sktime

sktime @ European Summer of Code and AI Challenge day at the German Center for Open Source AI We are excited to announce our summer projects at the European Summer of Code. Thanks to our partners and sponsors, we already have 4 stipend slots committed towards sktime and adjacent projects! We are still looking for sponsors or partners for joint projects by March 18, with mentoring capacity available. Applications open March 26, on our mentoring & internships portal: https://lnkd.in/eH8vmeaG - further details will be communicated on our discord https://lnkd.in/e_jhgKnh If you are located in Europe, we also invite you to join the in-person AI Challenge Day and kick-off event at the German Center for Open Source AI on March 25, with keynotes by Manuel Hagel, MP BW, and Alexander Baumann, head mayor. Companies and open source projects will present AI challenges, and form integrated teams to work over the summer. Register here for the in-person event: https://lnkd.in/e9nktHkx (the challenges will also be available on-line, links via discord and LinkedIn) Funded internship opportunities will be available, also internationally, and sktime will be present with our project challenges: https://lnkd.in/etTm-MKK Looking forward to an exciting European Summer of Code - in the words of Peter Diem, from the EU anthem: North and South will work together Just as friends and neighbours should. East and West will grow together - Brotherhood and sisterhood! #opensource #ai #sktime #python #timeseries #ml

sktime 0.36.0, 0.35.1, skpro 2.9.0, pytorch-forecasting 1.3.0, have been released! Highlights: - pytorch-forecasting compatibility with python 3.13 - forecasting metrics can now be passed a by_index arg at construction to return metrics by time index (Benedikt Heidrich) - forecasting benchmark now allows to return copies of fitted models per fold (Marc Rovira) - sktime native reducers (2nd generation) now support arbitrary imputation strategies (Lena Klosik) - tags and object API for datasets (Felipe Angelim) highlights, new models: - TiDE model, pytorch-forecasting (Sohaib Ahmed) - MAPA forecaster (Aryan Saini) - interface to skforecast recursive reducer (Anirban Ray) - new detection metrics: RAND index, windowed F1-score (Gavin Katz) - K-visibility clustering algorithm (seigpe) - auto-regressive process based lag estimator (Satvik Mishra) - skpro interface to xgboostLSS probabilistic regressor by Alexander M?rz/StatMixedML (Franz Király) Changelogs: sktime - https://lnkd.in/etekGzZ9 skpro - https://lnkd.in/d9tz7f8f pytorch-forecasting - https://lnkd.in/efvVffbT #sktime #opensource #python #timeseries #ai #ml #datascience

The talk with Matt Maloney - Developing Enterprise Forecasting Solutions for City of Hope - is now available as a youtube video on our channel: https://lnkd.in/e3b76VNi

Hi Everyone! Pretty excited to start my mentorship journey at the amazing open source organization sktime under the guidance of Franz Király and other #core_devs. sktime is an amazing Python package for time series forecasting, classification, and regression. It provides a unified framework for different forecasting models, making it easier to experiment with and compare various approaches. Whether you’re working with classical statistical models or deep learning-based forecasters, sktime offers a flexible and modular design to help you build better time series solutions. A heartfelt thank you to sktime team for having me onboard! Looking forward to learning, collaborating and contributing to the open-source community!! #OpenSouce #Time_Series #Machine_Learning

The integration of deep learning and foundation models into the sktime framework represents a significant leap forward in time series analysis. Franz Király and Benedikt Heidrich delve into the progress, challenges, and newest features of sktime, highlighting how it brings a unified interface architecture to the forefront of AI. sktime, a widely used scikit-learn compatible library, is known for its extensibility and interoperability with the pydata/numfocus stack. The talk discusses the framework's evolution to include deep learning and foundation models, addressing inconsistent formal interfaces, different deep learning backends, and vendor-specific APIs. The speakers emphasize the importance of creating modular, extensible building blocks with a simple specification language, drawing inspiration from hugging face and scikit-learn. The talk also covers the technical mission of sktime, which aims to integrate the ecosystem of time series modeling behind a simple, unified interface. This includes hundreds of algorithms for tasks such as forecasting and classification, all of which follow consistent, exchangeable interfaces. The speakers discuss the challenges of integrating next-generation algorithms, including the need to handle pre-training, fine-tuning, and back-end abstractions for data containers and deep learning frameworks. sktime's commitment to open source and community-driven development is evident in its permissive license, open governance, and mentoring programs. The framework invites contributions and donations, providing opportunities for anyone worldwide to get involved. Watch the full talk here: https://lnkd.in/e6baHZ59

