 3x faster S3 reads, 2x… | ||
| Hypothesis Testing Stop Criterion with “Precision Is The Goal” | Eyal Kazin | 30m |
Claim: Bayesian HDI with ROPE does not magically fix sequential stopping bias; simulations with Bernoulli coin flips and Python notebooks show HDI+ROPE can… | ||
| ML in Production–What does “Production” even mean? | Dean Pleban | 25m |
Defines production as model predictions directly affecting a product or users and notes the famous 87% stat probably wrong; breaks deployment into four… | ||
| BastionAI: Towards an Easy-to-use Privacy-preserving Deep Learning Framework | Daniel Huynh | 24m |
Claims an easy-to-use privacy-preserving deep learning framework that runs models inside secure enclaves + Differential Privacy and reports concrete results… | ||
| Object Detection with KerasCV | Lucas Wood | 22m |
KerasCV provides a full object detection pipeline that trains a RetinaNet on custom datasets, with native support for ragged bounding boxes, TPU‑compatible… | ||
| Revolutionizing the Big Data Age With Compute over Data | David Aronchick | 28m |
Claims a usable project called Bacalhau that runs Docker/WASM jobs where the data lives, cutting the need to move 100 terabytes across networks. Demonstrates… | ||
| Quokka-Rewriting SparkSQL in Python | Ziheng Wang | 08m |
Quokka is a distributed Python SQL engine that executes a DataFrame/SQL flavor API and recovers from worker loss so queries survive spot instance… | ||
| Everything That You Wanted To Know About P-Values But Were Afraid To Ask | Eyal Kazin | 11m |
Claim: p-values have no acceptance signal. Evidence: simulated thousands of binomial trials with 256 tosses per experiment and binned p-values, showing a flat… | ||
| Introducing Meadowrun: write code locally, run on the cloud | Kurt Schelfthout, Richard Lee | 10m |
Open source Meadowrun claims to let local Python code run on cloud VMs by pip install meadowrun, auto-creating EC2/Azure/Kubernetes compute, mirroring… | ||
| It might look normal but this distribution will ruin your stats | Allan Campopiano | 08m |
Contaminated normals are mixtures that add heavy tails and a few outliers and they cut statistical power dramatically — the demo shows 90% vs 20% power between… | ||
| an introduction to approximate nearest neighbors | Kacper Łukawski | 10m |
Explains HNSW (hierarchical navigable small world) as a layered graph index that finds neighbors by greedy descent, claiming sublinear search via layered… | ||
| OpenTeams Score: A Way to Assess Open Source Projects | Fatma Tarlaci, Brian Skinn, Dale Tovar | 07m |
OpenTeams Score assesses open source projects using five concrete components: popularity, community, security, license permissiveness and versioning. It pulls… | ||
| Lessons learned from using the C Foreign Function Interface | Caroline Arnold | 06m |
Build compiles Python ML into a dynamic library and links it into Fortran via cffi, uses buffer sharing for arrays, and replaces one Fortran routine with a… | ||
| Running Apache Airflow at Scale | Jean-Martin Archer, Michael Petro | 25m |
Shopify runs Apache Airflow with 12,000 DAGs and about 400,000 tasks per day, and describes concrete fixes: a hybrid GCS+mounted filestore sync to cut DAG… | ||
| Single node shared memory comes to dask | Martin Durant | 26m |
A Dask fork copies pickled buffers into shared memory so processes map the same bytes, demonstrating that scatter/replicate reduces memory from six workers to… | ||
| Parallelization of code in Python for beginners | Cheryl Roberts | 30m |
Shows concrete recipes: use joblib with a number-of-workers ≈ physical CPU cores and tune chunk size, prefer multiprocessing for blackbox Python code, beware… | ||
| On copies and views: updating pandas' internals | Joris Van den Bossche | 27m |
Proposal simplifies pandas’ copy/view rules so any derived DataFrame or Series behaves as a copy while under‑the‑hood uses views until mutation to save memory… | ||
| Create text classifiers in a few hours using the open-source Label Sleuth system | Katsis, Shnarch | 31m |
Claims a usable Label Sleuth no‑code system that combines an intuitive UI, active learning and automated background model training so domain experts build text… | ||
| Deploying Dask | Matthew Rocklin | 28m |
