使用Pandas Timestamp计算企业运营月数时遇int64溢出错误的解决

问题场景 #

你的DataFrame如下：

执行以下代码时触发OverflowError: Overflow in int64 addition：

today = pd.Timestamp.now()
df["founded_on"]=pd.to_datetime(df["founded_on"])
df["Time_Since_founded_on"] = (today - df["founded_on"]).dt.days // 30

解决方案 #

方法一：用Period计算自然月份差（推荐） #

直接基于月份周期计算差值，既避免溢出，又保证计算的是真实自然月数（比按30天估算更准确）：

import pandas as pd

# 初始化DataFrame（已有可跳过）
data = {
    "company_name": ["a", "b", "c", "d"],
    "founded_on": ["2004-01-01 00:00:00", "2013-01-01 00:00:00", "2008-01-01 00:00:00", "1997-01-01 00:00:00"]
}
df = pd.DataFrame(data)

df["founded_on"] = pd.to_datetime(df["founded_on"])
today = pd.Timestamp.now().floor('D')  # 取当日日期，忽略时分秒

# 转换为月份周期后计算差值
df["Time_Since_founded_on"] = (today.to_period('M') - df["founded_on"].dt.to_period('M')).apply(lambda x: x.n)

方法二：用relativedelta计算精确月份差 #

借助dateutil库的relativedelta直接计算年、月差值，转换为总月数：

import pandas as pd
from dateutil.relativedelta import relativedelta

# 初始化DataFrame（已有可跳过）
data = {
    "company_name": ["a", "b", "c", "d"],
    "founded_on": ["2004-01-01 00:00:00", "2013-01-01 00:00:00", "2008-01-01 00:00:00", "1997-01-01 00:00:00"]
}
df = pd.DataFrame(data)

df["founded_on"] = pd.to_datetime(df["founded_on"])
today = pd.Timestamp.now().floor('D')

df["Time_Since_founded_on"] = df["founded_on"].apply(
    lambda x: relativedelta(today, x).years * 12 + relativedelta(today, x).months
)

错误原因说明 #

原代码通过(today - df["founded_on"]).dt.days计算天数差再估算月数，当时间差对应的天数超出int64类型存储范围（或pandas内部处理触发溢出）时，就会抛出OverflowError。上述两种方法直接基于月份维度计算，不涉及大整数天数运算，从根源避免了溢出问题，同时结果更符合真实运营月数的定义。

内容的提问来源于stack exchange，提问作者Rebeka