I would like to convert a column which contains strings like:
["ABC","def","ghi"]
["Jkl","ABC","def"]
["Xyz","ABC"]
Into a encoded column like this:
[1,1,1,0,0]
[1,1,0,1,0]
[0,1,0,0,1]
Is there a class for that in pyspark.ml.feature?
Edit: In the encoded column the first entry always corresponds to the value "ABC" etc. 1 means "ABC" is present while 0 means it is not present in the corresponding row.
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<double>')
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 |
+---------------+-------------------------+-------------------------+---+---+---+---+---+
You will have to expand the list in a single column to multiple n columns (where n is the number of items in the given list). Then you can use the OneHotEncoderEstimator class to convert it into One hot encoded features.
Please follow the example in the documentation:
from pyspark.ml.feature import OneHotEncoderEstimator
df = spark.createDataFrame([
(0.0, 1.0),
(1.0, 0.0),
(2.0, 1.0),
(0.0, 2.0),
(0.0, 1.0),
(2.0, 0.0)
], ["categoryIndex1", "categoryIndex2"])
encoder = OneHotEncoderEstimator(inputCols=["categoryIndex1", "categoryIndex2"],
outputCols=["categoryVec1", "categoryVec2"])
model = encoder.fit(df)
encoded = model.transform(df)
encoded.show()
OneHotEncoder class has been deprecated since v2.3 because it is a stateless transformer, it is not usable on new data where the number of categories may differ from the training data.
This will help you to split the list: How to split a list to multiple columns in Pyspark?
