PySpark 在数据框中按组删除前导零值

PySpark drop leading zero values by group in dataframe

我有这个数据框 -

data = [(0,1,5,5,0,4),
        (1,1,5,6,0,7),
        (2,1,5,7,1,1), 
        (3,1,4,8,1,8), 
        (4,1,5,9,1,1), 
        (5,1,5,10,1,0),
        (6,2,3,4,0,2),
        (7,2,3,5,0,6),
        (8,2,3,6,3,8),
        (9,2,3,7,0,2),
        (10,2,3,8,0,6),
        (11,2,3,9,6,1)
      ]
data_cols = ["id","item","store","week","sales","inventory"]
data_df = spark.createDataFrame(data=data, schema = data_)
display(deptDF)

我想要的是对项目、商店和周进行分组,然后删除每组销售额中前导 0 的所有行,就像这样

data_new = [(2,1,5,7,1,1), 
        (3,1,4,8,1,8), 
        (4,1,5,9,1,1), 
        (5,1,5,10,1,0),
        (8,2,3,6,3,8),
        (9,2,3,7,0,2),
        (10,2,3,8,0,6),
        (11,2,3,9,6,1)
      ]
dep_cols = ["id","item","store","week","sales","inventory"]
data_df_new = spark.createDataFrame(data=data_new, schema = dep_cols)
display(data_df_new)

我需要在 PySpark 中执行此操作,而且我是新手。请帮忙!

使用窗口函数,按递增求和或collect_list排序。

  1. 筛选总和大于 0 的地方

2 个过滤器列表,这里有任何大于 0 的值。我更喜欢求和,因为它更快。

w=Window.partitionBy('item','store').orderBy(F.asc('week')).rowsBetween(Window.unboundedPreceding, Window.currentRow)

df.withColumn("sums", F.sum('Sales').over(w)).filter(col('sums')>0).drop('sums').show()

+---+----+-----+----+-----+---+
| id|item|store|week|sales|inv|
+---+----+-----+----+-----+---+
|  2|   1|    5|   7|    1|  1|
|  3|   1|    5|   8|    1|  8|
|  4|   1|    5|   9|    1|  1|
|  5|   1|    5|  10|    1|  0|
|  8|   2|    3|   6|    3|  8|
|  9|   2|    3|   7|    0|  2|
| 10|   2|    3|   8|    0|  6|
| 11|   2|    3|   9|    6|  1|
+---+----+-----+----+-----+---+