如何根据熊猫滚动窗口中的多列查找重复项？

Question

样本数据

{"transaction": {"merchant": "merchantA", "amount": 20, "time": "2019-02-13T10:00:00.000Z"}}
{"transaction": {"merchant": "merchantB", "amount": 90, "time": "2019-02-13T11:00:01.000Z"}}
{"transaction": {"merchant": "merchantC", "amount": 90, "time": "2019-02-13T11:00:10.000Z"}}
{"transaction": {"merchant": "merchantD", "amount": 90, "time": "2019-02-13T11:00:20.000Z"}}
{"transaction": {"merchant": "merchantE", "amount": 90, "time": "2019-02-13T11:01:30.000Z"}}
{"transaction": {"merchant": "merchantE", "amount": 90, "time": "2019-02-13T11:02:30.000Z"}}
.
.

我有一些这样的代码

    df = pd.DataFrame()
for line in sys.stdin:
    data = json.loads(line)
    # df1 = pd.DataFrame(data["transaction"], index=[len(df.index)])
    df1 = pd.DataFrame(data["transaction"], index=[data['transaction']['time']])
    df1['time'] = pd.to_datetime(df1['time'])
    df = df.append(df1)
    # df['count'] = df.rolling('2min', on='time', min_periods=1)['amount'].count()

print(df)
print(len(df[df.merchant.eq(data['transaction']['merchant']) & df.amount.eq(data['transaction']['amount'])].index))

电流输出

2019-02-13T10:00:00.000Z  merchantA      20 2019-02-13 10:00:00
2019-02-13T11:00:01.000Z  merchantB      90 2019-02-13 11:00:01
2019-02-13T11:00:10.000Z  merchantC      90 2019-02-13 11:00:10
2019-02-13T11:00:20.000Z  merchantD      90 2019-02-13 11:00:20
2019-02-13T11:01:30.000Z  merchantE      90 2019-02-13 11:01:30
2019-02-13T11:02:30.000Z  merchantE      90 2019-02-13 11:02:30

2

预期输出

2019-02-13T10:00:00.000Z  merchantA      20 2019-02-13 10:00:00
2019-02-13T11:00:01.000Z  merchantB      90 2019-02-13 11:00:01
2019-02-13T11:00:10.000Z  merchantC      90 2019-02-13 11:00:10
2019-02-13T11:00:20.000Z  merchantD      90 2019-02-13 11:00:20
2019-02-13T11:01:30.000Z  merchantE      90 2019-02-13 11:01:30

由于数据正在流式传输。我想检查是否有重复记录（其商家和金额值相同）在两分钟内到达，所以我将其丢弃并且不对其进行处理。将其打印为副本。

我必须对索引压缩或 groupby 做些什么吗？但是然后如何等同于多列。 或者两列上有一些滚动条件，但找不到任何方法。

我在这里错过了什么？

谢谢

编辑

#dup = df[df.duplicated(subset=['merchant', 'amount'], keep=False)]
     res = df.loc[(df.merchant == data['transaction']['merchant']) & (df.amount == data['transaction']['amount'])]
        # res['timediff'] = pd.to_timedelta((data['transaction']['time'] - res['time']), unit='T')
        res['timediff'] = (data['transaction']['time'] - res['time'])
        if len(res.index) >1:
           print(res)

所以我尝试这样的事情，如果结果小于 120 秒，我可以处理它。 但生成的df目前以

的形式

                      merchant  amount                time       concat          timediff
2019-02-13 11:03:00  merchantF      10 2019-02-13 11:03:00  merchantF10 -1 days +23:59:20
2019-02-13 11:02:20  merchantF      10 2019-02-13 11:02:20  merchantF10          00:00:00

2019-02-13 11:01:30  merchantE      10 2019-02-13 11:01:30  merchantE10 00:01:00
2019-02-13 11:02:00  merchantE      10 2019-02-13 11:02:00  merchantE10 00:00:30
2019-02-13 11:02:30  merchantE      10 2019-02-13 11:02:30  merchantE10 00:00:00

-1 天 +23:59:20 这种格式我觉得可以用绝对值代替？

如何将时间转换为可以与 120 秒比较的格式？ pd.to_deltatime() 对我不起作用，或者我使用错误。

Answer 1

首先，您可以形成滚动的 120 秒数据块。 然后你可以申请；

使用重复的块和评估： df = df[df.duplicated(subset=['val1','val2','val3'], keep=False)]

或分组： df.groupby(['val1','val2','val3']).count()

甚至是不同的 SQL。 https://www.w3schools.com/sql/sql_distinct.asp

请发布您尝试过的内容。上述方法适用于字符串、浮点数、日期时间和整数数据类型。

Answer 2

所以我让它工作但不是滚动窗口，因为它不支持字符串类型。该功能也在 Pandas Repo 上报告和请求。

我对这个问题的解决方案：

    if len(df.index) > 0:
        res = df.loc[(df.merchant == data['transaction']['merchant']) & (df.amount == data['transaction']['amount'])]
        res['timediff'] = (data['transaction']['time'] - res['time']).dt.total_seconds().abs() <= 120
        if res.timediff.any():
            continue
    df = df.append(df1)
print(df)

样本数据：

{"transaction": {"merchant": "merchantA", "amount": 20, "time": "2019-02-13T10:00:00.000Z"}}
{"transaction": {"merchant": "merchantB", "amount": 90, "time": "2019-02-13T11:00:01.000Z"}}
{"transaction": {"merchant": "merchantC", "amount": 10, "time": "2019-02-13T11:00:10.000Z"}}
{"transaction": {"merchant": "merchantD", "amount": 10, "time": "2019-02-13T11:00:20.000Z"}}
{"transaction": {"merchant": "merchantE", "amount": 10, "time": "2019-02-13T11:01:30.000Z"}}
{"transaction": {"merchant": "merchantF", "amount": 10, "time": "2019-02-13T11:03:00.000Z"}}
{"transaction": {"merchant": "merchantE", "amount": 10, "time": "2019-02-13T11:02:00.000Z"}}
{"transaction": {"merchant": "merchantF", "amount": 10, "time": "2019-02-13T11:02:20.000Z"}}
{"transaction": {"merchant": "merchantE", "amount": 10, "time": "2019-02-13T11:02:30.000Z"}}
{"transaction": {"merchant": "merchantF", "amount": 10, "time": "2019-02-13T11:05:20.000Z"}}
{"transaction": {"merchant": "merchantE", "amount": 10, "time": "2019-02-13T11:00:30.000Z"}}

输出：

                      merchant  amount                time
2019-02-13 10:00:00  merchantA      20 2019-02-13 10:00:00
2019-02-13 11:00:01  merchantB      90 2019-02-13 11:00:01
2019-02-13 11:00:10  merchantC      10 2019-02-13 11:00:10
2019-02-13 11:00:20  merchantD      10 2019-02-13 11:00:20
2019-02-13 11:01:30  merchantE      10 2019-02-13 11:01:30
2019-02-13 11:03:00  merchantF      10 2019-02-13 11:03:00
2019-02-13 11:05:20  merchantF      10 2019-02-13 11:05:20