Overview

TimeGPT is a generative pre-trained transformer model specialized in prediction tasks. TimeGPT was trained on the largest collection of public time series data in history – over 100 billion rows of financial, weather, energy, and web data – to democratize the power of time-series analysis. This tool is capable of discerning patterns and predicting future data points in a matter of seconds.

Moreover, TimeGPT supports fine-tuning, which further specializes the model on specific prediction tasks. This process is akin to training a machine learning model on a subset of data to get better performance at a specific task, making TimeGPT a versatile tool for predictive analytics.

Guides and Recipes (Python)

Use the API reference to create requests using HTTPS in any language you want. Or follow these guides or recipes to use our Python SDK (more coming soon).

Quickstart guide

Check this quick-start guide to start making predictions for a single series.

Uncertainty Quantification

Learn how to quantify uncertainty in your predictions.

Fine-tuning

Follow this recipe to fine-tune on your own data.

Forecasting multiple time series

Forecast multiple series at once.

Calendar and Date Variables

Include date Features or calendar variables to improve accuracy

Exogenous Variables

Include important information like special dates or prices to improve your forecasts.

Anomaly Detection

Detect historical anomalies with TimeGPT's prediction intervals.

Key Concepts

Key Concepts