ALSTM contains a temporal attentive aggregation layer based on normal LSTM.
Paper: A dual-stage attention-based recurrent neural network for time series prediction.
https://www.ijcai.org/Proceedings/2017/0366.pdf
NOTE: Current version of implementation is just a simplified version of ALSTM. It is an LSTM with attention.
