Discover the architecture and inner workings of Moirai and apply it in a forecasting project using Python
We have entered an era where large base models are now common. Base models have revolutionized many fields, such as computer vision and natural language processing, with models and applications that can generate text, images, and videos.
The field of time series forecasting has not been immune to this movement, and many basic models for forecasting have appeared. This marks a major paradigm shift, as we can now generate predictions from time series data at a moment’s notice and avoid the costs and development time of data-specific training models.
In October 2023, TimeGPT-1 was released, marking the appearance of one of the first prediction models on the foundation. Then, in February 2024, Lag-Llama was released, quickly followed by Chronos in March 2024.
In May 2024, a new forecasting model from the open source foundation Moirai was published. In the article Unified training of universal time series prediction transformersSalesforce researchers propose a baseline model capable of zero-impact probabilistic forecasting while allowing for exogenous features.
