TimeGPT

TimeGPT is a generative pre-trained transformer model for prediction tasks, developed by Nixtla. The model was trained on over 100 billion rows of financial, weather, energy, and web data, and Nixtla states it is the largest collection of public time series data in history. TimeGPT can discern complex patterns and predict future data points to help businesses make data-driven decisions. The model can also be fine-tuned with user data for specialization in specific prediction tasks. TimeGPT has use cases across a number of sectors (finance, healthcare, meteorology, social sciences, retail, energy, IoT, etc.) where identifying temporal patterns, trends, and variations is critical to forecasting future values and informing decision-making processes.

A time series dataset consists of a sequence of numerical values in successive order. In the context of TimeGPT, each data point is considered an input to the model, which learns and identifies patterns to make future predictions and detect anomalies. Time series data arises across a number of industries. Examples include stock prices, weather records, or sales data over time. Forecasting future outcomes of these datasets is critical to a number of fields. TimeGPT applies machine learning, and in particular deep learning, to the problem of predicting future values in time series datasets. TimeGPT can analyze multiple time series datasets concurrently and can also be fine-tuned using a broad time series collection, enabling users to tailor the model to their specific needs. The model processes time series data in chunks or tokens, similar to natural language processing. The number of tokens used to forecast future data points depends on the frequency of the time series data. TimeGPT is available through an API, which offers a range of time-series-related tasks, including what-if scenarios, anomaly detection, and more.

TimeGPT was unveiled on August 21, 2023, at the MindsDB San Francisco AI Collective, with the model available in beta. Businesses can express interest in the product and join a waiting list to receive access. The product's parent company, Nixtla, is a time series research and development company aiming to democratize AI tools. The company advocates for open-source development with a number of models, time series forecasting tools, and a Python software development kit for TimeGPT available on its GitHub. Prior to the unveiling of TimeGPT, Nixtla had already built a time series ecosystem. With over 5 million downloads, Nixtla's software is used by a number of large companies, including Amazon, Walmart, and Lyft. On October 5, 2023, researchers from Nixtla submitted a paper describing and evaluating the TimeGPT model.