Team Member: Wang Luozhou, Zhang He, Ji Yeon Fung, Ren Xinzhu
Python3.6.9
pip install -r requirements.txt
PS: If you get an error report in the dgl_cu101 ==0.5.2 command, please delete it and run the following command:
pip install --verbose --no-cache-dir torch-sparse torch-scatter
Because you probably don't have a graphics card or version of the driver, you cannot use CUDA to train your model. Therefore, you'll need to install the environment packages(Torch-Sparse and Scatter) to make sure that the code works (This model also can be trained directly with the CPU).
Run three datasets by type command in terminal
- Cora Dataset
python run.py --dataset Cora
- CiteSeer Dataset
python run.py --dataset CiteSeer
- PubMed Dataset
python run.py --dataset PubMed
The output will plot a curve of accuracy and loss firstly. When user switches off this window, the output will be two figures showing nodes position through t-SNE visualization before input into model and after respectively. Here are samples.
- gcn aggregator
python run_aggregator.py --aggregator gcn
- MaxPooling aggregator
python run_aggregator.py --aggregator pool
- LSTM aggregator
python run_aggregator.py --aggregator lstm