Requesting 20 machines with two cores each on AWS provisions T3 mediums in 20–30 seconds, pip/conda packages and local Git code get installed, and the Optuna +… | ||
| Trying No GIL on Scientific Programming | Cheuk Ting Ho | 28m |
Claim: nogil fork on Python 3.9 can remove the GIL and speed CPU‑bound scientific code. Experiment uses scikit‑learn, NumPy and SciPy examples, C extensions… | ||
| Crowd-Kit: A Scikit-Learn for Crowdsourced Annotations | Evgeniya | 26m |
Open-source Crowd-Kit claims to be a scikit-learn–style quality control library for crowdsourced labels, implemented as pandas-friendly estimators with… | ||
| A Gentle Introduction To Causal Inference | Eyal Kazin | 31m |
Claims that causal relationships are recoverable from observational data without running RCTs, demonstrates with a 10,000‑patient example where the population… | ||
| Data Storytelling through Visualization | Marysia Winkels | 30m |
Argues that visualization must move from analysis to action and demonstrates concrete choices that change conclusions: used basic matplotlib and mentions… | ||
| A Practitioner’s Guide to Opening the AI Black Box for Humans | Kilian Kluge | 28m |
Argues that there are no true black boxes and promotes user‑centric XAI that prioritizes technical accuracy plus human meaningfulness. Demonstrates concrete… | ||
| Detecting anomalous sequences using text processing methods | Liron Faybish | 29m |
Claims an architecture that converts sequences to vectors with Word2vec + RNN autoencoder or sentence vectors from Bert + Autoencoder and flags anomalies by… | ||
| Explaining Why You have a Favorite Cereal | Gatha | 30m |
Linear regression on a cereal ratings dataset (sourced from kegel.com) yields decent fit but interpretation matters: the model uses raw coefficients that… | ||
| Interpretable and realistic generative models in data science? | Narendra Mukherjee | 28m |
A marketplace agent‑based simulator models eight interpretable parameters per product (costs, herding, five‑star bias) and applies likelihood-free Bayesian… | ||
| Data Validation for Feature pipelines: Using Great Expectations and Hopsworks | Moritz Meister | 28m |
Feature pipelines validate at write time into a feature store by attaching an expectation suite to a Hopsworks feature group so data is checked before… | ||
| ARCH-GARCH Models Tour | Kalyan Prasad | 29m |
Overview gives concrete choices: uses Python with pandas-datareader to pull Tesla 2015–2020 returns, fits volatility models with the arch package, initially… | ||
| Managing Python Dependencies at scale | Jarek Potiuk | 28m |
Apache Airflow manages ~80 packages (about 630–650 transitive dependencies, 70+ provider packages) by generating a golden set of pinned dependencies and… | ||
| The Dask at Hand: Using Dask to Speed up the High-Quality Transit Areas dataset | Tiffany Chu | 20m |
Builds a reproducible pipeline that derives California’s legally defined high quality transit corridors from GTFS and publishes them on the CA open data portal… | ||
| The 10 commandments of reliable data science | Isaac Slavitt | 30m |
Ten concrete rules: start projects with a repeatable layout (example: cookiecutter-data-science), keep raw extracts immutable, run the DAG from raw input to… | ||
| Data Prep for Graphs | Paco Nathan | 27m |
Shows concrete data‑prep steps: deduplication at scale with Datasketch and nearest‑neighbor tricks, numeric ID assignment with Keyvi/Redis, cycle detection and… | ||
| Using feedback loops to tune predictive models in a video ad marketplace | Emily Hopper | 25m |
Combines predictive models with PID controllers and cascading feedback loops to tune decision thresholds using historical completion rates plus real‑time… | ||
| Urdu poems to Shakespearean English: Machine Translation | Sidra Effendi | 19m |
Claims an automated pipeline that converts Urdu poetry into Shakespearean English by chaining models: Urdu→Modern English via Helsinki trained on ~14,371 Bible… | ||
| Reactive data processing in Python | Adrian Kosowski | 28m |
Pathway provides a Python reactive framework that keeps batch-style code but applies incremental computation so models and predictions update as new data… | ||
| Embracing multi-lingual data science | Hadley Wickham | 59m |
Posit renames RStudio and ships concrete multilingual tooling: tidyverse code uses dplyr and ggplot2 to process 300,000 rows of flight data, computing speed… | ||
