thuml / Large-Time-Series-Model

Official code, datasets and checkpoints for "Timer: Generative Pre-trained Transformers Are Large Time Series Models" (ICML 2024)
https://arxiv.org/abs/2402.02368
MIT License
222 stars 15 forks source link

Zero-shot forecasting without training #6

Open ozanbarism opened 2 months ago

ozanbarism commented 2 months ago

Hi, is there a way for me to do zero-shot forecasting with timer without training it?

WenWeiTHU commented 1 month ago

Of course, you can load the pre-trained model as you would with any other large time series model and directly input a time series for prediction (currently Timer supports no longer than 1440 sequence lengths). For longer sequence outputs, you can try using the autoregression provided in our code.