sktime

Apply for a full-time Open Source AI developer job with sktime! We are partnering with ecoSPECS GmbH and the German Center for Open Source AI (GC.OS) to build the next generation of open, modular, and transparent AI tools for time series analysis and forecasting - embedded with the BioPharma Cluster South Germany, world-leading manufacturing champions, and energy companies. Multiple roles are available. If you have experience with Python, machine learning, time series analysis, or open-source engineering, and you want to make a real impact across pharma, life sciences, and energy industries, apply today! Apply with your CV, GitHub profile (or notable open-source contributions), and a short note on why you’re excited about this role: https://lnkd.in/e9RW48AN (applications open on rolling basis at least until Nov 30, 2025; relocation packages are available for non-EU applicants) As active contributors to the open-source community, we believe in advancing scientific and engineering excellence through collaboration, direct engagement with end users, and openness. We are looking for multiple Open Source AI Engineers to help lead development and innovation around sktime - the open-source Python library for AI with time series - and the GC.OS software stack. You'll work at the intersection of software engineering, industry applications, and community development, helping shape the future of time series AI. Apply today! #sktime #opensource #python #timeseries #ai

PyTorch Tabular joins the German Center for Open Source AI! We’re thrilled to welcome and support PyTorch Tabular, the powerful deep learning library for tabular data built on top of PyTorch and PyTorch Lightning. Created by Manu Joseph, PyTorch Tabular brings deep learning to tabular datasets, offering a unified, easy-to-use interface for practitioners - making it simple to experiment, compare, and deploy advanced architectures for tabular data. PyTorch Tabular is fully open source under a permissive license, helping make deep learning more accessible and interpretable for real-world data. Since its inception, PyTorch Tabular has gained growing adoption and contributions worldwide, becoming a go-to toolkit for deep learning on tabular data in Python. Explore PyTorch Tabular, contribute to the project, or dive into its tutorials and documentation! GitHub: https://lnkd.in/ep9iMcj And stay tuned for upcoming collaborations with pytorch-forecasting and state-of-the-art libraries for tabular data! #torch #opensource #python #ai #ki #gcos

🔥🌍🔥 Breaking news from the world of open source automation! How Microsoft’s GHA default CI settings for forks are single-handedly causing global warming🔥🌍🔥 Our developer Nirbhai S. (absolute legend 🕵️♂️) just discovered that our monthly testing CRON jobs from the sktime head repository GHA are automatically running on every fork of the open source project - that’s roughly 1,7 thousand forks, all happily ticking away in the cloud. The worst part being, no one even looks at or needs these CRON jobs in the forks – and since public repositories run CI for free, no one seems to have taken notice either. 💸 Estimated energy cost: about $1,500/month in data center compute – PER FORK. Multiplied with 1,7k forks this is ca $2,600,000 million / month. 🔥Multiply this with the number of open source projects out there with significant footprint and a costly CRON job ... you see where this is going. 🔥🌍🔥 Estimated impact: somewhere between “mildly wasteful” and “Microsoft’s default fork CI settings are single-handedly causing global warming.” Yes, you read that right - this looks like it is the default GitHub Actions behavior. Every time someone forks your repo, your scheduled jobs may go on a little adventure of their own. 😅 We love automation as much as the next data scientist, but maybe not 17 hundred copies of it. Dear Microsoft/GitHub: maybe let’s keep the planetary-scale compute (or should we say: warming) experiments opt-in next time? 💚 Big thanks again to Nirbhai S. for spotting this - you may have just saved Microsoft billions of compute per year, a couple of Swiss glaciers, the ozone layer, and all polar bears. 🕵️♂️🙌 Microsoft, if you are reading this: we are a non-profit supporting open source worldwide (and sustainable use of AI). Donations are appreciated https://lnkd.in/eJ4uP9yD - we can issue tax deductible donation receipts. #OpenSource #GitHub #DevOps #Sustainability #Python #DataScience #AI #KI #sktime

Open-Source AI - Franz Király | E18 Watch now: https://lnkd.in/e8QyPW3s I talked to Franz Király, founder of the German Center for Open Source AI, to explore one of the most urgent questions of our time: Who’s shaping the direction of AI - and what does it mean for democracy?

sktime meetup October 24, 13 UTC – Shuvam Das (International Monetary Fund): macroframe-forecast - A Python Package to Assist Macroframework Forecasting Location: sktime discord (virtual), meet-up-presentations stage https://lnkd.in/e_jhgKnh In forecasting economic time series, statistical models often need to be complemented with a process to impose various constraints in a smooth manner. Systematically imposing constraints and retaining smoothness are important but challenging. The Python package macroframe-forecast provides a systematic way to incorporate such constraints into forecasts. It allows users to generate forecasts that are both smooth and aligned with user-specified targets. We demonstrate the package’s functionality with two examples about forecasting US GDP and fiscal variables. pip install macroframe-forecast GitHub: https://lnkd.in/esu-2J6S #sktime #opensource #python #timeseries #ai #ml #datascience

sktime meetup October 17, 13 UTC – Manu Joseph (Walmart Global Tech): PyTorch Tabular - Leveraging the Power of Deep Learning for Tabular Data Location: sktime discord (virtual), meet-up-presentations stage https://lnkd.in/e_jhgKnh Join us for an exciting session on PyTorch Tabular, a powerful library developed by Manu Joseph that brings the strengths of deep learning to tabular data. In a field traditionally dominated by methods like gradient boosting, PyTorch Tabular simplifies the implementation of state-of-the-art models with a user-friendly, scikit-learn-like interface. From data preprocessing to training and inference, this framework offers robust features like categorical encoders and experiment tracking with Tensorboard and Weights & Biases. Learn how PyTorch Tabular bridges the gap between traditional methods and modern deep learning, supporting models like TabNet, FTTransformer, and GANDALF. GitHub: https://lnkd.in/ep9iMcj #sktime #opensource #python #timeseries #ai #ml #datascience

