Explore the open source large time model Time-MoE and apply it in a small experiment using Python
Traditionally, the field of time series forecasting relied on specific data models, where a model was trained on a specific data set and task. If the data or the forecast horizon changed, the model also had to change.
Since October 2023, researchers have been actively developing foundation forecasting models. With these broad-time models, a single model can now handle different forecasting tasks from different domains, with different frequencies and with virtually any forecast horizon.
These great time models include:
- TimeGPT, accessed via API, making it easy to forecast and adjust without using local resources.
- Lag-Llama, an open source model for probabilistic forecasting that builds features from lagged values
- Chronos, a T5-based model that translated the unlimited time series domain to the limited language domain through tokenization and quantization.
- Moirai, a model that supports exogenous features and the first to publicly share its LOTSA dataset containing over 27 billion data points.