vanilsongomes / RNN-for-Time-Series

Recurrent neural network (RNN) methods for time series forecasting

RNN-for-Time-Series

Recurrent Neural Network Methods for Time Series Forecasting.

This repo contains code and notes on recurrent neural network methods for time series forecasting. The goal is to cover the different aspects of recent developed methods which will be added succesively. Methods are presumably implemented using either AWS SageMaker (either built-in algorithm or in conjunction with another framework) or using the GluonTS API of the Apache MxNet deep learning framework.
Methods will be applied to the M4Competition dataset. Statistical time series forecasting methods such as the competition's benchmark methods are applied in repo: "M4-Statistical-Methods".

Recurrent neural network (RNN) methods:

DeepAR

• Salinas, David, Valentin Flunkert, and Jan Gasthaus. "DeepAR: Probabilistic forecasting with autoregressive recurrent networks." arXiv preprint arXiv:1704.04110 (2017).

ES-RNN / Smyl (2019)

• Smyl, Slawek. "A hybrid method of exponential smoothing and recurrent neural networks for time series forecasting." International Journal of Forecasting (2019).

DeepFactor

• Wang, Yuyang, et al. "Deep Factors for Forecasting." arXiv preprint arXiv:1905.12417 (2019).

DeepState

• Rangapuram, Syama Sundar, et al. "Deep state space models for time series forecasting." Advances in Neural Information Processing Systems. 2018.