Awesome

Large-scale Dynamic Linear Model (LSDLM)

This is a Python implementation of Large-scale dynamic linear model in the following paper: https://arxiv.org/abs/2104.13414

Requirements

pandas==1.3.0
numpy==1.21.0
scikit-learn==0.24.2
scipy==1.7.0
tqdm==4.61.2

Dependency can be installed using the following command (recommanded to install in a venv):

pip install -r requirements.txt

Train and evaluation

python main.py --dataset PEMS-BAY --horizon 6 --train   # for METR-LA dataset: --dataset METR-LA

loading dataset... done.
pre-processing the dataset...: 100%|█████████| 325/325 [00:01<00:00, 321.49it/s]
splitting dataset to training and test set (8:2 ratio)... done.
tau_short = 0.0001
tau_long = 1000
model created... start to train...
100%|█████████████████████████████████████████| 288/288 [16:28<00:00,  3.43s/it]
training finished!
prediction for h=6...: 288it [00:01, 149.84it/s]
RMSE: 3.77

Total computation for prediction: 2.15 sec