🚀 𝗕𝗿𝗮𝗻𝗱 𝗔𝘄𝗮𝗿𝗲𝗻𝗲𝘀𝘀, 𝗟𝗮𝘁𝗲𝗻𝘁 𝗖𝗼𝗺𝗽𝗼𝗻𝗲𝗻𝘁𝘀, 𝗮𝗻𝗱 𝗣𝗿𝗼𝘅𝘆 𝗩𝗮𝗿𝗶𝗮𝗯𝗹𝗲𝘀 in MMM — now in Prophetverse 0.10.0! (𝘋𝘰𝘤𝘶𝘮𝘦𝘯𝘵𝘢𝘵𝘪𝘰𝘯 𝘭𝘪𝘯𝘬 𝘪𝘯 𝘤𝘰𝘮𝘮𝘦𝘯𝘵𝘴) The new release introduces 𝗯𝗿𝗲𝗮𝗸𝘁𝗵𝗿𝗼𝘂𝗴𝗵 𝗶𝗻𝗻𝗼𝘃𝗮𝘁𝗶𝗼𝗻𝘀 in Marketing Mix Modeling: Prophetverse can now handle 𝗹𝗮𝘁𝗲𝗻𝘁 𝗰𝗼𝗺𝗽𝗼𝗻𝗲𝗻𝘁𝘀 — such as brand awareness — and use 𝗽𝗿𝗼𝘅𝘆 𝘃𝗮𝗿𝗶𝗮𝗯𝗹𝗲𝘀  (like surveys, search trends, or social metrics) to calibrate them. 💡 Traditional MMMs often struggle with awareness because: 1️⃣ They fail to capture the joint effect between awareness investments and last-click actions. 2️⃣ The impact of awareness on sales is unobservable and hard to test, so models can’t be properly calibrated. With Prophetverse 0.10.0, you can now 𝗹𝗶𝗻𝗸 𝘀𝘂𝗿𝘃𝗲𝘆 𝗼𝗿 𝗲𝘅𝘁𝗲𝗿𝗻𝗮𝗹 𝗱𝗮𝘁𝗮 𝗱𝗶𝗿𝗲𝗰𝘁𝗹𝘆 𝘁𝗼 𝘆𝗼𝘂𝗿 𝗹𝗮𝘁𝗲𝗻𝘁 𝗮𝘄𝗮𝗿𝗲𝗻𝗲𝘀𝘀 𝗰𝗼𝗺𝗽𝗼𝗻𝗲𝗻𝘁 — adding an extra likelihood term that improves inference and prevents upper-funnel under-attribution, one of the hardest challenges in MMM. #MarketingMixModel #MMM #MarketingScience

I recently wrote my first Medium article reflecting on my journey in open source. How it shaped me as a developer, the communities that welcomed me, and the lessons I’ve picked up along the way. From nervously opening my first pull request to becoming a core developer at sktime, it’s been about growth, collaboration, and finding a place to give back. I shared the full story here: https://lnkd.in/efXFG_Ke Would love to hear your thoughts, and if you’ve been meaning to contribute to OSS, maybe this will give you a little nudge to start.

sktime 0.38.4, 0.38.5, 0.39.0 have been released! Highlights: - TiRex foundation model forecaster (Sinem Kilicdere Moschos) - Toto foundation model forecaster (Shrivaths S Nair) - prophetverse (https://lnkd.in/ejybWVfn) Bayesian hierarchical forecasters for mixed & intermittent demand (Felipe Angelim, Victor Gruselius) - hyperactive (https://lnkd.in/dXbG9S8) tuners using any tuning engine for forecasting and time series classification (Franz Király, Simon Blanke) - ConvTimeNet time series classifier, torch based (Tanuj Taneja) - residual boosting forecaster (Sanchay Singh) - Loader classes for UCR UEA classification and Monash forecasting dataset from sktime huggingface repo (Jigyasu Krishnan‎) - elementwise operation combiner for transformations (Orest Alickolli) - TinyTimeMixer exogenous data support (Akshath Mangudi) - Jarque-Bera, variance break, and Ljung-Box tests/statistics (Harshvir Sandhu) - percentage forecast metrics relative to prediction (Avaya Aggarwal) sktime changelog: https://lnkd.in/etekGzZ9 #sktime #opensource #python #timeseries #ai #ml #datascience #hyperactive #prophetverse

Hi, Linked fam. If you are in Lagos, catch me at PyCon Nigeria 2025. I'll be giving a talk on how big retailers use sktime (Python's best forecasting library) to forecast demand and how you can, too (spoiler alert: sktime is easy to use).