Function gen_real_pred_errors

utils.py:370–400

(df, test_periods=6)

    return cluster_labels.astype(str)

def gen_real_pred_errors(df, test_periods=6):
    errors = []
    for region in df['REGIONAL_OFFICE_NAME'].unique():
        region_data = df[df['REGIONAL_OFFICE_NAME'] == region]
        for bnf_code in region_data['BNF_CHAPTER_PLUS_CODE'].unique():
            bnf_data = region_data[region_data['BNF_CHAPTER_PLUS_CODE'] == bnf_code]
            ts_data = bnf_data.groupby('YEAR_MONTH')['TOTAL_COST'].sum().reset_index()
            if len(ts_data) < test_periods + 3:
                continue  
            train = ts_data.iloc[:-test_periods]
            test = ts_data.iloc[-test_periods:]
            try:
                forecast_df = train_arima(train, test_periods)
                if len(forecast_df) != len(test):
                    continue
                y_true = test['TOTAL_COST'].values
                y_pred = forecast_df['FORECAST'].values
                mae = np.mean(np.abs(y_true - y_pred))
                bias = np.mean(y_pred - y_true)
                mape = np.mean(np.abs((y_true - y_pred) / y_true)) * 100 if np.all(y_true != 0) else np.nan
                errors.append({
                    'REGIONAL_OFFICE_NAME': region,
                    'BNF_CATEGORY': bnf_code.split(':')[0].strip(),
                    'Mean_Actual': np.mean(y_true),
                    'MAE': mae,
                    'Bias': bias,
                    'MAPE': mape
                })
            except Exception:
                continue
    return pd.DataFrame(errors)