PySpark：如何按固定日期范围分组，另一列使用窗口函数计算值列的总和？

Question

I have a Spark DataFrame consisting of three columns: Date , Item and Value of types Date , String and Double respectively. I would like to group by date range (where every range's duration is 7 days starting from the first date in the dataframe and up) and Item, and calculate Value's sums for each such group defined by the date range (week number actually) and Item.

I suspect PySpark's Window functions should be used at some point here for date ranges but can't figure out how to implement them in this case.

Answer 1

Let's first define approach for this -

(a) Add week_start_date column for row (each date)

(b) Use week_start_date column in group by (along with 'item') and calculate sum of "value"

Generate some test data

from pyspark.sql.types import *

schema = StructType([StructField('date', StringType(),True),
                     StructField('item', StringType(),True),
                     StructField('value', DoubleType(),True)
    ]
    )

data = [('2019-01-01','I1',1.1),
        ('2019-01-02','I1',1.1),
        ('2019-01-10','I1',1.1),
        ('2019-01-10','I2',1.1),
        ('2019-01-11','I2',1.1),
        ('2019-01-11','I3',1.1)]

df = spark.createDataFrame(data, schema)

Python function to generate week_start_date

from datetime import datetime, timedelta

def week_start_date(day):
    dt = datetime.strptime(day, '%Y-%m-%d')
    start = dt - timedelta(days=dt.weekday())
    end = start + timedelta(days=6)
    return start.strftime('%Y-%m-%d')

spark.udf.register('week_start_date',week_start_date)

Use function to generate week_start_date and then group by on week_start_date and item

 df.selectExpr("week_start_date(date) as start_date","date","item as item","value as value" ).\
        groupBy("start_date","item").\
        agg(sum('value').alias('value_sum')).\
        orderBy("start_date").\
        show()