shu-hai / ts_ensemble_sunspot

XGBoost-DL: A XGBoost-based ensemble method that combines deep learning models for sunspot number prediction

A comparative study of non-deep learning, deep learning, and ensemble learning methods for sunspot number prediction

Link to full paper: Y. Dang, Z. Chen, H. Li, H. Shu (2022). Applied Artificial Intelligence, 36(1), e2074129

Requirements

• Python 3.8
• xgboost==1.5.1
• ray==1.9.0
• ray[tune]==1.9.0
• torch==1.9.0
• numpy==1.20.3
• pandas==1.1.4
• matplotlib==3.4.2
• seaborn==0.11.2

Run the following command to install the required dependencies:

pip install -r requirements.txt

Follow the commands below to reproduce results in this study:

1. Generate predictions on the test portion with provided pre-trained models:

python informer_result.py --use_pre_trained --use_nasa_test_range --pre_trained_file_name ../../train_models/best_informer.pth
python transformer_result.py --use_pre_trained --use_nasa_test_range --pre_trained_file_name ../../train_models/best_transformer.pth
python lstm_result.py --use_pre_trained --use_nasa_test_range --pre_trained_file_name ../../train_models/best_lstm.pth
python gru_result.py --use_pre_trained --use_nasa_test_range --pre_trained_file_name ../../train_models/best_gru.pth

2. Generate predictions on the future portion with provided pre-trained models:

python informer_future.py --use_pre_trained --pre_trained_file_name ../../train_models/best_informer_future.pth
python transformer_future.py --use_pre_trained --pre_trained_file_name ../../train_models/best_transformer_future.pth
python lstm_future.py --use_pre_trained --pre_trained_file_name ../../train_models/best_lstm_future.pth
python gru_future.py --use_pre_trained --pre_trained_file_name ../../train_models/best_gru_future.pth

3. Combine predictions on both test and future portions with pre-trained XGBoost models:

python xgboost_ensemble.py --test_pre_trained_file_name ../../train_models/xgboost_dl.pth --future_pre_trained_file_name ../../train_models/xgboost_future.pth