torch_scatter==2.0.9
tqdm==4.62.3
torch==1.10.0
numpy==1.20.3
torch_sparse==0.6.12
torch_geometric==2.0.2
PyYAML==6.0
scikit_learn==1.0.2
The code in this repo is a node classification example on Cora of the template. Try
python main.py --model GCN --dataset Cora
to run code.
Config files are in config/{dataset_name}.yml format:
dataset: "Cora"
model_name: "GCN" # Name of the used baseline model, which can be change to 'GAT' of others.
# config for each baseline model
GCN:
epochs: 150
multirun: 10
dropout: 0.5
cuda: 0
feat_norm: True
hidden_dim: 64
multilabel: False
patience: 50
seed: 1234
lr: 0.005
weight_decay: 0.0005
lr_scheduler: False
monitor: "val_acc"
recache: False
optimizer: "Adam"
num_layers: 2
activation: "relu"
GAT:
epochs: 100
multirun: 10
dropout: 0.6
cuda: 0
feat_norm: True
hidden_dim: 64
multilabel: False
heads: 1
patience: 50
seed: 1234
lr: 0.005
weight_decay: 0.0005
lr_scheduler: False
monitor: "val_acc"
recache: False
num_layers: 2
optimizer: "Adam"
activation: "leaky_relu"
For each dataset, we need a specific config file. In each file, we config all baseline models.
All baseline models are in model/
In training_procedure/prepare.py, we config optimizer, loss function and model init parameters.
In training_procedure/train.py, we train the model.
In training_procedure/evaluate.py, we test the model.
This project is inspired by the project TWIRLS and IFM_Lab_Program_Template