Refacto: model_config

[vviers]

Still TODO 👷🏼‍♂️ :

• ~fill the config file with the correct values from the March model~
• ~create the CLI / script that consumes this config file and runs the model~

[vviers]

To run the script :

Install the package (as always) pip install -e .

and run 🙂 python -m predictsignauxfaibles

[slebastard]

Added some functionalities:

• a step was added to remove SIREN from the train dataset in the test dataset
• predictsignauxfaibles/main.py now uses argparse to take arguments for variables that a non data-scientist user may want to play with when training models and outputing lists. Models can now be trained like this: python3 predictsignauxfaibles --model_name my_model --config_file my_model_config --train_from 01/01/2015 --train_to 01/09/2015. When specifying a custom model and/or configuration file, we use the specific parameters declared in configuration file my_model/my_model_config. Those parameters can be overwritten à la carte using specific options:
  • "model_name" and "config_file" allow to use a different model and/or model configuration
  • oversampling
  • train_spl_size
  • train_from, train_to
  • test_spl_size
  • test_from, test_to
  • predict_on
• main.run() stores model stats in a dictionary and saves them in csv format @vviers can you let me know what you think?