Pyspark string array of dynamic length in dataframe column to onehot-encoded

Question

I would like to convert a column which contains strings like:

 [\"ABC\",\"def\",\"ghi\"] 
 [\"Jkl\",\"ABC\",\"def\"]
 [\"Xyz\",\"ABC\"]

Into a

Answer 1

You can probably use CountVectorizer, Below is an example:

Update: removed the step to drop duplicates in arrays, you can set binary=True when setting up CountVectorizer:

from pyspark.ml.feature import CountVectorizer
from pyspark.sql.functions import udf, col

df = spark.createDataFrame([
        (["ABC","def","ghi"],)
      , (["Jkl","ABC","def"],)
      , (["Xyz","ABC"],)
    ], ['arr']
)

create the CountVectorizer model:

cv = CountVectorizer(inputCol='arr', outputCol='c1', binary=True)

model = cv.fit(df)

vocabulary = model.vocabulary
# [u'ABC', u'def', u'Xyz', u'ghi', u'Jkl']

Create a UDF to convert a vector into array

udf_to_array = udf(lambda v: v.toArray().tolist(), 'array')

Get the vector and check the content:

df1 = model.transform(df)

df1.withColumn('c2', udf_to_array('c1')) \
   .select('*', *[ col('c2')[i].astype('int').alias(vocabulary[i]) for i in range(len(vocabulary))]) \
   .show(3,0)
+---------------+-------------------------+-------------------------+---+---+---+---+---+
|arr            |c1                       |c2                       |ABC|def|Xyz|ghi|Jkl|
+---------------+-------------------------+-------------------------+---+---+---+---+---+
|[ABC, def, ghi]|(5,[0,1,3],[1.0,1.0,1.0])|[1.0, 1.0, 0.0, 1.0, 0.0]|1  |1  |0  |1  |0  |
|[Jkl, ABC, def]|(5,[0,1,4],[1.0,1.0,1.0])|[1.0, 1.0, 0.0, 0.0, 1.0]|1  |1  |0  |0  |1  |
|[Xyz, ABC]     |(5,[0,2],[1.0,1.0])      |[1.0, 0.0, 1.0, 0.0, 0.0]|1  |0  |1  |0  |0  |
+---------------+-------------------------+-------------------------+---+---+---+---+---+