Traffic Forecasting on Highways using Graph Neural Networks
installing Pytorch Lightning Temporal, which depends on pytorch Scatter & pytroch Sparse can be quite cumbersome, but it's worth it.
run training:
python TLS_GNN_mytune.py
run evaluation:
python TLS_GNN_eval.py
data exploration is done in:
TLS_GNN_A01.ipynb
and the results are analysed in:
analysis.ipynb
results:
| minutes | direction | baseline 0 | GCN_LSTM 50 | GCN_LSTM 100 | A3TGCN 50 | A3TGCN 100 |
|---|---|---|---|---|---|---|
| 5 | 1 | 1.31067 | 0.456955 | 1.0536 | 1.05083 | 1.04685 |
| 5 | 1,2 | 1.28284 | 1.01506 | 1.02451 | 1.03516 | 1.03057 |
| 5 | 2 | 1.2576 | 1.0034 | 0.590801 | 1.02898 | 1.02508 |
| 30 | 1 | 0.425698 | 1.02005 | 0.233556 | 1.00553 | 0.517924 |
| 30 | 1,2 | 0.413225 | 0.983687 | 0.267276 | 0.994697 | 0.992599 |
| 30 | 2 | 0.404753 | 1.01366 | 0.255877 | 0.523816 | 0.993287 |
| 60 | 1 | 0.591821 | 0.297151 | 0.290184 | 1.01491 | 1.00851 |
| 60 | 1,2 | 0.566442 | 0.317912 | 0.301587 | 0.991007 | 0.988065 |
| 60 | 2 | 0.544157 | 0.988103 | 0.285369 | 0.984486 | 0.979044 |