It seems that the only data we will get when ranking is "Operating Reserve (OR)". Therefore, TML and DL methods based on full version data will not be adopted in Ranking Algorithm.
- Use
python app.pydirectly. The prediction will be saved assubmission.csv - Or use
python app.py --output "predict_name.csv"to specify a special output filename.
The full version of app.py can be found in 1_model_selection.ipynb, 2_1days_version.ipynb, and 3_2days_version.ipynb
- GradientBoostingRegression and 1-day preview approaches are used in
app.py.
- (2021/01/01 ~ 2022/01/31) Taiwan Power Company_Past Power Supply and Demand Information (https://data.gov.tw/dataset/19995) was pre-download in
data/0_raw_elec.csv - (2022/01/01 ~ ) Taiwan Power Company_Daily peak standby capacity rate for this year (https://data.gov.tw/dataset/25850) will be download while exec.
app.pyis a causal model, which means it will predict future information without peeking. Details are in "Causality".
For app.py algorithm, we will use the OR of day_(N) to predic the OR of day_(N+1). Therefore, the model is causal.
- Some data analysis and preprocessing were adoped in folder
\1_traditinal_ML - Traditional Machine Learning including K-Neighbor, SupportVevtor, NuSVR, DecisionTree, RandomForest, Adaboost, GradientBoosting, and ARIMA approaches were structured in folder
\1_Traditional_ML. - Deep Learning approaches including CNN, DNN, RNN, LSTM, BLSTM, CRNN, FCN (Fully Convolutional Network), Transformer were structured in folder
\2_DL.