Zero-shot time series forecasting in R

We’re excited to present nixtlar v0.6.1, our R package for working with our TimeGPT API. As the R equivalent to nixtla, the Python SDK for TimeGPT, nixtlar was already available on CRAN and has now been updated to version 0.6.1. This new version incorporates the new TimeGPT API-v2 and extends support to tibbles.

TimeGPT is the first foundation model for time series forecasting. With TimeGPT you can quickly and easily generate accurate forecasts or conduct anomaly detection in just a few lines of code. nixtlar is designed to be user-friendly, allowing anyone within the R community to leverage generative AI for time series.

Zero-shot forecasting in R

By bypassing the need for training on specific datasets, zero-shot methods save time and computational resources, and allow you to make predictions when you have less data available. This approach complements existing traditional forecasting methods and also lowers the barrier to entry for people new to forecasting.

Capabilities of nixtlar include:

• Zero-shot inference: Generate accurate forecasts with no prior model training.
• Fine-tuning: Enhance your predictions by fine-tuning the model to your specific dataset.
• Uncertainty quantification: Capture the full distribution of your predictions, either via quantile forecasts or prediction intervals.
• R ecosystem integration: Works with your favorite R data structures—data frames, tibbles, or tsibbles.

You can make use of all the TimeGPT capabilities including forecasting multiple time series simultaneously, anomaly detection to identify unexpected changes in your data, cross -validation for model evaluation, historical and future exogenous variables to improve your predictions and custom loss functions for tailored model performance.

Get started

Whether you're just starting with time series forecasting or an expert who wants to explore options, get started with:

✅ Visit the nixtlar Getting Started Guide

📒 Get your TimeGPT API key

🚀 Install nixtlar from CRAN with 'install.packages("nixtlar")'

📈 Run your first forecast!

👀 You can see the whole process in Mariana Menchero's Getting Started with nixtlar video

To help you make the most of nixtlar, the package includes comprehensive vignettes that document its capabilities and features.

nixtlar embodies Nixtla’s core philosophy to democratize access to state-of-the-art predictive insights, making advanced forecasting tools accessible to all.

And of course, nixtlar already has its own hex sticker! Thank you for all the ideas and feedback from the R community that have contributed to nixtlar!

‍