[SPARK-7535] [.0] [MLLIB] Audit the pipeline APIs for 1.4

Some changes to the pipeilne APIs:

1. Estimator/Transformer/ doesn’t need to extend Params since PipelineStage already does.
1. Move Evaluator to ml.evaluation.
1. Mention larger metric values are better.
1. PipelineModel doc. “compiled” -> “fitted”
1. Hide object PolynomialExpansion.
1. Hide object VectorAssembler.
1. Word2Vec.minCount (and other) -> group param
1. ParamValidators -> DeveloperApi
1. Hide MetadataUtils/SchemaUtils.

jkbradley

Author: Xiangrui Meng <meng@databricks.com>

Closes #6322 from mengxr/SPARK-7535.0 and squashes the following commits:

9e9c7da [Xiangrui Meng] move JavaEvaluator to ml.evaluation as well
e179480 [Xiangrui Meng] move Evaluation to ml.evaluation in PySpark
08ef61f [Xiangrui Meng] update pipieline APIs

(cherry picked from commit 8f11c6116b)
Signed-off-by: Xiangrui Meng <meng@databricks.com>

mllib/src/main/scala/org/apache/spark/ml/Estimator.scala

@@ -28,7 +28,7 @@ import org.apache.spark.sql.DataFrame
  * Abstract class for estimators that fit models to data.
  */
 @AlphaComponent
-abstract class Estimator[M <: Model[M]] extends PipelineStage with Params {
+abstract class Estimator[M <: Model[M]] extends PipelineStage {
 
   /**
    * Fits a single model to the input data with optional parameters.

mllib/src/main/scala/org/apache/spark/ml/Pipeline.scala

@@ -170,7 +170,7 @@ class Pipeline(override val uid: String) extends Estimator[PipelineModel] {
 
 /**
  * :: AlphaComponent ::
- * Represents a compiled pipeline.
+ * Represents a fitted pipeline.
  */
 @AlphaComponent
 class PipelineModel private[ml] (

mllib/src/main/scala/org/apache/spark/ml/Transformer.scala

@@ -32,7 +32,7 @@ import org.apache.spark.sql.types._
  * Abstract class for transformers that transform one dataset into another.
  */
 @AlphaComponent
-abstract class Transformer extends PipelineStage with Params {
+abstract class Transformer extends PipelineStage {
 
   /**
    * Transforms the dataset with optional parameters

mllib/src/main/scala/org/apache/spark/ml/evaluation/BinaryClassificationEvaluator.scala

@@ -18,7 +18,7 @@
 package org.apache.spark.ml.evaluation
 
 import org.apache.spark.annotation.AlphaComponent
-import org.apache.spark.ml.Evaluator
+import org.apache.spark.ml.evaluation.Evaluator
 import org.apache.spark.ml.param._
 import org.apache.spark.ml.param.shared._
 import org.apache.spark.ml.util.{Identifiable, SchemaUtils}

mllib/src/main/scala/org/apache/spark/ml/evaluation/Evaluator.scala

@@ -15,7 +15,7 @@
  * limitations under the License.
  */
 
-package org.apache.spark.ml
+package org.apache.spark.ml.evaluation
 
 import org.apache.spark.annotation.AlphaComponent
 import org.apache.spark.ml.param.{ParamMap, Params}
@@ -29,7 +29,7 @@ import org.apache.spark.sql.DataFrame
 abstract class Evaluator extends Params {
 
   /**
-   * Evaluates the output.
+   * Evaluates model output and returns a scalar metric (larger is better).
    *
    * @param dataset a dataset that contains labels/observations and predictions.
    * @param paramMap parameter map that specifies the input columns and output metrics

mllib/src/main/scala/org/apache/spark/ml/feature/PolynomialExpansion.scala

@@ -75,7 +75,7 @@ class PolynomialExpansion(override val uid: String)
  * To handle sparsity, if c is zero, we can skip all monomials that contain it. We remember the
  * current index and increment it properly for sparse input.
  */
-object PolynomialExpansion {
+private[feature] object PolynomialExpansion {
 
   private def choose(n: Int, k: Int): Int = {
     Range(n, n - k, -1).product / Range(k, 1, -1).product

mllib/src/main/scala/org/apache/spark/ml/feature/VectorAssembler.scala

@@ -78,8 +78,7 @@ class VectorAssembler(override val uid: String)
   }
 }
 
-@AlphaComponent
-object VectorAssembler {
+private object VectorAssembler {
 
   private[feature] def assemble(vv: Any*): Vector = {
     val indices = ArrayBuilder.make[Int]

mllib/src/main/scala/org/apache/spark/ml/feature/Word2Vec.scala

@@ -37,6 +37,7 @@ private[feature] trait Word2VecBase extends Params
 
   /**
    * The dimension of the code that you want to transform from words.
+   * @group param
    */
   final val vectorSize = new IntParam(
     this, "vectorSize", "the dimension of codes after transforming from words")
@@ -47,6 +48,7 @@ private[feature] trait Word2VecBase extends Params
 
