fire_pred_data_and_models_to_truera.py

project_name = 'Fire_Party'
tru.set_environment('local')
tru.add_project(project_name, score_type='probits')
extra_data_columns = ['year']
train_split_name = 'train'
burned_fraction_th = 0.01
for window_size in range(1,11):
    key = f'{window_size}year_window'
    print(key)
    tru.add_data_collection(key)
    tru.add_data_split(train_split_name, data_train_x[key], label_data=(data_train_y[key]>burned_fraction_th), extra_data_df=data_train[key][extra_data_columns])
    for year in range(2001,2017):
        print(year)
        data_tmp = data_yearly[key][data_yearly[key]['year']==year]
        x = data_tmp.drop(columns=['year', 'burned_fraction'])
        y = data_tmp['burned_fraction']
        extra = data_tmp[extra_data_columns]
        tru.add_data_split(f'{year}', x, label_data=(y>burned_fraction_th), extra_data_df=extra)
    for model_name in models:
        if key not in model_name:
            continue
        print(model_name)
        model_params = {
            'model_type': type(models[model_name]).__name__,
        }
        model_prefix = model_name.split('_')[0]
        if model_prefix == 'linear':
            model_params['max_iter'] = 1000
            model_params['solver'] = 'saga'
        elif model_prefix == 'gb':
            model_params['n_estimators'] = int(model_name.split('_')[1])
        elif model_prefix == 'SVC':
            model_params['c_weight'] = {0:1.0, 1:int(test.split('SVC')[1].split('_')[0])}
        tru.add_python_model(model_name, models[model_name], train_split_name=train_split_name, train_parameters=model_params)