Python内部函数无法访问外部函数局部变量df的问题咨询

你碰到的"Unresolved reference 'df'"错误，核心原因是子函数accumulate_df的else分支里写了df = df.append(...)——Python会把这种赋值操作认定为你要在子函数内创建局部变量df，因此整个子函数里的df都会被当作局部变量处理。但if分支里先访问了df，此时局部的df还未定义，就触发了未解析引用的错误。

有两种直接的修复方式：

方式一：用nonlocal声明变量 #

在子函数开头加上nonlocal df，明确告诉Python：这个df不是子函数的局部变量，而是外层函数func1中的变量，这样赋值操作就会直接修改外层的df：

import pandas as pd
import numpy as np

def func1():
    
    # Initialization for accumulated df
    df = pd.DataFrame(columns=['col1',
                               'col2',
                               'col3',
                               'col4'])

    df['col1'] = df['col1'].astype('string')
    df['col2'] = df['col2'].astype(np.float32)
    df['col3'] = df['col3'].astype(np.int32)
    df['col4'] = df['col4'].astype(np.float64)

    def accumulate_df(result_tuple):
        nonlocal df  # 声明df来自外层函数
        col1_value = result_tuple[0]
        col2_value = result_tuple[1]
        col3_value = result_tuple[2]
        col4_value = result_tuple[3]
        
        if df[(df['col1'] == col1_value) & (df['col2'] == col2_value)].values.any():
            # Update Accumulate
            df.loc[(df['col1'] == col1_value) & (df['col2'] == col2_value), 'col3'] += col3_value
            df.loc[(df['col1'] == col1_value) & (df['col2'] == col2_value), 'col4'] += col4_value

        else:
            df = df.append({'col1': col1_value,
                            'col2': col2_value,
                            'col3': col3_value,
                            'col4': col4_value}, ignore_index=True)

方式二：避免重新赋值df #

df.append()会返回新的DataFrame，你可以直接修改原df的行，不用重新赋值。比如用df.loc[len(df)] = ...添加新行，这样子函数里不会出现对df的赋值操作，自然会引用外层的df：

import pandas as pd
import numpy as np

def func1():
    
    # Initialization for accumulated df
    df = pd.DataFrame(columns=['col1',
                               'col2',
                               'col3',
                               'col4'])

    df['col1'] = df['col1'].astype('string')
    df['col2'] = df['col2'].astype(np.float32)
    df['col3'] = df['col3'].astype(np.int32)
    df['col4'] = df['col4'].astype(np.float64)

    def accumulate_df(result_tuple):
        col1_value = result_tuple[0]
        col2_value = result_tuple[1]
        col3_value = result_tuple[2]
        col4_value = result_tuple[3]
        
        mask = (df['col1'] == col1_value) & (df['col2'] == col2_value)
        if mask.any():
            # Update Accumulate
            df.loc[mask, 'col3'] += col3_value
            df.loc[mask, 'col4'] += col4_value

        else:
            # 直接给原df添加行，不重新赋值df
            df.loc[len(df)] = {
                'col1': col1_value,
                'col2': col2_value,
                'col3': col3_value,
                'col4': col4_value
            }

你提到要用多处理器累加结果，需要注意：多进程模式下，每个子进程会复制一份df的内存空间，进程间的变量修改不会共享。因此最后需要把各个进程的结果df收集起来，再合并成最终的df，而非直接在子进程里修改同一个df。

内容的提问来源于stack exchange，提问作者burcak