fit_predictions.py

def fit_predictions(model_fit, steps_out_to_predict, actual_values):
    """
    This function predicts the SARIMA model out a certain designated number of steps,
    and compares the predictions to the actual values. The root mean squared error and
    the mean absolute error are calculated, comparing the predicted and actual values.
    The function returns the predicted values and their respective confidence intervals.
    Args:
        model_fit:  SARIMA model.
        steps_out_to_predict: Int. Number of steps out to predict the time series.
        actual_values: Series of actual time series values.
    Outputs:
        mean_predicted_values: Series of predicted time series values.
        confidence_interval_predicted_values: Dataframe, containing upper and lower thresholds of the
        confidence interval
    """
    predicted_values = model_fit.get_forecast(steps=steps_out_to_predict)
    mean_predicted_values = predicted_values.predicted_mean
    confidence_interval_predicted_values = predicted_values.conf_int()
    #Compare the actual to the predicted values using RMSE and MAE metrics
    rmse, mae = quantify_rmse_mae(mean_predicted_values, actual_values)
    print("Root mean squared error: ", str(rmse))
    print("Mean absolute error: ", str(mae))
    return mean_predicted_values, confidence_interval_predicted_values
    
def quantify_rmse_mae(predicted_values, actual_values):
    """
    This function calculates the root mean squared error and mean absolute error for 
    the predicted values, when compared to the actual values. These helps help us to
    gauge model performance. 
    Args:
        predicted_values: Series of predicted time series values.
        actual_values: Corresponding series of actual time series values.
    Outputs:
        rmse: Float. Root mean squared error.
        mae: Float. Mean absolute error.
    """
    #calcuate the mean squared error of the model
    rmse = math.sqrt(mean_squared_error(actual_values, predicted_values))
    #Calculate the mean absolute error of the model 
    mae = mean_absolute_error(actual_values, predicted_values)
    #Return the MSE and MAE for the model
    return rmse, mae
  
### EXECUTE IN THE MAIN FUNCTION ###
#Run the data on the test set to gauge model performance
mean_predicted_values, confidence_interval_predicted_values = fit_predictions(best_model, 
                                                                              len(test_set), 
                                                                              test_set)