Github user zapletal-martin commented on a diff in the pull request:
https://github.com/apache/spark/pull/8377#discussion_r38040278
--- Diff: docs/ml-guide.md ---
@@ -801,6 +801,173 @@ jsc.stop();
</div>
+## Example: Model Selection via Train Validation Split
+In addition to `CrossValidator` Spark also offers
+[`TrainValidationSplit`](api/scala/index.html#org.apache.spark.ml.tuning.TrainValidationSplit)
for hyper-parameter tuning.
+It randomly splits the input dataset into train and validation sets based
on ratio passed as parameter
+and use evaluation metric on the validation set to select the best model.
+The use is similar to `CrossValidator`, but simpler and less
computationally expensive.
+
+`TrainValidationSplit` takes an `Estimator`, a set of `ParamMap`s, and an
+[`Evaluator`](api/scala/index.html#org.apache.spark.ml.Evaluator).
+It begins by splitting the dataset into two parts using *trainRatio*
parameter
+which are used as separate training and test datasets. For example with
`$trainRatio=0.75$` (default),
+`TrainValidationSplit` will generate training and test dataset pair where
75% of the data is used for training and 25% for validation.
+Similarly to `CrossValidator`, `TrainValidationSplit` also iterates
through the set of `ParamMap`s.
+For each combination of parameters, it trains the given `Estimator` and
evaluates it using the given `Evaluator`.
+The `ParamMap` which produces the best evaluation metric is selected as
the best option.
+`TrainValidationSplit` finally fits the `Estimator` using the best
`ParamMap` and the entire dataset.
+
+`TrainValidationSplit` only evaluates each combination of parameters once
as opposed to k times in
+ case of `CrossValidator`. It is therefore less expensive, but will not
produce as reliable results.
+
+<div class="codetabs">
+
+<div data-lang="scala">
+{% highlight scala %}
+import org.apache.spark.ml.evaluation.RegressionEvaluator
+import org.apache.spark.ml.regression.LinearRegression
+import org.apache.spark.ml.tuning.{ParamGridBuilder, TrainValidationSplit}
+import org.apache.spark.mllib.linalg.{Vector, Vectors}
+import org.apache.spark.mllib.regression.LabeledPoint
+import org.apache.spark.sql.{Row, SQLContext}
+import org.apache.spark.{SparkConf, SparkContext}
+
+val conf = new SparkConf().setAppName("TrainValidationSplitExample")
+val sc = new SparkContext(conf)
+val sqlContext = new SQLContext(sc)
+import sqlContext.implicits._
+
+val training = sc.parallelize(Seq(
+ LabeledPoint(1.0, Vectors.dense(0.0, 1.1, 0.1)),
+ LabeledPoint(0.0, Vectors.dense(2.0, 1.0, -1.0)),
+ LabeledPoint(0.0, Vectors.dense(2.0, 1.3, 1.0)),
+ LabeledPoint(1.0, Vectors.dense(0.0, 1.2, -0.5))))
+
+val lr = new LinearRegression()
+
+// In this case the estimator is simply the linear regression.
+// A TrainValidationSplit requires an Estimator, a set of Estimator
ParamMaps, and an Evaluator.
+val trainValidationSplit = new TrainValidationSplit()
+ .setEstimator(lr)
+ .setEvaluator(new RegressionEvaluator)
+
+// We use a ParamGridBuilder to construct a grid of parameters to search
over.
+// TrainValidationSplit will try all combinations of values and determine
best model using
+// the evaluator.
+val paramGrid = new ParamGridBuilder()
+ .addGrid(lr.regParam, Array(0.1, 0.01))
--- End diff --
Simplified to 2x2x3.
I wanted the example to be different from CrossValidator example. It only
evaluates parameters of the LinearRegression rather than whole pipeline, it
uses RegressionEvaluator as opposed to BinaryClassificationEvaluator and I also
wanted to show how multiple parameter combinations can be evaluated.
---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at [email protected] or file a JIRA ticket
with INFRA.
---
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