   /**
    * Number of partitions for sentences of words.
+   * @group param
    */
   final val numPartitions = new IntParam(
     this, "numPartitions", "number of partitions for sentences of words")
@@ -58,6 +60,7 @@ private[feature] trait Word2VecBase extends Params
   /**
    * The minimum number of times a token must appear to be included in the word2vec model's
    * vocabulary.
+   * @group param
    */
   final val minCount = new IntParam(this, "minCount", "the minimum number of times a token must " +
     "appear to be included in the word2vec model's vocabulary")

mllib/src/main/scala/org/apache/spark/ml/param/params.scala

@@ -24,7 +24,7 @@ import scala.annotation.varargs
 import scala.collection.mutable
 import scala.collection.JavaConverters._
 
-import org.apache.spark.annotation.AlphaComponent
+import org.apache.spark.annotation.{DeveloperApi, AlphaComponent}
 import org.apache.spark.ml.util.Identifiable
 
 /**
@@ -92,9 +92,11 @@ class Param[T](val parent: String, val name: String, val doc: String, val isVali
 }
 
 /**
+ * :: DeveloperApi ::
  * Factory methods for common validation functions for [[Param.isValid]].
  * The numerical methods only support Int, Long, Float, and Double.
  */
+@DeveloperApi
 object ParamValidators {
 
   /** (private[param]) Default validation always return true */
@@ -529,11 +531,13 @@ trait Params extends Identifiable with Serializable {
 }
 
 /**
+ * :: DeveloperApi ::
  * Java-friendly wrapper for [[Params]].
  * Java developers who need to extend [[Params]] should use this class instead.
  * If you need to extend a abstract class which already extends [[Params]], then that abstract
  * class should be Java-friendly as well.
  */
+@DeveloperApi
 abstract class JavaParams extends Params
 
 /**

mllib/src/main/scala/org/apache/spark/ml/tuning/CrossValidator.scala

@@ -22,6 +22,7 @@ import com.github.fommil.netlib.F2jBLAS
 import org.apache.spark.Logging
 import org.apache.spark.annotation.AlphaComponent
 import org.apache.spark.ml._
+import org.apache.spark.ml.evaluation.Evaluator
 import org.apache.spark.ml.param._
 import org.apache.spark.ml.util.Identifiable
 import org.apache.spark.mllib.util.MLUtils

mllib/src/main/scala/org/apache/spark/ml/util/MetadataUtils.scala

@@ -19,18 +19,14 @@ package org.apache.spark.ml.util
 
 import scala.collection.immutable.HashMap
 
-import org.apache.spark.annotation.Experimental
 import org.apache.spark.ml.attribute._
 import org.apache.spark.sql.types.StructField
 
 
 /**
- * :: Experimental ::
- *
  * Helper utilities for tree-based algorithms
  */
-@Experimental
-object MetadataUtils {
+private[spark] object MetadataUtils {
 
   /**
    * Examine a schema to identify the number of classes in a label column.

mllib/src/main/scala/org/apache/spark/ml/util/SchemaUtils.scala

@@ -17,15 +17,13 @@
 
 package org.apache.spark.ml.util
 
-import org.apache.spark.annotation.DeveloperApi
 import org.apache.spark.sql.types.{DataType, StructField, StructType}
 
+
 /**
- * :: DeveloperApi ::
  * Utils for handling schemas.
  */
-@DeveloperApi
-object SchemaUtils {
+private[spark] object SchemaUtils {
 
   // TODO: Move the utility methods to SQL.
 

python/pyspark/ml/__init__.py

@@ -15,6 +15,6 @@
 # limitations under the License.
 #
 
-from pyspark.ml.pipeline import Transformer, Estimator, Model, Pipeline, PipelineModel, Evaluator
+from pyspark.ml.pipeline import Transformer, Estimator, Model, Pipeline, PipelineModel
 
