PySpark sampleBy using multiple columns

Question

I want to carry out a stratified sampling from a data frame on PySpark. There is a sampleBy(col, fractions, seed=None) function, but it seems to only use one column as a strata. Is there any way to use multiple columns as a strata?

Answer 1

based on the answer here

after converting it to python, I think an answer might look like:

#create a dataframe to use
df = sc.parallelize([ (1,1234,282),(1,1396,179),(2,8620,178),(3,1620,191),(3,8820,828) ] ).toDF(["ID","X","Y"])

#we are going to use the first two columns as our key (strata)
#assign sampling percentages to each key # you could do something cooler here
fractions = df.rdd.map(lambda x: (x[0],x[1])).distinct().map(lambda x: (x,0.3)).collectAsMap()

#setup how we want to key the dataframe
kb = df.rdd.keyBy(lambda x: (x[0],x[1]))

#create a dataframe after sampling from our newly keyed rdd
#note, if the sample did not return any values you'll get a `ValueError: RDD is empty` error

sampleddf = kb.sampleByKey(False,fractions).map(lambda x: x[1]).toDF(df.columns)
sampleddf.show()
+---+----+---+
| ID|   X|  Y|
+---+----+---+
|  1|1234|282|
|  1|1396|179|
|  3|1620|191|
+---+----+---+
#other examples
kb.sampleByKey(False,fractions).map(lambda x: x[1]).toDF(df.columns).show()
+---+----+---+
| ID|   X|  Y|
+---+----+---+
|  2|8620|178|
+---+----+---+


kb.sampleByKey(False,fractions).map(lambda x: x[1]).toDF(df.columns).show()
+---+----+---+
| ID|   X|  Y|
+---+----+---+
|  1|1234|282|
|  1|1396|179|
+---+----+---+

Is this the kind of thing you were looking for?