How to merge multiple feature vectors in DataFrame

问题:

Using Spark ML transformers I arrived at a DataFrame where each row looks like this:

Row(object_id, text_features_vector, color_features, type_features)

where text_features is a sparse vector of term weights, color_features is a small 20-element (one-hot-encoder) dense vector of colors, and type_features is also a one-hot-encoder dense vector of types.

What would a good approach be (using Spark's facilities) to merge these features in one single, large array, so that I measure things like the cosine distance between any two objects?

回答1:

You can use VectorAssembler:

import org.apache.spark.ml.feature.VectorAssembler
import org.apache.spark.sql.DataFrame

val df: DataFrame = ???

val assembler = new VectorAssembler()
  .setInputCols(Array("text_features", "color_features", "type_features"))
  .setOutputCol("features")

val transformed = assembler.transform(df)

For PySpark example see: Encode and assemble multiple features in PySpark