sktime meetup February 7, 13 UTC – Matt Maloney (City of Hope): Developing Enterprise Forecasting Solutions for City of Hope Location: sktime discord (virtual), meet-up-presentations stage https://lnkd.in/e_jhgKnh City of Hope, is a National Cancer Institute (NCI)-designated comprehensive cancer center, is one of the largest and most advanced cancer research and treatment organizations in the U.S. and its National Medical Centre named top 5 in the nation for cancer care by U.S. News & World Report. At City of Hope, Matt Maloney leads a small team of data scientists who develop applications to support business & operational workflows. Sktime has enabled the efficient development of enterprise forecasting capabilities. Matt will discuss how sktime features have enabled the application of more accurate and modern forecasting methods without the need to develop an extensive custom code base. This presentation will have a very applied focus; the goal is to share Matt’s experience with implementing a production solution using sktime. #sktime #opensource #python #timeseries #ai #ml #datascience

Excited to Share! ?? I’m thrilled to announce that I’ve been selected for the sktime's Mentorship Program! ?? This opportunity means a lot to me as it allows me to contribute to open-source development while deepening my understanding of time series forecasting. Through this program, I’ll be working with a community of incredible mentors and developers, learning how to tackle complex problems, and improving my skills in a professional setting. Looking forward to gaining valuable insights, collaborating with like-minded individuals, and making meaningful contributions to the sktime community. A big thank you to the sktime team for this opportunity.

I'm happy to share that I've become?a member of the core developer team at sktime! Since 2022, I've been an active user of the package, and it's a great honor to contribute back to the community. Sktime has always been my top choice for forecasting in Python, both in research and production workflows, thanks to its consistent interface, which has been a key factor in my productivity. It's important to emphasize that sktime is more than just a package — it's a welcoming community where experts and beginners collaborate, and where anyone can learn and enhance their statistical and programming skills. I invite you to join the Discord server and check the Github repo to connect with this environment and explore ways to contribute. Join the discord server: https://lnkd.in/dJhT_TeB Github repository: https://lnkd.in/dj9p7QNB.

Hi, I have been selected for sktime's mentorship program where I will be mentored by Dr. Franz Király and Felix Hirwa Nshuti and work on interfacing and implementing foundation models with PyTorch. I came across sktime while working on a hackathon where we needed to use time-series forecasting to predict the price of agricultural commodities in the Indian market. Earlier this month, I got my first PR merged which added the SplineTrendForecaster estimator to the library, based on the SplineTranformer from scikit-learn. I have been having a nice time with the community, thanks to Franz Király and Anirban Ray for helping me with my initial PR. I hope to make more impactful contributions to the library in the coming few months.

sktime 0.35.0, 0.34.1, skpro 2.8.0, pytorch-forecasting 1.2.0, have been released! Highlights: - Full rework of sktime detectors API, full integration with skchange, module has left "experimental" stage (Martin Tveten, Norsk Regnesentral, Alex Gregory, Franz Király, Alyssa D’Souza, Shivam Jindal, Gavin Katz, Julia Zhu, Christopher Risi, WAT.ai) - pytorch-forecasting is now compatible with numpy 2, maintenance is up-to-date (Franz Király, Benedikt Heidrich, Felix Hirwa Nshuti, Xinyu Wu, Ewan Thompson, Eugenio Mercuriali, Sai Komal Surisetti) - sktime forecasting metrics now accept callable as sample_weight for dynamic weight generation (Markus Sagen) highlights, new sktime/skpro models: - Interfaces to new neuralforecast estimators (Anirban Ray) - Box-Cox bias adjustment for forecasters (Sanskar Modi, Talat Khattatov) - SCINet forecaster (Sohaib Ahmed) - Spatio-Temporal DBSCAN clusterer (Valerii Chirkov, Rishabh Kamboj) - Signature moments transformer (Rishabh Kamboj) - Experimental pymc-facing Bayesian interface and Bayesian linear regression (Meraldo Antonio) Changelog: sktime - https://lnkd.in/etekGzZ9 skpro - https://lnkd.in/d9tz7f8f pytorch-forecasting - https://lnkd.in/efvVffbT #sktime #opensource #python #timeseries #ai #ml #datascience