Python中ETS预测函数出现UnboundLocalError问题求助

问题描述 #

我编写了用于ETS预测的函数，但运行时出现UnboundLocalError错误，代码和报错信息如下：

原代码 #

# evaluate an ETS model for a given parameters (t,d,s,p,b,r)
def train_ets_model(train, test, cfg):
    t,d,s,p,b,r = cfg
    # prepare training dataset
    history = [x for x in train.values]
    # make predictions
    predictions = list()
    for t in range(len(test.values)):
        model = ExponentialSmoothing(history, trend=t, damped=d, seasonal=s, seasonal_periods=p, use_boxcox=b)
        # fit model
        model_fit = model.fit(optimized=True, remove_bias=r)
        #one-step prediction
        yhat = model_fit.predict(len(history), len(history))[0]
        predictions.append(yhat)
        history.append(test.values[t])
    # calculate out of sample error
    error = np.round(mean_squared_error(test, predictions),2)
    return error, predictions

#evaluate configurations for ETS model
def evaluate_ets_models(train, test, cfg_list):
    best_score, best_cfg = float("inf"), None
    for cfg in cfg_list:
        try:
            mse, predictions = train_ets_model(train, test, cfg)
            if mse < best_score:
                best_score, best_cfg = mse, cfg
                #print('ETS%s RMSE=%.3f' % (order,rmse))
        except:
            continue
    return best_cfg, best_score, predictions
    #print('Best ETS%s RMSE=%.3f' % (best_cfg, best_score))

报错信息 #

---------------------------------------------------------------------------
UnboundLocalError                         Traceback (most recent call last)
Cell In[38], line 1
----> 1 evaluate_ets_models(train, test, cfg_list)

Cell In[37], line 32, in evaluate_ets_models(train, test, cfg_list)
     30     except:
     31         continue
---> 32 return best_cfg, best_score, predictions

UnboundLocalError: local variable 'predictions' referenced before assignment

错误原因与修复方案 #

错误原因 #

1. predictions变量仅在try代码块内赋值，如果所有配置都触发异常（比如cfg_list为空，或每个配置运行都报错），predictions会从未被定义，返回时触发未绑定变量错误。
2. 原逻辑中predictions仅保存最后一次循环的结果，而非最优配置对应的预测值，属于逻辑漏洞。

修改后的evaluate_ets_models函数 #

# evaluate configurations for ETS model
def evaluate_ets_models(train, test, cfg_list):
    best_score, best_cfg, best_predictions = float("inf"), None, []
    for cfg in cfg_list:
        try:
            mse, predictions = train_ets_model(train, test, cfg)
            if mse < best_score:
                best_score, best_cfg = mse, cfg
                best_predictions = predictions  # 保存最优配置对应的预测结果
                #print('ETS%s RMSE=%.3f' % (order,rmse))
        except:
            continue
    return best_cfg, best_score, best_predictions

修改说明 #

• 初始化best_predictions为空列表，确保无论是否有成功运行的配置，返回时变量都已定义。
• 每次找到更优MSE的模型时，同步更新best_predictions，保证返回的是最优模型的预测结果，而非最后一次循环的结果。
• 若所有配置运行失败，返回的best_predictions为空列表，best_cfg为None，best_score为inf，可根据需求添加额外的异常提示逻辑。

内容的提问来源于stack exchange，提问作者Mridul Bhardwaj