Introduction

TimeGPT-1: The first foundation model for time series forecasting and anomaly detection.

The nixtlar package is the R interface to TimeGPT, allowing you to perform state-of-the-art time series forecasting directly from R. TimeGPT is a production-ready, generative pretrained transformer for time series forecasting, developed by Nixtla. It is capable of accurately predicting various domains such as retail, electricity, finance, and IoT, with just a few lines of code. Additionally, it can detect anomalies in time series data.

Version 0.6.2 of nixtlar is now available on CRAN! This version introduces support for TimeGEN-1, TimeGPT optimized for Azure, along with enhanced date support, business-day frequency inference, and various bug fixes.

How to use

To learn how to use nixtlar, please refer to the documentation.

To view directly on CRAN, please use this link.

The nixtlar package requires an API key. Get yours on the Nixtla Dashboard.

Installation

# Install nixtlar from CRAN
install.packages("nixtlar")

# Then load it 
library(nixtlar)

# Set your API key
nixtla_set_api_key(api_key = "Your API key here")

Quick Example

# Load sample data
df <- nixtlar::electricity
head(df)

# Forecast the next 8 steps ahead
nixtla_client_fcst <- nixtla_client_forecast(df, h = 8, level = c(80,95))

# Optionally, plot the results
nixtla_client_plot(df, nixtla_client_fcst, max_insample_length = 200)

Anomaly Detection Example

# Detect anomalies
nixtla_client_anomalies <- nixtlar::nixtla_client_detect_anomalies(df)

# Plot with anomalies highlighted
nixtlar::nixtla_client_plot(df, nixtla_client_anomalies, plot_anomalies = TRUE)

Features and Capabilities

TimeGPT through the nixtlar package provides:

  • Zero-shot Inference: Generate forecasts and detect anomalies with no prior training
  • Fine-tuning: Enhance model performance for your specific datasets
  • Add Exogenous Variables: Incorporate additional variables like special dates or events to improve accuracy
  • Multiple Series Forecasting: Simultaneously forecast multiple time series
  • Custom Loss Function: Tailor the fine-tuning process with specific performance metrics
  • Cross Validation: Implement out-of-the-box validation techniques
  • Prediction Intervals: Quantify uncertainty in your predictions
  • Irregular Timestamps: Handle data with non-uniform intervals

How to Cite

If you find TimeGPT useful for your research, please consider citing:

Garza, A., Challu, C., & Mergenthaler-Canseco, M. (2024). TimeGPT-1.
arXiv preprint arXiv:2310.03589. Available at
https://arxiv.org/abs/2310.03589

Support

If you have questions or need support, please email support@nixtla.io.

TimeGPT is closed source. However, this SDK is open source and available under the Apache 2.0 License.