如何对DataFrame按交易ID与连续日号序列分组计算预测值的cumulative sum

首先构造示例DataFrame：

import pandas as pd

data = {
    "transaction ID（交易ID）": [12,12,12,12,13,13,13,13],
    "day number（日号）": [1,2,1,2,1,2,3,4],
    "Predicted value（预测值）": [0.001,0.002,0.001,0.002,0.001,0.002,0.002,0.003]
}
df = pd.DataFrame(data)

接下来生成分组标识：

• 按交易ID分组后，判断当前行的日号是否为1，若是则标记为新组起点
• 对每个交易ID内的标记进行累加，得到唯一的分组ID

# 生成分组键：同一交易ID下，日号为1时触发新分组
df['group_id'] = df.groupby('transaction ID（交易ID）')['day number（日号）'].apply(
    lambda x: (x == 1).cumsum()
)

最后按交易ID和group_id分组，对预测值求和：

result = df.groupby(['transaction ID（交易ID）', 'group_id'])['Predicted value（预测值）'].sum().tolist()
print(result)  # 输出: [0.003, 0.003, 0.008]

说明 #

• 核心逻辑是通过(x == 1).cumsum()在每个交易ID内部生成递增的分组标识，把从1开始的连续日号序列归为一组
• 分组后直接求和就能得到每组的累计总和，完全匹配需求结果

内容的提问来源于stack exchange，提问作者rushi