PredAir
PredAir copied to clipboard
Forecasting air pollution using temporal attention mechanism in Beijing
Air-Pollution-Forecasting-using-Machine-Learning (APF Model)
First step towards solving a real-life problem - air pollution forecasting in Delhi, using deep learning
Will implement the following versions of the APF (Air Pollution Forecasting) Model:
- vanilla RNN encoder-decoder where both the RNNs will be plain RNNs
- LSTM-RNN encoder-decoder where both the RNNs will be LSTM
- vanilla RNN encoder - attention-based decoder where both the RNNs will be plain RNNs
- vanilla RNN bi-directional encoder - attention-based decoder where both the RNNs will be plain RNNs
- LSTM-RNN encoder - attention-based decoder where both the RNNs will be LSTM
- LSTM-RNN bi-directional encoder - attention-based decoder where both the RNNs will be LSTM
Will be trying with Bahdanau and Luong attention mechanisms individually
- Model based on the temporal-based attention where attention is given to tensors across time steps and also values of features of each tensor at every time step using the reference below: https://arxiv.org/abs/1809.04206v2 (Shun-Yao Shih, Fan-Keng Sun, Hung-yi Lee, 2018: "Temporal Pattern Attention for Multivariate Time Series Forecasting")
Will implement the model first in Keras and then in TensorFlow