HTNet: Dynamic WLAN Performance Prediction using Heterogenous Temporal GNN

Dependencies

Older versions may also work but are not tested.

• python >= 3.9.7
• numpy >= 1.20.3
• pytorch >= 1.10.2
• dgl >= 0.8.0
• xgboost >= 1.5.2
• scikit-learn >= 0.24.2

Dataset

All six setups can be downloaded on this Google Drive Link. After downloading, unzip the file and copy them to /data/ folder. The file structure should be

├── data
│   ├── setup1
│   │   ├── processed
│   │   │   ├── train_0.bin
│   │   │   ├── valid_0.bin
│   │   │   ├── test_0.bin
│   │   │   ├── ...
│   ├── ...
├── train.py
├── ...

Run

To run HTNet and the baseline methods, first specify a setup in {setup1, setup2, setup3, setup4, setup5, setup6}. Ramon, ATARI, Ramon+LSTM, ATARI+LSTM, and HTNet require to train on GPU. For these methods, specify which GPU to use using the --gpu option where gpu 0 is the default value.

For SINR:

python train.py --data <setup> --sinr

For GBRT:

python train.py --data <setup> --gbrt

For Ramon:

python train.py --data <setup> 

For ATARI:

python train.py --data <setup> --graph

For Ramon+LSTM:

python train.py --data <setup> --dynamic

For ATARI+LSTM:

python train.py --data <setup> --graph --dynamic

For HTNet:

python train.py --data <setup> --graph --hetero --dynamic