A dual stage attention LSTM network used for time series prediction.
This is an implementation of the paper for my use case of predicting solar irradiation values. The input sequence is multivariate with various meterological parameters.
The model code is heavily inspired by this blogpost by Chandler Zuo.
This code has added functionality:
- To define various training runs with different set of hyperparameters
- Keep track of all the runs during training using
tensorboard