Pyspark - how to do case insensitive dataframe joins?

Question

Is there any nice looking code to perform case insensitive join in Pyspark? Something like:

df3 = df1.join(df2, 
               ["col1", "col2", "col3"],
               "left_outer",
               "case-insensitive")

Or what is your working solutions to this?

Answer 1

It's not exactly elegant, but you could create new lower-case versions of those columns purely for joining.

import pyspark.sql.functions as F
df1_l = df1 \
    .withColumn("col1_l", F.lower(df1.col1)) \
    .withColumn("col2_l", F.lower(df1.col2)) \
    .withColumn("col3_l"), F.lower(df1.col3)

df2_l = df2 \
    .withColumn("col1_l", F.lower(df2.col1)) \
    .withColumn("col2_l", F.lower(df2.col2)) \
    .withColumn("col3_l"), F.lower(df2.col3)

df3 = df1_l.join(df2_l, 
           ["col1_l", "col2_l", "col3_l"],
           "left_outer")

And you could also try doing this same transformation in a join predicate, e.g.:

df3 = df1.join(df2, 
           (F.lower(df1.col1) == F.lower(df2.col1))
            & (F.lower(df1.col2) == F.lower(df2.col2))
            & (F.lower(df1.col3) == F.lower(df2.col3))
           "left_outer")