一个 dataframe 列与另一个 dataframe 列的倍数基于条件

Question

There are two dataframe, one is info table, and another one is reference table. I need to multiple two columns based on the conditions, here is the details:

Dataframe (Info)

+-----+-----+
|  key|value|
+-----+-----+
|    a|   10|
|    b|   20|
|    c|   50|
|    d|   40|
+-----+-----+

Dataframe (Reference)

+-----+----------+
|  key|percentage|
+-----+----------+
|    a|       0.1|
|    b|       0.5|
+-----+----------+

Dataframe (this is the output I want)

+-----+------+
|  key|result|
+-----+------+
|    a|     1|   (10 * 0.1 = 1)
|    b|    10|   (20 * 0.5 = 10)
|    c|    50|   (because there are no key matching in reference table, then remain the same)
|    d|    40|   (because there are no key matching in reference table, then remain the same)
+-----+------+

I have try the below code but failed.

df_cal = (
    info
    .withColumn('result', f.when(f.col('key')==reference.withColumn(f.col('key')), \
                          f.col('value)*reference.withColumn(f.col('percentage')) ))
    .select('key', 'result')
)

df_cal.show()

Answer 1

Join and multiply. Code and logic below

new_info = (info.join(broadcast(Reference), on='key', how='left')#Join the two dataframes
 .na.fill(1.0)#Fill null with 1
 .withColumn('result', col('value')*col('percentage'))#multiply the columns and store in results
 .drop('value','percentage')#drop unwanted columns
)

new_info.show()

Answer 2

a slight difference from wwnde's solution, with the overall logic remaining same, would be to use coalesce instead of the fillna . fillna , if used without subset, can fill unwanted columns as well - and in any case, it generates a new projection in the spark plan.

example using coalesce

data1_sdf. \
    join(data2_sdf, ['key'], 'left'). \
    withColumn('result', 
               func.coalesce(func.col('value') * func.col('percentage'), func.col('value'))
               ). \
    show()

# +---+-----+----------+------+
# |key|value|percentage|result|
# +---+-----+----------+------+
# |  d|   40|      null|  40.0|
# |  c|   50|      null|  50.0|
# |  b|   20|       0.5|  10.0|
# |  a|   10|       0.1|   1.0|
# +---+-----+----------+------+

Answer 3

If you are willing to use Spark SQL instead of the DataFrame API, you can do this approach:

Create dataframes. (Optional since you already have the data)

from pyspark.sql.types import StructType,StructField, IntegerType, FloatType, StringType

# create info dataframe
info_data = [
  ("a",10),
  ("b",20),
  ("c",50),
  ("d",40),
]
info_schema = StructType([
  StructField("key",StringType()),
  StructField("value",IntegerType()),
])
info_df = spark.createDataFrame(data=info_data,schema=info_schema)

# create reference dataframe
reference_data = [
  ("a",.1),
  ("b",.5)
]
reference_schema = StructType([
  StructField("key",StringType()),
  StructField("percentage",FloatType()),
])
reference_df = spark.createDataFrame(data=reference_data,schema=reference_schema)
reference_df.show()

Next we need to create views of the 2 dataframes to run sql queries. Below we create a view called info from info_df and a view called reference from reference_df

# create views: info and reference
info_df.createOrReplaceTempView("info")
reference_df.createOrReplaceTempView("reference")

Finally we write a query to perform the multiplication. The query performs a left join between info and reference and then multiplies value by percentage . The key part is that we coalesce percentage with 1. Thus if percentage is null, then value is multiplied by 1.

from pyspark.sql.functions import coalesce

my_query = """
select
  i.key,
  -- coalese the percentage with 1. If percentage is null then it gets replaced by 1
  i.value * coalesce(r.percentage,1) as result
from info i
left join reference r
  on i.key = r.key
"""

final_df = spark.sql(my_query)
final_df.show()

Output:

+---+------+
|key|result|
+---+------+
|  a|   1.0|
|  b|  10.0|
|  c|  50.0|
|  d|  40.0|
+---+------+