Next up at Nixtla? Tutorials, new Features, and what's coming next

As we kick off 2025, we’re excited to share some of the latest resources, updates, and insights from Nixtla and the forecasting field. Whether you’re looking to fine-tune your forecasts, explore innovative tools, or stay on top of the latest trends, this edition has something for you.

Let us know what you think and what you’d love to see in upcoming editions. Comment below and subscribe to stay in the loop ??

New from Nixtla

• ??Looking to enhance your time-series forecasts? Yibei Hu’s tutorial demonstrates how to improve accuracy using fine-tuning, adding exogenous variables, and long-horizon forecasting. She uses hourly electricity price data to share the context and code for implementing these approaches with TimeGPT.
• ??Persist Fine-Tuned Models – It’s Here! The wait is over – we’ve just launched the feature you've been asking for! Now, you can save your fine-tuned models and reuse them on fresh data. Curious how it works? Check out how to save your fine-tuned model and set yourself up for even more success in the future.?

• ?? Why Choose TimeGPT for Forecasting? If you’re exploring why TimeGPT stands out, this notebook is a must-read. It provides a comprehensive introduction to the key benefits and capabilities that make TimeGPT a powerful tool for forecasters.
• ?? Unpacking SHAP Values for TimeGPT. Understand how SHAP values bring transparency and interpretability to your TimeGPT forecasts. This blog dives into the mechanics and benefits of using SHAP for better decision-making.
• HierarchicalForecast v1.0.0 Released! This release introduces Polars support for seamless compatibility and a unified evaluation API aligned with utilsforecast. Note: input data must now be flat tables (no unique_id as an index). Coming soon: temporal hierarchical reconciliation methods—share your feature ideas on GitHub!
• ?? VN1 Benchmark Announced! Win second place with 5 lines of code. Explore the performance of TimeGPT in our latest VN1 benchmark results. The detailed breakdown is available for both Python and R versions. Python Version | R Version.
• ??? Podcast Feature with Max! Max Mergenthaler Canseco, Co-Founder & CEO of Nixtla, shares insights on building robust forecasting algorithms, common mistakes, and the future of foundation models like TimeGPT. Listen here.

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. ??