| Text to Data: Make Your Code Malleable, Not Brittle | Martha L Escobar-Molano, David Barrett | 26m |
A concrete extractor that tags text then runs a three‑step pipeline of tuple relational calculus queries to select, transform and project locations is… | ||
| Mischief Managed: What hackers can do on your Jupyter instance | Joseph Lucas | 25m |
Jupyter instances expose code execution, defaulting to HTTP on port 888 with an authentication token that becomes a cookie and lasts 30 days, and shells and… | ||
| PyScript and Data Science: a love story | Fabio Pliger | 29m |
PyScript claims to run Python in the browser and brings numpy, pandas, matplotlib, scikit‑learn and seaborn into client side demos, using Pyodide (packages… | ||
| An Evolving Jupyter Notebook | Rosio Reyes | 27m |
Claim: Notebook 7 replaces the old front end with Notebook 7 uses JupyterLab components and runs on Jupyter Server, so pip install notebook will soon pull a… | ||
| Algorithms at Scale: Raising Awareness on Latent Inequities in Our Data | Lalitha Krishnamoorthy | 21m |
Claims that >90% of businesses have AI embedded and that AI drives revenue, cost and profitability are supported with a concrete pricing/IoT/edge use case and… | ||
| Keynote: Ada Nduka Oyom | Ada Nduka Oyom | 44m |
Community-building claims get concrete here: a volunteer core team grows She Code Africa to 18,000 members, hits milestones like trained 1,000 women, and spins… | ||
| Data Science Project Patterns that Work | Ian Ozsvald | 28m |
Practical patterns for the five project stages: choose good projects, de‑risk data by talking to users and using schemas and profilers, do productive research… | ||
| The Beauty of Zarr | Sanket Verma | 28m |
Zarr is an open specification and Python implementation for chunked, compressed N‑dimensional arrays that actually stores chunks as key/value blobs (default… | ||
| Creating custom engines for flexible notebook execution | Eduardo Blancas | 28m |
Shows how to build and register papermill engines as Python entry points, with an execute method that replaces the default subprocess model. Demonstrates a… | ||
| Knowing what you don’t know matters- Uncertainty-aware model rating | Malte Tichy | 30m |
Claims a concrete uncertainty-aware model rating that scores probabilistic forecasts against an ideal Poisson baseline using Python tooling and M5 data. Method… | ||
| Supercharge your training on TPUs | Kaushik Bokka | 28m |
Claims end-to-end TPU training with concrete tools and commands: shows using PyTorch XLA tensors, xm.mark_step and spawn to run a single core versus an… | ||
| Data-Centric AI Cookbook-let's prep that data | Marysia Winkels | 30m |
Shows that tiny concrete fixes beat vague promises: a penguin dataset of ~5000 images contains unit errors (cm vs mm) that break models, imputing missing… | ||
| Inequality Joins in Pandas with Pyjanitor | Samuel Oranyeli | 27m |
pyjanitor adds a conditional_join to pandas that implements inequality joins via sorting and binary search so the code actually reduces search space instead of… | ||
| Expressive & fast dataframes in Python with polars | Juan Luis Cano Rodríguez | 28m |
Polars is a Rust + Arrow dataframe that claims much faster, lower‑memory ops than pandas and demonstrates it on a 1.8 GB StackOverflow CSV (~1.26M question… | ||
| Measurement of Trust in AI | Shashank Shekhar | 31m |
A concrete framework claims trust = four pillars (reliability safety transparency responsibility and accountability) mapped to performance fairness… | ||
| Generate Actionable Counterfactuals w/ Multi-objective Particle Swarm Optimization | Niranjan G. S. | 30m |
A simple counterfactual generator uses multi-objective particle swarm optimization with a mixed discrete PSO variant, box constraints and a diversity… | ||
| Bayesian Decision Analysis [Tutorial] | Allen Downey | 01h 28m |
Hands-on walkthrough of Bayesian decision analysis with runnable Jupyter notebooks (Colab), using NumPy, SciPy and Pandas to represent PMFs and implement… | ||
| The State of the Art for Probabilistic Programming | Thomas Wiecki | 47m |
Survey of concrete advances and demos: PyMC moves from monolith to layered stack, compiling to PyTensor and BlackJAX or Rust samplers; examples show JAX and a… | ||
| nothing matches | ||