Spectral Temporal Graph Neural Network for Multivariate Time-series Forecasting

This repository is the wooheon's reproducing of StemGNN(NeurIPS20)

Requirements

pip install --upgrade pip
pip install -r requirements.txt

Datasets

https://github.com/microsoft/StemGNN

Training and Evaluation

python main.py --dataset <name of csv file> --window_size <length of sliding window> --horizon <predict horizon> --norm_method z_score --batch_size 64 --train_length 7 --validate_length 2 --test_length 1

The detailed descriptions about the parameters are as following: New parameter are bold type.

Parameter name	Description of parameter
dataset	file name of input csv
window_size	length of sliding window, default 12
horizon	predict horizon, default 3
train_length	length of training data, default 7
validate_length	length of validation data, default 2
test_length	length of testing data, default 1
epoch	epoch size during training
optimizer	optimizer, default RMSProp
lr	learning rate, default 1e-3
decay_rate	decay rate, default 0.7
exponential_decay_step	exponential decay step, default 5
randomwalk_laplacian	determine whether to use randomwalk normalized laplacian matrix
attention_channel	hyper parameter of latent correlation layer, default 32
kernel_size	hyper parameter of Gated CNN's kernel size, default 3
gcnn_channel	hyper parameter of Gated CNN's channel, default 32
gconv_channel	hyper parameter of Graph Convolution channel, default 64
multi_channel	hyper parameter of StemBlock's forecast, backcast output channel, default 128
device	device that the code works on, 'cpu' or 'cuda:x'
validate_freq	frequency of validation
batch_size	batch size, default 64
dropout_rate	dropout_rate, default 0.2
leakyrelu_rate	leakyrelu rate, default 0.5
norm_method	method for normalization, 'z_score' or 'min_max'
early_stop	whether to enable early stop, default False

Results

My reproducing model shows following performance on the 10 datasets:

Table 1 Configuration and perforamance for all datasets

Dataset	window_size	horizon	norm_method	MAE	RMSE	MAPE(%)
METR-LA	12	3	z_score
PEMS-BAY	12	3	z_score
PEMS03	12	3	z_score
PEMS04	12	3	z_score
PEMS07	12	3	z_score
PEMS08	12	3	z_score
COVID-19	28	28	z_score