This document discusses time series forecasting models including autoregressive (AR) models, DeepAR, and LSTNet. It provides the following information:
- Autoregressive (AR) models forecast a variable using its past values in a linear combination. The DeepAR model is based on autoregressive RNNs that learn from multiple time series datasets.
- LSTNet is designed to capture both long-term and short-term patterns in multivariate time series using CNNs, RNNs, and an autoregressive component. It combines the outputs from recurrent and recurrent-skip layers.
- The goal of models like DeepAR and LSTNet is to learn from similar time series to generalize without overfitting