如何在Pandas中向前填充DataFrame的A列NaN直至三列值相等

原始数据 #

原始DataFrame定义如下：

import pandas as pd
import numpy as np

df = pd.DataFrame({
    "A": ["sell", np.nan, np.nan, np.nan, np.nan, "buy", np.nan, np.nan, np.nan, np.nan],
    "B": ["buy", "buy", "sell", "buy", "buy", np.nan, "buy", "buy", "buy", np.nan],
    "C": ["sell", "sell", "sell", "sell", "buy", "sell", "sell", "buy", "buy", np.nan]
}, index=pd.date_range("2025-05-22", periods=10, freq="15min"))

输出结果：

A     B     C
2025-05-22 00:00:00  sell   buy  sell
2025-05-22 00:15:00   NaN   buy  sell
2025-05-22 00:30:00   NaN  sell  sell
2025-05-22 00:45:00   NaN   buy  sell
2025-05-22 01:00:00   NaN   buy   buy
2025-05-22 01:15:00   buy   NaN  sell
2025-05-22 01:30:00   NaN   buy  sell
2025-05-22 01:45:00   NaN   buy   buy
2025-05-22 02:00:00   NaN   buy   buy
2025-05-22 02:15:00   NaN   NaN   buy

需求说明 #

对A列的NaN值进行向前填充，但填充仅持续到A、B、C三列的值完全相等为止，超出该位置的NaN保持不变。

期望结果 #

A     B     C
2025-05-22 00:00:00  sell   buy  sell
2025-05-22 00:15:00  sell   buy  sell
2025-05-22 00:30:00  sell  sell  sell
2025-05-22 00:45:00   NaN   buy  sell
2025-05-22 01:00:00   NaN   buy   buy
2025-05-22 01:15:00   buy   NaN  sell
2025-05-22 01:30:00   buy   buy  sell
2025-05-22 01:45:00   buy   buy   buy
2025-05-22 02:00:00   NaN   buy   buy
2025-05-22 02:15:00   NaN   NaN   buy 

核心思路是先完成全量向前填充，再通过条件筛选保留符合要求的填充值，超出三列相等位置的部分恢复为NaN。具体代码如下：

import pandas as pd
import numpy as np

# 原始DataFrame（重复定义仅为完整演示，实际可省略）
df = pd.DataFrame({
    "A": ["sell", np.nan, np.nan, np.nan, np.nan, "buy", np.nan, np.nan, np.nan, np.nan],
    "B": ["buy", "buy", "sell", "buy", "buy", np.nan, "buy", "buy", "buy", np.nan],
    "C": ["sell", "sell", "sell", "sell", "buy", "sell", "sell", "buy", "buy", np.nan]
}, index=pd.date_range("2025-05-22", periods=10, freq="15min"))

# 1. 对A列做向前填充，生成临时列
df['A_ffill'] = df['A'].ffill()

# 2. 标记三列值完全相等的行（自动排除B/C为NaN的情况，因NaN不与任何值相等）
mask_equal = (df['A_ffill'] == df['B']) & (df['A_ffill'] == df['C'])

# 3. 按A列非NaN值分组，找到每个分组内第一个三列相等的位置
groups = df['A'].notna().cumsum()
first_equal = mask_equal.groupby(groups).idxmax()

# 4. 生成最终A列：仅保留从非NaN值到第一个相等行之间的填充值，其余设为NaN
df['A'] = np.where(df.index <= first_equal.loc[groups].values, df['A_ffill'], np.nan)

# 清理临时列
df.drop('A_ffill', axis=1, inplace=True)

# 输出结果
print(df)

代码逻辑说明 #

• 步骤1：先对A列做全量向前填充，得到所有可能的填充候选值；
• 步骤2：通过布尔掩码标记出A（填充后）、B、C三列值完全一致的行；
• 步骤3：以A列的非NaN值为分组起点，找到每个分组内第一个满足三列相等的行索引；
• 步骤4：通过np.where判断，仅保留分组内从非NaN值到第一个相等行之间的填充值，其余位置恢复为NaN。

内容的提问来源于stack exchange，提问作者beni