Spark - StringIndexer Vs OneHotEncoderEstimator

Question

I am learning Spark and I have below code in one of the tutorial. I understand the dataframe is one hot encoded in below code but what I don't understand is why StringIndexer is used? Is StringIndexer should be used in conjunction with OneHotEncoderEstimator? val si = new StringIndexer() .setHandleInvalid("keep") .setInputCol(ProcuctTypeCol) .setOutputCol(ProcuctTypeSIOutCol)

val ohe = new OneHotEncoderEstimator()
      .setHandleInvalid("keep")
      .setInputCols(Array(si.getOutputCol))
      .setOutputCols(Array(ProductTypeOHEOutCol))

val pipeline = new Pipeline()
  .setStages(Array(si, ohe))

Thanks

Answer 1

SI transform the string value to integer and OHE make integer in ohe encoding, if your column is in int like 1,2,3 you can apply OHE directly. But if your label in string like A,B,C you have to use SI first then chain OHE