Pandas SettingWithCopy警告及DataFrame修改不生效问题求助

Pandas新手（首次使用）开发算法交易回测框架时遇到两个问题：

1. 触发SettingWithCopyWarning警告；
2. 在backtest函数中通过df.at修改transactions和wallet列后，返回的DataFrame中这些列仍为NaN，修改未生效。

相关代码与警告输出如下：

原代码 #

import pandas as pd
import numpy as np
import matplotlib as mpl
import random

def backtest(df):
    # Set wallet properties
    wallet = 1000 # dollars

    # iterate through day by day
    for i in range(len(df)):
        today = df.iloc[i,:]
        print(f"i = {str(i)}")

        signal = random.randint(-1,1)
        print('Signal: ' + str(signal))

        # hold signal
        if signal == 0:
            pass
        # buy signal
        elif signal == 1:
            print('Buying on: ' + str(today['timestamp']) + ' for ' + str(today['close']))
            df.at[i,'transactions'] = 1
            wallet -= today['close']
        # sell signal
        elif signal == -1:
            print('Selling on: ' + str(today['timestamp']) + ' for ' + str(today['close']))
            wallet += today['close']
            df.at[i,'transactions'] = -1
            print(df.at[i,'transactions'])
            print(f"Wallet at ${str(wallet)}")
        df.at[i,'wallet'] = wallet
        print()
    
    return df


# Import the dataset and get it ready to work with
df = pd.read_csv('Data/S&P500_2015-2018.csv')
df['timestamp'] = pd.to_datetime(df['timestamp'])

# Choose which stock(s) you're using
stock = df[df['symbol'] == 'AMD']

out = backtest(stock)
print('All done backtesting')
out.plot(x = 'timestamp', y = 'transactions')

警告输出 #

i = 0
Signal: 1
Buying on: 2015-12-01 05:00:00+00:00 for 2.34

i = 1
Signal: -1
Selling on: 2015-12-02 05:00:00+00:00 for 2.27
-1.0
Wallet at $999.93

...

i = 398
Signal: -1
Selling on: 2017-06-30 04:00:00+00:00 for 12.48
<ipython-input-10-c0810f02fee9>:25: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  df.loc[i,'transactions'] = 1

核心问题是操作了DataFrame的视图而非独立副本：

• 代码中stock = df[df['symbol'] == 'AMD']是对原DataFrame的切片操作，Pandas返回的是原数据的视图（或潜在副本），而非完全独立的新DataFrame。
• 当在backtest函数中修改这个对象时，Pandas无法确定你要修改的是原DataFrame还是临时副本，因此抛出SettingWithCopyWarning；同时修改可能仅作用于临时副本，导致最终返回的DataFrame中修改未生效，列仍为NaN。

方法1：切片时创建独立副本 #

在筛选股票时，添加.copy()显式创建独立的DataFrame，后续操作完全独立于原数据：

# 替换原筛选代码
stock = df[df['symbol'] == 'AMD'].copy()

方法2：在函数内部复制传入的DataFrame #

如果不想修改外部的筛选逻辑，可以在backtest函数开头复制传入的df，确保操作的是独立副本：

def backtest(df):
    # 新增：复制传入的DataFrame
    df = df.copy()
    wallet = 1000 # dollars
    # 后续代码不变...

额外优化：提前初始化目标列 #

为避免列初始为NaN的问题，可提前初始化transactions和wallet列：

# 在调用backtest前添加
stock['transactions'] = np.nan
stock['wallet'] = np.nan

import pandas as pd
import numpy as np
import matplotlib as mpl
import random

def backtest(df):
    # 复制传入的DataFrame，确保操作独立副本
    df = df.copy()
    wallet = 1000 # dollars
    # 初始化目标列
    df['transactions'] = np.nan
    df['wallet'] = np.nan

    for i in range(len(df)):
        today = df.iloc[i,:]
        print(f"i = {str(i)}")

        signal = random.randint(-1,1)
        print('Signal: ' + str(signal))

        if signal == 0:
            pass
        elif signal == 1:
            print('Buying on: ' + str(today['timestamp']) + ' for ' + str(today['close']))
            df.at[i,'transactions'] = 1
            wallet -= today['close']
        elif signal == -1:
            print('Selling on: ' + str(today['timestamp']) + ' for ' + str(today['close']))
            wallet += today['close']
            df.at[i,'transactions'] = -1
            print(df.at[i,'transactions'])
            print(f"Wallet at ${str(wallet)}")
        df.at[i,'wallet'] = wallet
        print()
    
    return df


df = pd.read_csv('Data/S&P500_2015-2018.csv')
df['timestamp'] = pd.to_datetime(df['timestamp'])

# 筛选时创建独立副本
stock = df[df['symbol'] == 'AMD'].copy()

out = backtest(stock)
print('All done backtesting')
out.plot(x = 'timestamp', y = 'transactions')

内容的提问来源于stack exchange，提问作者Jack Butler