解决干旱预测模型中LinearRegression的0样本ValueError问题

问题背景 #

构建干旱预测算法，输入起止年份、月份、城市后输出干旱指数柱状图，但运行时触发错误：

ValueError: Found array with 0 sample(s) (shape=(0, 5)) while a minimum of 1 is required by LinearRegression.

错误原因 #

问题出在predict_drought函数：调用predict_drought(best_model, 'Anuradhapura', 2030, 2040)时，grouped_data[(grouped_data['City'] == city) & (grouped_data['Year'].between(start_year, end_year))]返回空DataFrame。因为训练数据grouped_data中没有2030-2040年的该城市数据，导致输入模型的特征数组X_city样本量为0，无法满足模型预测的最低要求。

原代码逻辑错误：当前predict_drought仅从已有历史数据中筛选对应城市年份的数据做预测，而非用ARIMA生成的未来气候数据推导未来CDI。

解决方案 #

核心修正方向 #

先通过ARIMA按城市预测未来气候变量（温度、降水、蒸散量），再生成对应年份的特征数据输入回归模型，得到未来CDI预测结果。

修改后的代码示例 #

1. 调整ARIMA预测函数，支持按城市预测

def forecast_arima_by_city(data, city, column, order, steps):
    city_data = data[data['City'] == city]
    model = ARIMA(city_data[column], order=order)
    model_fit = model.fit()
    forecast = model_fit.forecast(steps=steps)
    return forecast

2. 重写predict_drought函数，实现未来预测逻辑

def predict_drought(model, city, start_year, end_year):
    # 生成目标年月序列
    years = range(start_year, end_year + 1)
    months = range(1, 13)
    future_dates = [pd.to_datetime(f"{y}-{m}-01") for y in years for m in months]
    total_steps = len(future_dates)
    
    # 预测该城市未来气候变量
    temp_forecast = forecast_arima_by_city(df, city, 'Average_temperature_month', order=(5,1,0), steps=total_steps)
    precip_forecast = forecast_arima_by_city(df, city, 'Average_precipitation_month', order=(5,1,0), steps=total_steps)
    evap_forecast = forecast_arima_by_city(df, city, 'Average_evapotranspiration_month', order=(5,1,0), steps=total_steps)
    
    # 构建模型输入特征
    future_features = pd.DataFrame({
        'Year': [d.year for d in future_dates],
        'Month': [d.month for d in future_dates],
        'Average_temperature_month': temp_forecast.values,
        'Average_precipitation_month': precip_forecast.values,
        'Average_evapotranspiration_month': evap_forecast.values
    })
    
    # 预测CDI并计算实际CDI（用于对比）
    future_features['Predicted_CDI'] = model.predict(future_features)
    future_features['Norm_Temperature'] = (future_features['Average_temperature_month'] - df['Average_temperature_month'].min()) / (df['Average_temperature_month'].max() - df['Average_temperature_month'].min())
    future_features['Norm_Precipitation'] = (future_features['Average_precipitation_month'] - df['Average_precipitation_month'].min()) / (df['Average_precipitation_month'].max() - df['Average_precipitation_month'].min())
    future_features['Norm_Evapotranspiration'] = (future_features['Average_evapotranspiration_month'] - df['Average_evapotranspiration_month'].min()) / (df['Average_evapotranspiration_month'].max() - df['Average_evapotranspiration_month'].min())
    future_features['CDI'] = (future_features['Norm_Precipitation'] + (1 - future_features['Norm_Temperature']) + future_features['Norm_Evapotranspiration']) / 3
    
    return future_features

3. 防御性编程补充（可选）

如果需要保留历史数据查询功能，可在原函数中加入空数据检查：

def predict_drought(model, city, start_year, end_year):
    city_data = grouped_data[(grouped_data['City'] == city) & (grouped_data['Year'].between(start_year, end_year))]
    if city_data.empty:
        print(f"Error: No data found for city {city} between {start_year} and {end_year}")
        return None
    X_city = city_data[['Year', 'Month', 'Average_temperature_month', 'Average_precipitation_month', 'Average_evapotranspiration_month']]
    predictions = model.predict(X_city)
    city_data['Predicted_CDI'] = predictions
    return city_data

关键说明 #

• 原代码逻辑误区：试图从不存在的未来年份历史数据中筛选特征，导致输入模型的特征为空。
• 正确的未来预测逻辑：先通过时间序列模型生成未来气候特征，再输入回归模型预测CDI。

内容的提问来源于stack exchange，提问作者Viper Rate