CODEWITHSUNDEEP
pythonhadoopapache-sparkhivepyspark
user3292373
user3292373

Reputation: 533

Pyspark iterate over dataframe by group on lookup of previous row

Please help me in this I am new to spark. Below is mydataframe

type col1 col2 col3
1    0    41   0
1    27   0    0
1    1    0    0 
1    183  0    2
2    null 0    0
2    null 10   0
3    0    126  0
3    2    0    1
3    4    0    0
3    5    0    0

Below should be my output

type col1 col2 col3 result
1    0    41   0    0
1    27   0    0    14
1    1    0    0    13
1    183  0    2    -168
2    null 0    0
2    null 10   0
3    0    126  0    0
3    2    0    1    125
3    4    0    0    121
3    5    0    0    116

The challenge is this has to be done for every group of type column the formula is like prev(col2)-col1+col3

I tried to use window and lag function on col2 to populate result column but it did not work.

Below was my code

part = Window().partitionBy().orderBy('type')
DF = DF.withColumn('result',lag("col2").over(w)-DF.col1+DF.col3)

Now I am struggling to try with map function please help

Upvotes: 1

Views: 3742

Answers (1)

Ramesh Maharjan
Ramesh Maharjan

Reputation: 41957

The logic is a bit tricky and complex.

You can do the following in pyspark

pyspark

from pyspark.sql import functions as F
from pyspark.sql import Window
import sys
windowSpec = Window.partitionBy("type").orderBy("type")
df = df.withColumn('result', F.lag(df.col2, 1).over(windowSpec) - df.col1 + df.col3)
df = df.withColumn('result', F.when(df.result.isNull(), F.lit(0)).otherwise(df.result))
df = df.withColumn('result', F.sum(df.result).over(windowSpec.rowsBetween(-sys.maxsize, -1)) + df.result)
df = df.withColumn('result', F.when(df.result.isNull(), F.lit(0)).otherwise(df.result))

scala

import org.apache.spark.sql.expressions._
import org.apache.spark.sql.functions._
val windowSpec = Window.partitionBy("type").orderBy("type")
df.withColumn("result", lag("col2", 1).over(windowSpec) - $"col1"+$"col3")
  .withColumn("result", when($"result".isNull, lit(0)).otherwise($"result"))
  .withColumn("result", sum("result").over(windowSpec.rowsBetween(Long.MinValue, -1)) +$"result")
  .withColumn("result", when($"result".isNull, lit(0)).otherwise($"result"))

You should have the following result.

+----+----+----+----+------+
|type|col1|col2|col3|result|
+----+----+----+----+------+
|1   |0   |41  |0   |0.0   |
|1   |27  |0   |0   |14.0  |
|1   |1   |0   |0   |13.0  |
|1   |183 |0   |2   |-168.0|
|3   |0   |126 |0   |0.0   |
|3   |2   |0   |1   |125.0 |
|3   |4   |0   |0   |121.0 |
|3   |5   |0   |0   |116.0 |
|2   |null|0   |0   |0.0   |
|2   |null|10  |0   |0.0   |
+----+----+----+----+------+

Edited

the first withColumn applies the formula prev(col2) - col1 + col3. The second withColumn changes null to 0 for result column. The third withColumn is for cumulative sum i.e. adding all the values until the current row of result column. so the three withColumn is equivalent to prev(col2) + prev(results) 1 col1 + col3. The last withColumn is changing the null values to 0 in result column.

Upvotes: 3

Related Questions