Execute Cross Validation and Performance Loop

execute_cross_validation_and_performance_loop.py

def execute_cross_validation_and_performance_loop(cross_valid_params, metric = 'mse'):
""" Execute Cross Validation and Performance Loop
    To complement Medium Blog Post: Prophet Auto Selection with Cross-Validation 
    Link https://medium.com/@jeanphilippemallette/prophet-auto-selection-with-cross-validation-7ba2c0a3beef

    Parameters
    ----------
    cross_valid_params: List of dict
      dict value same as cross_validation function argument
      model, horizon, period, initial
    metric: string 
      sort the dataframe in ascending order base on the 
      performance metric of your choice either mse, rmse, mae or mape

    Returns
    -------
    A pd.DataFrame with cross_validation result. One row
    per different configuration sorted ascending base on
    the metric inputed by the user.
    
    Example
    --------
    >>> m = Prophet()
    >>> df = pd.read_csv('/examples/example_wp_log_peyton_manning.csv')
    >>> m.fit(df)
    >>> cross_valid_params = [{'model': m,
                      'initial': '730 days',
                      'period': '180 days',
                      'horizon': '365 days'},
                      {'model': m,
                       'initial': '500 days',
                        'period': '180 days',
                        'horizon': '365 days'}]
    
    index     initial	horizon	    period	    mse	        rmse	    mae	        mape	    coverage
    4332	  500 days	365 days	180 days	0.663628	0.814634	0.627102	0.075824	0.572352
    3987	  730 days	365 days	180 days	0.670460	0.818816	0.628407	0.075577	0.589017
    
    """
    
    assert metric in ['mse','rmse','mae','mape'], \
                    'metric must be either mse, rmse, mae or mape'

    df_ps = pd.DataFrame()

    for cross_valid_param in cross_valid_params:
        df_cv = cross_validation(**cross_valid_param)
        df_p = performance_metrics(df_cv, rolling_window=1)
        df_p['initial'] = cross_valid_param['initial']
        df_p['period'] = cross_valid_param['period']
        df_ps = df_ps.append(df_p)
    
    df_ps = df_ps[['initial','horizon','period','mse'
                    ,'rmse','mae','mape','coverage']]
    return df_ps.sort_values(metric)