Next up from Nixtla? A newsletter, new releases, and forecasting features.

The forecasting field is always looking into the future, and especially now, there’s even more going on. So, how to keep up? We’re starting this newsletter 'NextUp' to share news from Nixtla, what’s new with releases, guides and how-to’s, and what’s new in forecasting, approaches, implementations and examples.?

We hope this helps keep you up to date, learn some things, and have some fun too. We’d love to hear from you what you think and what you might like to see.?? Comment here and subscribe to follow the newsletter.

New from Nixtla

• ?? Nixtla named as one of Fast Company’s Next Big Things in Tech. Proud to share that we’ve been named to Fast Company’s fourth annual Next Big Things in Tech list, honoring emerging technology that has a profound impact for industries—from education and sustainability to robotics and artificial intelligence. We are honored to be chosen alongside other game-changing organizations for the #FCTechAwards.??
• ? Version 2 of the TimeGPT API: faster, smarter and more powerful. The nixtla API lets you seamlessly connect with TimeGPT. This new API brings speed improvements, the ability to distinguish between future and historical exogenous variables, enhanced model explainability with SHAP values and new integration with polars.
• ?? New release of NeuralForecast. New features include updates to conformal prediction, support for providing DataLoader arguments to optimize GPU usage, set activation function in GRN of TFT, and a new tutorial on cross-validation.
• ??? New release of Hierarchical Forecast. New features include sparse middle-out reconciliation via MiddleOutSparse, support for exogenous variables in utils.aggregate, efficient Schafer-Strimmer for MinT, improved residuals-based reconciliation stability and faster ma.cov. Shout out to our community members Christopher Titchen and Kurai Maingi for their contributions!
• ?? Forecasting at Scale: One Billion Time Series. Imagine you're tasked with forecasting for one billion unique time series—ranging from retail sales across thousands of stores to sensor data from millions of IoT devices. Here’s a blueprint for scaling such a task, utilizing TimeGPT and orchestrating the process efficiently using Python and AWS S3.?

Forecasting Features

• ?? VN1 Forecasting Accuracy Challenge. In Nicolas Vandeput’s datathon, participants were tasked with accurately forecasting future sales using historical sales and pricing data for different clients and warehouses. Submissions were evaluated based on accuracy and bias against actual sales figures. ?? Congratulations to all the winners!? See how the winners won and check out our own Olivier Sprangers’ example notebooks using MLForecast, NeuralForecast and StatsForecast.
• ??Time-series forecasting through recurrent topology. Taylor Chomiak & Bin Hu introduce a learning algorithm, Forecasting through Recurrent Topology (FReT) that can generate multi-step-ahead forecasts of unseen data. The simplicity of FReT could offer an attractive alternative to complex models where increased model complexity may limit interpretability/explainability and has high computational requirements.?

If you’re ready to explore the future of forecasting, you can try Nixtla today. Get started for free, no credit card required. Visit our TimeGPT dashboard to get started. ??