- This project has been validated here with a detailed tutorial for experiment duplication.
- Check out the slides AAG 2021 Presentation for more project details.
Dataset can be downloaded here. Please contact the author Jingchao Yang (jyang43@gmu.edu) for direct access if link expires.
- Place the dataset in the data folder to avoid additional path setup before running the code
Note: All data has been preprocessed to csv format, raw data can be accessed from weather underground and GeoTab. Toolset for preprocessing raw data can be accessed upon request.
- Python 3.7
- PyTorch 1.7.0 (code has GPU support, but can run without)
- Pandas 1.0.1
- scikit-learn
- scipy
- numpy
- matplotlib
- tqdm
- pmdarima
- xgboost
- multistep_lstm indludes python files for LSTM model building and training.
- multistep_others includes comparison model ARIMA and XGBoost.
To run our LSTM model for regional training, go to the directory and use the command
python run_auto.py
LSTM was also developed to support transfer learning with command
python run_auto.py --transLearn
Note: Model training can take much longer time without GPU support. LA Dataset already includes trained models and ready for transfer learning, user can delete the content inside the LA/output to retrain
Model output will be store in the data/output folder
Creat result folder under multistep_others for model output. ARIMA and XGBoost are for model comparison and were not developed for transfer learning
To use our ARIMA model, go to multistep_others and use the command
python auto_arima_run.py
To use our XGBoost model, go to multistep_others and use the command
python xgboost_run.py