================================================================

# This work is presented in http://www.theses.fr/2019PESC1031/document

================================================================

# It can be installed with :
cd ~/Downloads
git clone https://github.com/BCD65/electricityLoadForecasting.git
cd electricityLoadForecasting
conda create --name elec python=3.7 pip
conda activate elec
pip install -e .
# [Optional] To be used, the packages xgboost and spams have to be installed separately :
conda install -c conda-forge python-spams xgboost sklearn-contrib-py-earth ipdb openblas
# [Optional] Also, rpy2 in Python should be available if the GAM are to be tested.
pip install rpy2==3.3.1
# [Optional] If you do not want to use the default R installation and the mgcv library :
# conda install -c r r r-mgcv

================================================================

# It can be tested with
python scripts/preprocessing_eCO2mix.py
python scripts/main_forecasting.py

================================================================

The script preprocessing_eCO2mix.py downloads, formats and saves public data from MeteoFrance and RTE websites. The load data and the weather data are saved in two separate multiindexed pandas dataframes. The objective is to have a public dataset that can then be used as the input of forecasting algorithms. In preprocessing/eCO2mix/config.py, you can choose either the national database 'France' or the 'administrative_regions'.

Running the script main_forecasting.py, with the dataset selected in forecasting/hyperparameters/choose_dataset.py, launches the whole learning process : selecting the inputs according to the chosen hyperparameters, computing the features and optimizing the coefficients.

================================================================