Package org.apache.spark.ml.feature

Feature transformers The `ml.feature` package provides common feature transformers that help convert raw data or features into more suitable forms for model fitting.

See: Description

Package org.apache.spark.ml.feature Description

Feature transformers The `ml.feature` package provides common feature transformers that help convert raw data or features into more suitable forms for model fitting. Most feature transformers are implemented as Transformers, which transforms one DataFrame into another, e.g., HashingTF. Some feature transformers are implemented as Estimator}s, because the transformation requires some aggregated information of the dataset, e.g., document frequencies in IDF. For those feature transformers, calling Estimator.fit(org.apache.spark.sql.DataFrame, org.apache.spark.ml.param.ParamPair<?>, org.apache.spark.ml.param.ParamPair<?>...) is required to obtain the model first, e.g., IDFModel, in order to apply transformation. The transformation is usually done by appending new columns to the input DataFrame, so all input columns are carried over. We try to make each transformer minimal, so it becomes flexible to assemble feature transformation pipelines. Pipeline can be used to chain feature transformers, and VectorAssembler can be used to combine multiple feature transformations, for example:
 
   import java.util.Arrays;

   import org.apache.spark.api.java.JavaRDD;
   import static org.apache.spark.sql.types.DataTypes.*;
   import org.apache.spark.sql.types.StructType;
   import org.apache.spark.sql.DataFrame;
   import org.apache.spark.sql.RowFactory;
   import org.apache.spark.sql.Row;

   import org.apache.spark.ml.feature.*;
   import org.apache.spark.ml.Pipeline;
   import org.apache.spark.ml.PipelineStage;
   import org.apache.spark.ml.PipelineModel;

  // a DataFrame with three columns: id (integer), text (string), and rating (double).
  StructType schema = createStructType(
    Arrays.asList(
      createStructField("id", IntegerType, false),
      createStructField("text", StringType, false),
      createStructField("rating", DoubleType, false)));
  JavaRDD rowRDD = jsc.parallelize(
    Arrays.asList(
      RowFactory.create(0, "Hi I heard about Spark", 3.0),
      RowFactory.create(1, "I wish Java could use case classes", 4.0),
      RowFactory.create(2, "Logistic regression models are neat", 4.0)));
  DataFrame df = jsql.createDataFrame(rowRDD, schema);
  // define feature transformers
  RegexTokenizer tok = new RegexTokenizer()
    .setInputCol("text")
    .setOutputCol("words");
  StopWordsRemover sw = new StopWordsRemover()
    .setInputCol("words")
    .setOutputCol("filtered_words");
  HashingTF tf = new HashingTF()
    .setInputCol("filtered_words")
    .setOutputCol("tf")
    .setNumFeatures(10000);
  IDF idf = new IDF()
    .setInputCol("tf")
    .setOutputCol("tf_idf");
  VectorAssembler assembler = new VectorAssembler()
    .setInputCols(new String[] {"tf_idf", "rating"})
    .setOutputCol("features");

  // assemble and fit the feature transformation pipeline
  Pipeline pipeline = new Pipeline()
    .setStages(new PipelineStage[] {tok, sw, tf, idf, assembler});
  PipelineModel model = pipeline.fit(df);

  // save transformed features with raw data
  model.transform(df)
    .select("id", "text", "rating", "features")
    .write().format("parquet").save("/output/path");
 
 
Some feature transformers implemented in MLlib are inspired by those implemented in scikit-learn. The major difference is that most scikit-learn feature transformers operate eagerly on the entire input dataset, while MLlib's feature transformers operate lazily on individual columns, which is more efficient and flexible to handle large and complex datasets.
See Also:
scikit-learn.preprocessing