-__all__ = ["Transformer", "Estimator", "Model", "Pipeline", "PipelineModel", "Evaluator"]
+__all__ = ["Transformer", "Estimator", "Model", "Pipeline", "PipelineModel"]

python/pyspark/ml/evaluation.py

@@ -15,13 +15,72 @@
 # limitations under the License.
 #
 
-from pyspark.ml.wrapper import JavaEvaluator
+from abc import abstractmethod, ABCMeta
+
+from pyspark.ml.wrapper import JavaWrapper
 from pyspark.ml.param import Param, Params
 from pyspark.ml.param.shared import HasLabelCol, HasRawPredictionCol
 from pyspark.ml.util import keyword_only
 from pyspark.mllib.common import inherit_doc
 
-__all__ = ['BinaryClassificationEvaluator']
+__all__ = ['Evaluator', 'BinaryClassificationEvaluator']
+
+
+@inherit_doc
+class Evaluator(Params):
+    """
+    Base class for evaluators that compute metrics from predictions.
+    """
+
+    __metaclass__ = ABCMeta
+
+    @abstractmethod
+    def _evaluate(self, dataset):
+        """
+        Evaluates the output.
+
+        :param dataset: a dataset that contains labels/observations and
+               predictions
+        :return: metric
+        """
+        raise NotImplementedError()
+
+    def evaluate(self, dataset, params={}):
+        """
+        Evaluates the output with optional parameters.
+
+        :param dataset: a dataset that contains labels/observations and
+                        predictions
+        :param params: an optional param map that overrides embedded
+                       params
+        :return: metric
+        """
+        if isinstance(params, dict):
+            if params:
+                return self.copy(params)._evaluate(dataset)
+            else:
+                return self._evaluate(dataset)
+        else:
+            raise ValueError("Params must be a param map but got %s." % type(params))
+
+
+@inherit_doc
+class JavaEvaluator(Evaluator, JavaWrapper):
+    """
+    Base class for :py:class:`Evaluator`s that wrap Java/Scala
+    implementations.
+    """
+
+    __metaclass__ = ABCMeta
+
+    def _evaluate(self, dataset):
+        """
+        Evaluates the output.
+        :param dataset: a dataset that contains labels/observations and predictions.
+        :return: evaluation metric
+        """
+        self._transfer_params_to_java()
+        return self._java_obj.evaluate(dataset._jdf)
 
 
 @inherit_doc

python/pyspark/ml/pipeline.py

@@ -219,40 +219,3 @@ class PipelineModel(Model):
     def copy(self, extra={}):
         stages = [stage.copy(extra) for stage in self.stages]
         return PipelineModel(stages)
-
-
-class Evaluator(Params):
-    """
-    Base class for evaluators that compute metrics from predictions.
-    """
-
-    __metaclass__ = ABCMeta
-
-    @abstractmethod
-    def _evaluate(self, dataset):
-        """
-        Evaluates the output.
-
-        :param dataset: a dataset that contains labels/observations and
-               predictions
-        :return: metric
-        """
-        raise NotImplementedError()
-
-    def evaluate(self, dataset, params={}):
-        """
-        Evaluates the output with optional parameters.
-
-        :param dataset: a dataset that contains labels/observations and
-                        predictions
-        :param params: an optional param map that overrides embedded
-                       params
-        :return: metric
-        """
-        if isinstance(params, dict):
-            if params:
-                return self.copy(params)._evaluate(dataset)
-            else:
-                return self._evaluate(dataset)
-        else:
-            raise ValueError("Params must be a param map but got %s." % type(params))

python/pyspark/ml/wrapper.py

@@ -20,7 +20,7 @@ from abc import ABCMeta
 from pyspark import SparkContext
 from pyspark.sql import DataFrame
 from pyspark.ml.param import Params
-from pyspark.ml.pipeline import Estimator, Transformer, Evaluator, Model
+from pyspark.ml.pipeline import Estimator, Transformer, Model
 from pyspark.mllib.common import inherit_doc, _java2py, _py2java
 
 
@@ -185,22 +185,3 @@ class JavaModel(Model, JavaTransformer):
         sc = SparkContext._active_spark_context
         java_args = [_py2java(sc, arg) for arg in args]
         return _java2py(sc, m(*java_args))
-
-
-@inherit_doc
-class JavaEvaluator(Evaluator, JavaWrapper):
-    """
-    Base class for :py:class:`Evaluator`s that wrap Java/Scala
-    implementations.
-    """
-
-    __metaclass__ = ABCMeta
-
-    def _evaluate(self, dataset):
-        """
-        Evaluates the output.
-        :param dataset: a dataset that contains labels/observations and predictions.
-        :return: evaluation metric
-        """
-        self._transfer_params_to_java()
-        return self._java_obj.evaluate(dataset._jdf)